Models

hbayesdm.models.bandit2arm_delta(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

2-Armed Bandit Task - Rescorla-Wagner (Delta) Model

Hierarchical Bayesian Modeling of the 2-Armed Bandit Task [Erev2010], [Hertwig2004] using Rescorla-Wagner (Delta) Model with the following parameters: “A” (learning rate), “tau” (inverse temperature).

[Erev2010]Erev, I., Ert, E., Roth, A. E., Haruvy, E., Herzog, S. M., Hau, R., et al. (2010). A choice prediction competition: Choices from experience and from description. Journal of Behavioral Decision Making, 23(1), 15-47. http://doi.org/10.1002/bdm.683
[Hertwig2004]Hertwig, R., Barron, G., Weber, E. U., & Erev, I. (2004). Decisions From Experience and the Effect of Rare Events in Risky Choice. Psychological Science, 15(8), 534-539. http://doi.org/10.1111/j.0956-7976.2004.00715.x

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the 2-Armed Bandit Task, there should be 3 columns of data with the labels “subjID”, “choice”, “outcome”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer value representing the option chosen on the given trial: 1 or 2.
  • “outcome”: Integer value representing the outcome of the given trial (where reward == 1, and loss == -1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “outcome”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘bandit2arm_delta’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- bandit2arm_delta(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.bandit4arm2_kalman_filter(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

4-Armed Bandit Task (modified) - Kalman Filter

Hierarchical Bayesian Modeling of the 4-Armed Bandit Task (modified) using Kalman Filter [Daw2006] with the following parameters: “lambda” (decay factor), “theta” (decay center), “beta” (inverse softmax temperature), “mu0” (anticipated initial mean of all 4 options), “sigma0” (anticipated initial sd (uncertainty factor) of all 4 options), “sigmaD” (sd of diffusion noise).

[Daw2006]Daw, N. D., O’Doherty, J. P., Dayan, P., Seymour, B., & Dolan, R. J. (2006). Cortical substrates for exploratory decisions in humans. Nature, 441(7095), 876-879.

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the 4-Armed Bandit Task (modified), there should be 3 columns of data with the labels “subjID”, “choice”, “outcome”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer value representing the option chosen on the given trial: 1, 2, 3, or 4.
  • “outcome”: Integer value representing the outcome of the given trial (where reward == 1, and loss == -1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “outcome”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘bandit4arm2_kalman_filter’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- bandit4arm2_kalman_filter(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.bandit4arm_2par_lapse(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

4-Armed Bandit Task - 3 Parameter Model, without C (choice perseveration), R (reward sensitivity), and P (punishment sensitivity). But with xi (noise)

Hierarchical Bayesian Modeling of the 4-Armed Bandit Task using 3 Parameter Model, without C (choice perseveration), R (reward sensitivity), and P (punishment sensitivity). But with xi (noise) [Aylward2018] with the following parameters: “Arew” (reward learning rate), “Apun” (punishment learning rate), “xi” (noise).

[Aylward2018]Aylward, Valton, Ahn, Bond, Dayan, Roiser, & Robinson (2018) Altered decision-making under uncertainty in unmedicated mood and anxiety disorders. PsyArxiv. 10.31234/osf.io/k5b8m

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the 4-Armed Bandit Task, there should be 4 columns of data with the labels “subjID”, “choice”, “gain”, “loss”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer value representing the option chosen on the given trial: 1, 2, 3, or 4.
  • “gain”: Floating point value representing the amount of currency won on the given trial (e.g. 50, 100).
  • “loss”: Floating point value representing the amount of currency lost on the given trial (e.g. 0, -50).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “gain”, “loss”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘bandit4arm_2par_lapse’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- bandit4arm_2par_lapse(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.bandit4arm_4par(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

4-Armed Bandit Task - 4 Parameter Model, without C (choice perseveration)

Hierarchical Bayesian Modeling of the 4-Armed Bandit Task using 4 Parameter Model, without C (choice perseveration) [Seymour2012] with the following parameters: “Arew” (reward learning rate), “Apun” (punishment learning rate), “R” (reward sensitivity), “P” (punishment sensitivity).

[Seymour2012]Seymour, Daw, Roiser, Dayan, & Dolan (2012). Serotonin Selectively Modulates Reward Value in Human Decision-Making. J Neuro, 32(17), 5833-5842.

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the 4-Armed Bandit Task, there should be 4 columns of data with the labels “subjID”, “choice”, “gain”, “loss”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer value representing the option chosen on the given trial: 1, 2, 3, or 4.
  • “gain”: Floating point value representing the amount of currency won on the given trial (e.g. 50, 100).
  • “loss”: Floating point value representing the amount of currency lost on the given trial (e.g. 0, -50).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “gain”, “loss”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘bandit4arm_4par’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- bandit4arm_4par(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.bandit4arm_lapse(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

4-Armed Bandit Task - 5 Parameter Model, without C (choice perseveration) but with xi (noise)

Hierarchical Bayesian Modeling of the 4-Armed Bandit Task using 5 Parameter Model, without C (choice perseveration) but with xi (noise) [Seymour2012] with the following parameters: “Arew” (reward learning rate), “Apun” (punishment learning rate), “R” (reward sensitivity), “P” (punishment sensitivity), “xi” (noise).

[Seymour2012]Seymour, Daw, Roiser, Dayan, & Dolan (2012). Serotonin Selectively Modulates Reward Value in Human Decision-Making. J Neuro, 32(17), 5833-5842.

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the 4-Armed Bandit Task, there should be 4 columns of data with the labels “subjID”, “choice”, “gain”, “loss”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer value representing the option chosen on the given trial: 1, 2, 3, or 4.
  • “gain”: Floating point value representing the amount of currency won on the given trial (e.g. 50, 100).
  • “loss”: Floating point value representing the amount of currency lost on the given trial (e.g. 0, -50).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “gain”, “loss”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘bandit4arm_lapse’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- bandit4arm_lapse(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.bandit4arm_lapse_decay(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

4-Armed Bandit Task - 5 Parameter Model, without C (choice perseveration) but with xi (noise). Added decay rate (Niv et al., 2015, J. Neuro).

Hierarchical Bayesian Modeling of the 4-Armed Bandit Task using 5 Parameter Model, without C (choice perseveration) but with xi (noise). Added decay rate (Niv et al., 2015, J. Neuro). [Aylward2018] with the following parameters: “Arew” (reward learning rate), “Apun” (punishment learning rate), “R” (reward sensitivity), “P” (punishment sensitivity), “xi” (noise), “d” (decay rate).

[Aylward2018]Aylward, Valton, Ahn, Bond, Dayan, Roiser, & Robinson (2018) Altered decision-making under uncertainty in unmedicated mood and anxiety disorders. PsyArxiv. 10.31234/osf.io/k5b8m

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the 4-Armed Bandit Task, there should be 4 columns of data with the labels “subjID”, “choice”, “gain”, “loss”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer value representing the option chosen on the given trial: 1, 2, 3, or 4.
  • “gain”: Floating point value representing the amount of currency won on the given trial (e.g. 50, 100).
  • “loss”: Floating point value representing the amount of currency lost on the given trial (e.g. 0, -50).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “gain”, “loss”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘bandit4arm_lapse_decay’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- bandit4arm_lapse_decay(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.bandit4arm_singleA_lapse(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

4-Armed Bandit Task - 4 Parameter Model, without C (choice perseveration) but with xi (noise). Single learning rate both for R and P.

Hierarchical Bayesian Modeling of the 4-Armed Bandit Task using 4 Parameter Model, without C (choice perseveration) but with xi (noise). Single learning rate both for R and P. [Aylward2018] with the following parameters: “A” (learning rate), “R” (reward sensitivity), “P” (punishment sensitivity), “xi” (noise).

[Aylward2018]Aylward, Valton, Ahn, Bond, Dayan, Roiser, & Robinson (2018) Altered decision-making under uncertainty in unmedicated mood and anxiety disorders. PsyArxiv. 10.31234/osf.io/k5b8m

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the 4-Armed Bandit Task, there should be 4 columns of data with the labels “subjID”, “choice”, “gain”, “loss”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer value representing the option chosen on the given trial: 1, 2, 3, or 4.
  • “gain”: Floating point value representing the amount of currency won on the given trial (e.g. 50, 100).
  • “loss”: Floating point value representing the amount of currency lost on the given trial (e.g. 0, -50).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “gain”, “loss”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘bandit4arm_singleA_lapse’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- bandit4arm_singleA_lapse(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.bart_par4(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Balloon Analogue Risk Task - Re-parameterized version of BART model with 4 parameters

Hierarchical Bayesian Modeling of the Balloon Analogue Risk Task using Re-parameterized version of BART model with 4 parameters [van_Ravenzwaaij2011] with the following parameters: “phi” (prior belief of balloon not bursting), “eta” (updating rate), “gam” (risk-taking parameter), “tau” (inverse temperature).

[van_Ravenzwaaij2011]van Ravenzwaaij, D., Dutilh, G., & Wagenmakers, E. J. (2011). Cognitive model decomposition of the BART: Assessment and application. Journal of Mathematical Psychology, 55(1), 94-105.

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Balloon Analogue Risk Task, there should be 3 columns of data with the labels “subjID”, “pumps”, “explosion”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “pumps”: The number of pumps.
  • “explosion”: 0: intact, 1: burst

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “pumps”, “explosion”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘bart_par4’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- bart_par4(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.cgt_cm(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Cambridge Gambling Task - Cumulative Model

Hierarchical Bayesian Modeling of the Cambridge Gambling Task [Rogers1999] using Cumulative Model with the following parameters: “alpha” (probability distortion), “c” (color bias), “rho” (relative loss sensitivity), “beta” (discounting rate), “gamma” (choice sensitivity).

[Rogers1999]Rogers, R. D., Everitt, B. J., Baldacchino, A., Blackshaw, A. J., Swainson, R., Wynne, K., Baker, N. B., Hunter, J., Carthy, T., London, M., Deakin, J. F. W., Sahakian, B. J., Robbins, T. W. (1999). Dissociable deficits in the decision-making cognition of chronic amphetamine abusers, opiate abusers, patients with focal damage to prefrontal cortex, and tryptophan-depleted normal volunteers: evidence for monoaminergic mechanisms. Neuropsychopharmacology, 20, 322–339.

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Cambridge Gambling Task, there should be 7 columns of data with the labels “subjID”, “gamble_type”, “percentage_staked”, “trial_initial_points”, “assessment_stage”, “red_chosen”, “n_red_boxes”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “gamble_type”: Integer value representng whether the bets on the current trial were presented in descending (0) or ascending (1) order.
  • “percentage_staked”: Integer value representing the bet percentage (not proportion) selected on the current trial: 5, 25, 50, 75, or 95.
  • “trial_initial_points”: Floating point value representing the number of points that the subject has at the start of the current trial (e.g., 100, 150, etc.).
  • “assessment_stage”: Integer value representing whether the current trial is a practice trial (0) or a test trial (1). Only test trials are used for model fitting.
  • “red_chosen”: Integer value representing whether the red color was chosen (1) versus the blue color (0).
  • “n_red_boxes”: Integer value representing the number of red boxes shown on the current trial: 1, 2, 3,…, or 9.

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “gamble_type”, “percentage_staked”, “trial_initial_points”, “assessment_stage”, “red_chosen”, “n_red_boxes”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. For this model they are: “y_hat_col”, “y_hat_bet”, “bet_utils”.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – (Currently not available.) Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘cgt_cm’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.
  • model_regressor: Dict holding the extracted model-based regressors.

Examples

# Run the model and store results in "output"
output <- cgt_cm(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.choiceRT_ddm(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Choice Reaction Time Task - Drift Diffusion Model

Hierarchical Bayesian Modeling of the Choice Reaction Time Task using Drift Diffusion Model [Ratcliff1978] with the following parameters: “alpha” (boundary separation), “beta” (bias), “delta” (drift rate), “tau” (non-decision time).

Note

Note that this implementation is NOT the full Drift Diffusion Model as described in Ratcliff (1978). This implementation estimates the drift rate, boundary separation, starting point, and non-decision time; but not the between- and within-trial variances in these parameters.

Note

Code for this model is based on codes/comments by Guido Biele, Joseph Burling, Andrew Ellis, and potential others @ Stan mailing.

[Ratcliff1978]Ratcliff, R. (1978). A theory of memory retrieval. Psychological Review, 85(2), 59-108. http://doi.org/10.1037/0033-295X.85.2.59

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Choice Reaction Time Task, there should be 3 columns of data with the labels “subjID”, “choice”, “RT”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Choice made for the current trial, coded as 1/2 to indicate lower/upper boundary or left/right choices (e.g., 1 1 1 2 1 2).
  • “RT”: Choice reaction time for the current trial, in seconds (e.g., 0.435 0.383 0.314 0.309, etc.).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “RT”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – (Currently not available.) Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – For this model, it’s possible to set the following model-specific argument to a value that you may prefer.

    • RTbound: Floating point value representing the lower bound (i.e., minimum allowed) reaction time. Defaults to 0.1 (100 milliseconds).

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘choiceRT_ddm’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- choiceRT_ddm(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.choiceRT_ddm_single(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Choice Reaction Time Task - Drift Diffusion Model

Individual Bayesian Modeling of the Choice Reaction Time Task using Drift Diffusion Model [Ratcliff1978] with the following parameters: “alpha” (boundary separation), “beta” (bias), “delta” (drift rate), “tau” (non-decision time).

Note

Note that this implementation is NOT the full Drift Diffusion Model as described in Ratcliff (1978). This implementation estimates the drift rate, boundary separation, starting point, and non-decision time; but not the between- and within-trial variances in these parameters.

Note

Code for this model is based on codes/comments by Guido Biele, Joseph Burling, Andrew Ellis, and potential others @ Stan mailing.

[Ratcliff1978]Ratcliff, R. (1978). A theory of memory retrieval. Psychological Review, 85(2), 59-108. http://doi.org/10.1037/0033-295X.85.2.59

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Choice Reaction Time Task, there should be 3 columns of data with the labels “subjID”, “choice”, “RT”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Choice made for the current trial, coded as 1/2 to indicate lower/upper boundary or left/right choices (e.g., 1 1 1 2 1 2).
  • “RT”: Choice reaction time for the current trial, in seconds (e.g., 0.435 0.383 0.314 0.309, etc.).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “RT”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – (Currently not available.) Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – For this model, it’s possible to set the following model-specific argument to a value that you may prefer.

    • RTbound: Floating point value representing the lower bound (i.e., minimum allowed) reaction time. Defaults to 0.1 (100 milliseconds).

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘choiceRT_ddm_single’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- choiceRT_ddm_single(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.cra_exp(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Choice Under Risk and Ambiguity Task - Exponential Subjective Value Model

Hierarchical Bayesian Modeling of the Choice Under Risk and Ambiguity Task using Exponential Subjective Value Model [Hsu2005] with the following parameters: “alpha” (risk attitude), “beta” (ambiguity attitude), “gamma” (inverse temperature).

[Hsu2005]Hsu, M., Bhatt, M., Adolphs, R., Tranel, D., & Camerer, C. F. (2005). Neural systems responding to degrees of uncertainty in human decision-making. Science, 310(5754), 1680-1683. https://doi.org/10.1126/science.1115327

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Choice Under Risk and Ambiguity Task, there should be 6 columns of data with the labels “subjID”, “prob”, “ambig”, “reward_var”, “reward_fix”, “choice”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “prob”: Objective probability of the variable lottery.
  • “ambig”: Ambiguity level of the variable lottery (0 for risky lottery; greater than 0 for ambiguous lottery).
  • “reward_var”: Amount of reward in variable lottery. Assumed to be greater than zero.
  • “reward_fix”: Amount of reward in fixed lottery. Assumed to be greater than zero.
  • “choice”: If the variable lottery was selected, choice == 1; otherwise choice == 0.

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “prob”, “ambig”, “reward_var”, “reward_fix”, “choice”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. For this model they are: “sv”, “sv_fix”, “sv_var”, “p_var”.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘cra_exp’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.
  • model_regressor: Dict holding the extracted model-based regressors.

Examples

# Run the model and store results in "output"
output <- cra_exp(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.cra_linear(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Choice Under Risk and Ambiguity Task - Linear Subjective Value Model

Hierarchical Bayesian Modeling of the Choice Under Risk and Ambiguity Task using Linear Subjective Value Model [Levy2010] with the following parameters: “alpha” (risk attitude), “beta” (ambiguity attitude), “gamma” (inverse temperature).

[Levy2010]Levy, I., Snell, J., Nelson, A. J., Rustichini, A., & Glimcher, P. W. (2010). Neural representation of subjective value under risk and ambiguity. Journal of Neurophysiology, 103(2), 1036-1047.

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Choice Under Risk and Ambiguity Task, there should be 6 columns of data with the labels “subjID”, “prob”, “ambig”, “reward_var”, “reward_fix”, “choice”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “prob”: Objective probability of the variable lottery.
  • “ambig”: Ambiguity level of the variable lottery (0 for risky lottery; greater than 0 for ambiguous lottery).
  • “reward_var”: Amount of reward in variable lottery. Assumed to be greater than zero.
  • “reward_fix”: Amount of reward in fixed lottery. Assumed to be greater than zero.
  • “choice”: If the variable lottery was selected, choice == 1; otherwise choice == 0.

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “prob”, “ambig”, “reward_var”, “reward_fix”, “choice”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. For this model they are: “sv”, “sv_fix”, “sv_var”, “p_var”.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘cra_linear’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.
  • model_regressor: Dict holding the extracted model-based regressors.

Examples

# Run the model and store results in "output"
output <- cra_linear(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.dbdm_prob_weight(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Description Based Decison Making Task - Probability Weight Function

Hierarchical Bayesian Modeling of the Description Based Decison Making Task using Probability Weight Function [Erev2010], [Hertwig2004], [Jessup2008] with the following parameters: “tau” (probability weight function), “rho” (subject utility function), “lambda” (loss aversion parameter), “beta” (inverse softmax temperature).

[Erev2010]Erev, I., Ert, E., Roth, A. E., Haruvy, E., Herzog, S. M., Hau, R., … & Lebiere, C. (2010). A choice prediction competition: Choices from experience and from description. Journal of Behavioral Decision Making, 23(1), 15-47.
[Hertwig2004]Hertwig, R., Barron, G., Weber, E. U., & Erev, I. (2004). Decisions from experience and the effect of rare events in risky choice. Psychological science, 15(8), 534-539.
[Jessup2008]Jessup, R. K., Bishara, A. J., & Busemeyer, J. R. (2008). Feedback produces divergence from prospect theory in descriptive choice. Psychological Science, 19(10), 1015-1022.

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Description Based Decison Making Task, there should be 8 columns of data with the labels “subjID”, “opt1hprob”, “opt2hprob”, “opt1hval”, “opt1lval”, “opt2hval”, “opt2lval”, “choice”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “opt1hprob”: Possiblity of getting higher value of outcome(opt1hval) when choosing option 1.
  • “opt2hprob”: Possiblity of getting higher value of outcome(opt2hval) when choosing option 2.
  • “opt1hval”: Possible (with opt1hprob probability) outcome of option 1.
  • “opt1lval”: Possible (with (1 - opt1hprob) probability) outcome of option 1.
  • “opt2hval”: Possible (with opt2hprob probability) outcome of option 2.
  • “opt2lval”: Possible (with (1 - opt2hprob) probability) outcome of option 2.
  • “choice”: If option 1 was selected, choice == 1; else if option 2 was selected, choice == 2.

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “opt1hprob”, “opt2hprob”, “opt1hval”, “opt1lval”, “opt2hval”, “opt2lval”, “choice”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘dbdm_prob_weight’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- dbdm_prob_weight(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.dd_cs(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Delay Discounting Task - Constant-Sensitivity (CS) Model

Hierarchical Bayesian Modeling of the Delay Discounting Task using Constant-Sensitivity (CS) Model [Ebert2007] with the following parameters: “r” (exponential discounting rate), “s” (impatience), “beta” (inverse temperature).

[Ebert2007]Ebert, J. E. J., & Prelec, D. (2007). The Fragility of Time: Time-Insensitivity and Valuation of the Near and Far Future. Management Science. http://doi.org/10.1287/mnsc.1060.0671

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Delay Discounting Task, there should be 6 columns of data with the labels “subjID”, “delay_later”, “amount_later”, “delay_sooner”, “amount_sooner”, “choice”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “delay_later”: An integer representing the delayed days for the later option (e.g. 1, 6, 28).
  • “amount_later”: A floating point number representing the amount for the later option (e.g. 10.5, 13.4, 30.9).
  • “delay_sooner”: An integer representing the delayed days for the sooner option (e.g. 0).
  • “amount_sooner”: A floating point number representing the amount for the sooner option (e.g. 10).
  • “choice”: If amount_later was selected, choice == 1; else if amount_sooner was selected, choice == 0.

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “delay_later”, “amount_later”, “delay_sooner”, “amount_sooner”, “choice”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘dd_cs’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- dd_cs(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.dd_cs_single(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Delay Discounting Task - Constant-Sensitivity (CS) Model

Individual Bayesian Modeling of the Delay Discounting Task using Constant-Sensitivity (CS) Model [Ebert2007] with the following parameters: “r” (exponential discounting rate), “s” (impatience), “beta” (inverse temperature).

[Ebert2007]Ebert, J. E. J., & Prelec, D. (2007). The Fragility of Time: Time-Insensitivity and Valuation of the Near and Far Future. Management Science. http://doi.org/10.1287/mnsc.1060.0671

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Delay Discounting Task, there should be 6 columns of data with the labels “subjID”, “delay_later”, “amount_later”, “delay_sooner”, “amount_sooner”, “choice”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “delay_later”: An integer representing the delayed days for the later option (e.g. 1, 6, 28).
  • “amount_later”: A floating point number representing the amount for the later option (e.g. 10.5, 13.4, 30.9).
  • “delay_sooner”: An integer representing the delayed days for the sooner option (e.g. 0).
  • “amount_sooner”: A floating point number representing the amount for the sooner option (e.g. 10).
  • “choice”: If amount_later was selected, choice == 1; else if amount_sooner was selected, choice == 0.

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “delay_later”, “amount_later”, “delay_sooner”, “amount_sooner”, “choice”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘dd_cs_single’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- dd_cs_single(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.dd_exp(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Delay Discounting Task - Exponential Model

Hierarchical Bayesian Modeling of the Delay Discounting Task using Exponential Model [Samuelson1937] with the following parameters: “r” (exponential discounting rate), “beta” (inverse temperature).

[Samuelson1937]Samuelson, P. A. (1937). A Note on Measurement of Utility. The Review of Economic Studies, 4(2), 155. http://doi.org/10.2307/2967612

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Delay Discounting Task, there should be 6 columns of data with the labels “subjID”, “delay_later”, “amount_later”, “delay_sooner”, “amount_sooner”, “choice”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “delay_later”: An integer representing the delayed days for the later option (e.g. 1, 6, 28).
  • “amount_later”: A floating point number representing the amount for the later option (e.g. 10.5, 13.4, 30.9).
  • “delay_sooner”: An integer representing the delayed days for the sooner option (e.g. 0).
  • “amount_sooner”: A floating point number representing the amount for the sooner option (e.g. 10).
  • “choice”: If amount_later was selected, choice == 1; else if amount_sooner was selected, choice == 0.

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “delay_later”, “amount_later”, “delay_sooner”, “amount_sooner”, “choice”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘dd_exp’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- dd_exp(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.dd_hyperbolic(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Delay Discounting Task - Hyperbolic Model

Hierarchical Bayesian Modeling of the Delay Discounting Task using Hyperbolic Model [Mazur1987] with the following parameters: “k” (discounting rate), “beta” (inverse temperature).

[Mazur1987]Mazur, J. E. (1987). An adjustment procedure for studying delayed reinforcement.

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Delay Discounting Task, there should be 6 columns of data with the labels “subjID”, “delay_later”, “amount_later”, “delay_sooner”, “amount_sooner”, “choice”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “delay_later”: An integer representing the delayed days for the later option (e.g. 1, 6, 28).
  • “amount_later”: A floating point number representing the amount for the later option (e.g. 10.5, 13.4, 30.9).
  • “delay_sooner”: An integer representing the delayed days for the sooner option (e.g. 0).
  • “amount_sooner”: A floating point number representing the amount for the sooner option (e.g. 10).
  • “choice”: If amount_later was selected, choice == 1; else if amount_sooner was selected, choice == 0.

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “delay_later”, “amount_later”, “delay_sooner”, “amount_sooner”, “choice”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘dd_hyperbolic’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- dd_hyperbolic(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.dd_hyperbolic_single(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Delay Discounting Task - Hyperbolic Model

Individual Bayesian Modeling of the Delay Discounting Task using Hyperbolic Model [Mazur1987] with the following parameters: “k” (discounting rate), “beta” (inverse temperature).

[Mazur1987]Mazur, J. E. (1987). An adjustment procedure for studying delayed reinforcement.

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Delay Discounting Task, there should be 6 columns of data with the labels “subjID”, “delay_later”, “amount_later”, “delay_sooner”, “amount_sooner”, “choice”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “delay_later”: An integer representing the delayed days for the later option (e.g. 1, 6, 28).
  • “amount_later”: A floating point number representing the amount for the later option (e.g. 10.5, 13.4, 30.9).
  • “delay_sooner”: An integer representing the delayed days for the sooner option (e.g. 0).
  • “amount_sooner”: A floating point number representing the amount for the sooner option (e.g. 10).
  • “choice”: If amount_later was selected, choice == 1; else if amount_sooner was selected, choice == 0.

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “delay_later”, “amount_later”, “delay_sooner”, “amount_sooner”, “choice”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘dd_hyperbolic_single’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- dd_hyperbolic_single(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.gng_m1(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Orthogonalized Go/Nogo Task - RW + noise

Hierarchical Bayesian Modeling of the Orthogonalized Go/Nogo Task using RW + noise [Guitart-Masip2012] with the following parameters: “xi” (noise), “ep” (learning rate), “rho” (effective size).

[Guitart-Masip2012]Guitart-Masip, M., Huys, Q. J. M., Fuentemilla, L., Dayan, P., Duzel, E., & Dolan, R. J. (2012). Go and no-go learning in reward and punishment: Interactions between affect and effect. Neuroimage, 62(1), 154-166. http://doi.org/10.1016/j.neuroimage.2012.04.024

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Orthogonalized Go/Nogo Task, there should be 4 columns of data with the labels “subjID”, “cue”, “keyPressed”, “outcome”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “cue”: Nominal integer representing the cue shown for that trial: 1, 2, 3, or 4.
  • “keyPressed”: Binary value representing the subject’s response for that trial (where Press == 1; No press == 0).
  • “outcome”: Ternary value representing the outcome of that trial (where Positive feedback == 1; Neutral feedback == 0; Negative feedback == -1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “cue”, “keyPressed”, “outcome”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. For this model they are: “Qgo”, “Qnogo”, “Wgo”, “Wnogo”.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘gng_m1’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.
  • model_regressor: Dict holding the extracted model-based regressors.

Examples

# Run the model and store results in "output"
output <- gng_m1(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.gng_m2(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Orthogonalized Go/Nogo Task - RW + noise + bias

Hierarchical Bayesian Modeling of the Orthogonalized Go/Nogo Task using RW + noise + bias [Guitart-Masip2012] with the following parameters: “xi” (noise), “ep” (learning rate), “b” (action bias), “rho” (effective size).

[Guitart-Masip2012]Guitart-Masip, M., Huys, Q. J. M., Fuentemilla, L., Dayan, P., Duzel, E., & Dolan, R. J. (2012). Go and no-go learning in reward and punishment: Interactions between affect and effect. Neuroimage, 62(1), 154-166. http://doi.org/10.1016/j.neuroimage.2012.04.024

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Orthogonalized Go/Nogo Task, there should be 4 columns of data with the labels “subjID”, “cue”, “keyPressed”, “outcome”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “cue”: Nominal integer representing the cue shown for that trial: 1, 2, 3, or 4.
  • “keyPressed”: Binary value representing the subject’s response for that trial (where Press == 1; No press == 0).
  • “outcome”: Ternary value representing the outcome of that trial (where Positive feedback == 1; Neutral feedback == 0; Negative feedback == -1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “cue”, “keyPressed”, “outcome”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. For this model they are: “Qgo”, “Qnogo”, “Wgo”, “Wnogo”.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘gng_m2’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.
  • model_regressor: Dict holding the extracted model-based regressors.

Examples

# Run the model and store results in "output"
output <- gng_m2(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.gng_m3(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Orthogonalized Go/Nogo Task - RW + noise + bias + pi

Hierarchical Bayesian Modeling of the Orthogonalized Go/Nogo Task using RW + noise + bias + pi [Guitart-Masip2012] with the following parameters: “xi” (noise), “ep” (learning rate), “b” (action bias), “pi” (Pavlovian bias), “rho” (effective size).

[Guitart-Masip2012]Guitart-Masip, M., Huys, Q. J. M., Fuentemilla, L., Dayan, P., Duzel, E., & Dolan, R. J. (2012). Go and no-go learning in reward and punishment: Interactions between affect and effect. Neuroimage, 62(1), 154-166. http://doi.org/10.1016/j.neuroimage.2012.04.024

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Orthogonalized Go/Nogo Task, there should be 4 columns of data with the labels “subjID”, “cue”, “keyPressed”, “outcome”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “cue”: Nominal integer representing the cue shown for that trial: 1, 2, 3, or 4.
  • “keyPressed”: Binary value representing the subject’s response for that trial (where Press == 1; No press == 0).
  • “outcome”: Ternary value representing the outcome of that trial (where Positive feedback == 1; Neutral feedback == 0; Negative feedback == -1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “cue”, “keyPressed”, “outcome”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. For this model they are: “Qgo”, “Qnogo”, “Wgo”, “Wnogo”, “SV”.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘gng_m3’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.
  • model_regressor: Dict holding the extracted model-based regressors.

Examples

# Run the model and store results in "output"
output <- gng_m3(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.gng_m4(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Orthogonalized Go/Nogo Task - RW (rew/pun) + noise + bias + pi

Hierarchical Bayesian Modeling of the Orthogonalized Go/Nogo Task using RW (rew/pun) + noise + bias + pi [Cavanagh2013] with the following parameters: “xi” (noise), “ep” (learning rate), “b” (action bias), “pi” (Pavlovian bias), “rhoRew” (reward sensitivity), “rhoPun” (punishment sensitivity).

[Cavanagh2013]Cavanagh, J. F., Eisenberg, I., Guitart-Masip, M., Huys, Q., & Frank, M. J. (2013). Frontal Theta Overrides Pavlovian Learning Biases. Journal of Neuroscience, 33(19), 8541-8548. http://doi.org/10.1523/JNEUROSCI.5754-12.2013

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Orthogonalized Go/Nogo Task, there should be 4 columns of data with the labels “subjID”, “cue”, “keyPressed”, “outcome”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “cue”: Nominal integer representing the cue shown for that trial: 1, 2, 3, or 4.
  • “keyPressed”: Binary value representing the subject’s response for that trial (where Press == 1; No press == 0).
  • “outcome”: Ternary value representing the outcome of that trial (where Positive feedback == 1; Neutral feedback == 0; Negative feedback == -1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “cue”, “keyPressed”, “outcome”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. For this model they are: “Qgo”, “Qnogo”, “Wgo”, “Wnogo”, “SV”.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘gng_m4’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.
  • model_regressor: Dict holding the extracted model-based regressors.

Examples

# Run the model and store results in "output"
output <- gng_m4(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.igt_orl(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Iowa Gambling Task - Outcome-Representation Learning Model

Hierarchical Bayesian Modeling of the Iowa Gambling Task [Ahn2008] using Outcome-Representation Learning Model [Haines2018] with the following parameters: “Arew” (reward learning rate), “Apun” (punishment learning rate), “K” (perseverance decay), “betaF” (outcome frequency weight), “betaP” (perseverance weight).

[Ahn2008]Ahn, W. Y., Busemeyer, J. R., & Wagenmakers, E. J. (2008). Comparison of decision learning models using the generalization criterion method. Cognitive Science, 32(8), 1376-1402. http://doi.org/10.1080/03640210802352992
[Haines2018]Haines, N., Vassileva, J., & Ahn, W.-Y. (2018). The Outcome-Representation Learning Model: A Novel Reinforcement Learning Model of the Iowa Gambling Task. Cognitive Science. https://doi.org/10.1111/cogs.12688

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Iowa Gambling Task, there should be 4 columns of data with the labels “subjID”, “choice”, “gain”, “loss”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer indicating which deck was chosen on that trial (where A==1, B==2, C==3, and D==4).
  • “gain”: Floating point value representing the amount of currency won on that trial (e.g. 50, 100).
  • “loss”: Floating point value representing the amount of currency lost on that trial (e.g. 0, -50).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “gain”, “loss”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – For this model, it’s possible to set the following model-specific argument to a value that you may prefer.

    • payscale: Raw payoffs within data are divided by this number. Used for scaling data. Defaults to 100.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘igt_orl’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- igt_orl(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.igt_pvl_decay(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Iowa Gambling Task - Prospect Valence Learning (PVL) Decay-RI

Hierarchical Bayesian Modeling of the Iowa Gambling Task [Ahn2008] using Prospect Valence Learning (PVL) Decay-RI [Ahn2014] with the following parameters: “A” (decay rate), “alpha” (outcome sensitivity), “cons” (response consistency), “lambda” (loss aversion).

[Ahn2008]Ahn, W. Y., Busemeyer, J. R., & Wagenmakers, E. J. (2008). Comparison of decision learning models using the generalization criterion method. Cognitive Science, 32(8), 1376-1402. http://doi.org/10.1080/03640210802352992
[Ahn2014]Ahn, W.-Y., Vasilev, G., Lee, S.-H., Busemeyer, J. R., Kruschke, J. K., Bechara, A., & Vassileva, J. (2014). Decision-making in stimulant and opiate addicts in protracted abstinence: evidence from computational modeling with pure users. Frontiers in Psychology, 5, 1376. http://doi.org/10.3389/fpsyg.2014.00849

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Iowa Gambling Task, there should be 4 columns of data with the labels “subjID”, “choice”, “gain”, “loss”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer indicating which deck was chosen on that trial (where A==1, B==2, C==3, and D==4).
  • “gain”: Floating point value representing the amount of currency won on that trial (e.g. 50, 100).
  • “loss”: Floating point value representing the amount of currency lost on that trial (e.g. 0, -50).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “gain”, “loss”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – For this model, it’s possible to set the following model-specific argument to a value that you may prefer.

    • payscale: Raw payoffs within data are divided by this number. Used for scaling data. Defaults to 100.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘igt_pvl_decay’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- igt_pvl_decay(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.igt_pvl_delta(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Iowa Gambling Task - Prospect Valence Learning (PVL) Delta

Hierarchical Bayesian Modeling of the Iowa Gambling Task [Ahn2008] using Prospect Valence Learning (PVL) Delta [Ahn2008] with the following parameters: “A” (learning rate), “alpha” (outcome sensitivity), “cons” (response consistency), “lambda” (loss aversion).

[Ahn2008]Ahn, W. Y., Busemeyer, J. R., & Wagenmakers, E. J. (2008). Comparison of decision learning models using the generalization criterion method. Cognitive Science, 32(8), 1376-1402. http://doi.org/10.1080/03640210802352992

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Iowa Gambling Task, there should be 4 columns of data with the labels “subjID”, “choice”, “gain”, “loss”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer indicating which deck was chosen on that trial (where A==1, B==2, C==3, and D==4).
  • “gain”: Floating point value representing the amount of currency won on that trial (e.g. 50, 100).
  • “loss”: Floating point value representing the amount of currency lost on that trial (e.g. 0, -50).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “gain”, “loss”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – For this model, it’s possible to set the following model-specific argument to a value that you may prefer.

    • payscale: Raw payoffs within data are divided by this number. Used for scaling data. Defaults to 100.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘igt_pvl_delta’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- igt_pvl_delta(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.igt_vpp(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Iowa Gambling Task - Value-Plus-Perseverance

Hierarchical Bayesian Modeling of the Iowa Gambling Task [Ahn2008] using Value-Plus-Perseverance [Worthy2013] with the following parameters: “A” (learning rate), “alpha” (outcome sensitivity), “cons” (response consistency), “lambda” (loss aversion), “epP” (gain impact), “epN” (loss impact), “K” (decay rate), “w” (RL weight).

[Ahn2008]Ahn, W. Y., Busemeyer, J. R., & Wagenmakers, E. J. (2008). Comparison of decision learning models using the generalization criterion method. Cognitive Science, 32(8), 1376-1402. http://doi.org/10.1080/03640210802352992
[Worthy2013]Worthy, D. A., & Todd Maddox, W. (2013). A comparison model of reinforcement-learning and win-stay-lose-shift decision-making processes: A tribute to W.K. Estes. Journal of Mathematical Psychology, 59, 41-49. http://doi.org/10.1016/j.jmp.2013.10.001

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Iowa Gambling Task, there should be 4 columns of data with the labels “subjID”, “choice”, “gain”, “loss”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer indicating which deck was chosen on that trial (where A==1, B==2, C==3, and D==4).
  • “gain”: Floating point value representing the amount of currency won on that trial (e.g. 50, 100).
  • “loss”: Floating point value representing the amount of currency lost on that trial (e.g. 0, -50).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “gain”, “loss”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – For this model, it’s possible to set the following model-specific argument to a value that you may prefer.

    • payscale: Raw payoffs within data are divided by this number. Used for scaling data. Defaults to 100.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘igt_vpp’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- igt_vpp(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.peer_ocu(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Peer Influence Task - Other-Conferred Utility (OCU) Model

Hierarchical Bayesian Modeling of the Peer Influence Task [Chung2015] using Other-Conferred Utility (OCU) Model with the following parameters: “rho” (risk preference), “tau” (inverse temperature), “ocu” (other-conferred utility).

[Chung2015]Chung, D., Christopoulos, G. I., King-Casas, B., Ball, S. B., & Chiu, P. H. (2015). Social signals of safety and risk confer utility and have asymmetric effects on observers’ choices. Nature Neuroscience, 18(6), 912-916.

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Peer Influence Task, there should be 8 columns of data with the labels “subjID”, “condition”, “p_gamble”, “safe_Hpayoff”, “safe_Lpayoff”, “risky_Hpayoff”, “risky_Lpayoff”, “choice”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “condition”: 0: solo, 1: info (safe/safe), 2: info (mix), 3: info (risky/risky).
  • “p_gamble”: Probability of receiving a high payoff (same for both options).
  • “safe_Hpayoff”: High payoff of the safe option.
  • “safe_Lpayoff”: Low payoff of the safe option.
  • “risky_Hpayoff”: High payoff of the risky option.
  • “risky_Lpayoff”: Low payoff of the risky option.
  • “choice”: Which option was chosen? 0: safe, 1: risky.

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “condition”, “p_gamble”, “safe_Hpayoff”, “safe_Lpayoff”, “risky_Hpayoff”, “risky_Lpayoff”, “choice”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘peer_ocu’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- peer_ocu(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.prl_ewa(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Probabilistic Reversal Learning Task - Experience-Weighted Attraction Model

Hierarchical Bayesian Modeling of the Probabilistic Reversal Learning Task using Experience-Weighted Attraction Model [Ouden2013] with the following parameters: “phi” (1 - learning rate), “rho” (experience decay factor), “beta” (inverse temperature).

[Ouden2013]Ouden, den, H. E. M., Daw, N. D., Fernandez, G., Elshout, J. A., Rijpkema, M., Hoogman, M., et al. (2013). Dissociable Effects of Dopamine and Serotonin on Reversal Learning. Neuron, 80(4), 1090-1100. http://doi.org/10.1016/j.neuron.2013.08.030

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Probabilistic Reversal Learning Task, there should be 3 columns of data with the labels “subjID”, “choice”, “outcome”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer value representing the option chosen on that trial: 1 or 2.
  • “outcome”: Integer value representing the outcome of that trial (where reward == 1, and loss == -1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “outcome”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. For this model they are: “ev_c”, “ev_nc”, “ew_c”, “ew_nc”.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘prl_ewa’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.
  • model_regressor: Dict holding the extracted model-based regressors.

Examples

# Run the model and store results in "output"
output <- prl_ewa(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.prl_fictitious(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Probabilistic Reversal Learning Task - Fictitious Update Model

Hierarchical Bayesian Modeling of the Probabilistic Reversal Learning Task using Fictitious Update Model [Glascher2009] with the following parameters: “eta” (learning rate), “alpha” (indecision point), “beta” (inverse temperature).

[Glascher2009]Glascher, J., Hampton, A. N., & O’Doherty, J. P. (2009). Determining a Role for Ventromedial Prefrontal Cortex in Encoding Action-Based Value Signals During Reward-Related Decision Making. Cerebral Cortex, 19(2), 483-495. http://doi.org/10.1093/cercor/bhn098

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Probabilistic Reversal Learning Task, there should be 3 columns of data with the labels “subjID”, “choice”, “outcome”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer value representing the option chosen on that trial: 1 or 2.
  • “outcome”: Integer value representing the outcome of that trial (where reward == 1, and loss == -1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “outcome”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. For this model they are: “ev_c”, “ev_nc”, “pe_c”, “pe_nc”, “dv”.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘prl_fictitious’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.
  • model_regressor: Dict holding the extracted model-based regressors.

Examples

# Run the model and store results in "output"
output <- prl_fictitious(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.prl_fictitious_multipleB(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Probabilistic Reversal Learning Task - Fictitious Update Model

Multiple-Block Hierarchical Bayesian Modeling of the Probabilistic Reversal Learning Task using Fictitious Update Model [Glascher2009] with the following parameters: “eta” (learning rate), “alpha” (indecision point), “beta” (inverse temperature).

[Glascher2009]Glascher, J., Hampton, A. N., & O’Doherty, J. P. (2009). Determining a Role for Ventromedial Prefrontal Cortex in Encoding Action-Based Value Signals During Reward-Related Decision Making. Cerebral Cortex, 19(2), 483-495. http://doi.org/10.1093/cercor/bhn098

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Probabilistic Reversal Learning Task, there should be 4 columns of data with the labels “subjID”, “block”, “choice”, “outcome”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “block”: A unique identifier for each of the multiple blocks within each subject.
  • “choice”: Integer value representing the option chosen on that trial: 1 or 2.
  • “outcome”: Integer value representing the outcome of that trial (where reward == 1, and loss == -1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “block”, “choice”, “outcome”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. For this model they are: “ev_c”, “ev_nc”, “pe_c”, “pe_nc”, “dv”.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘prl_fictitious_multipleB’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.
  • model_regressor: Dict holding the extracted model-based regressors.

Examples

# Run the model and store results in "output"
output <- prl_fictitious_multipleB(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.prl_fictitious_rp(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Probabilistic Reversal Learning Task - Fictitious Update Model, with separate learning rates for positive and negative prediction error (PE)

Hierarchical Bayesian Modeling of the Probabilistic Reversal Learning Task using Fictitious Update Model, with separate learning rates for positive and negative prediction error (PE) [Glascher2009], [Ouden2013] with the following parameters: “eta_pos” (learning rate, +PE), “eta_neg” (learning rate, -PE), “alpha” (indecision point), “beta” (inverse temperature).

[Glascher2009]Glascher, J., Hampton, A. N., & O’Doherty, J. P. (2009). Determining a Role for Ventromedial Prefrontal Cortex in Encoding Action-Based Value Signals During Reward-Related Decision Making. Cerebral Cortex, 19(2), 483-495. http://doi.org/10.1093/cercor/bhn098
[Ouden2013]Ouden, den, H. E. M., Daw, N. D., Fernandez, G., Elshout, J. A., Rijpkema, M., Hoogman, M., et al. (2013). Dissociable Effects of Dopamine and Serotonin on Reversal Learning. Neuron, 80(4), 1090-1100. http://doi.org/10.1016/j.neuron.2013.08.030

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Probabilistic Reversal Learning Task, there should be 3 columns of data with the labels “subjID”, “choice”, “outcome”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer value representing the option chosen on that trial: 1 or 2.
  • “outcome”: Integer value representing the outcome of that trial (where reward == 1, and loss == -1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “outcome”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. For this model they are: “ev_c”, “ev_nc”, “pe_c”, “pe_nc”, “dv”.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘prl_fictitious_rp’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.
  • model_regressor: Dict holding the extracted model-based regressors.

Examples

# Run the model and store results in "output"
output <- prl_fictitious_rp(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.prl_fictitious_rp_woa(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Probabilistic Reversal Learning Task - Fictitious Update Model, with separate learning rates for positive and negative prediction error (PE), without alpha (indecision point)

Hierarchical Bayesian Modeling of the Probabilistic Reversal Learning Task using Fictitious Update Model, with separate learning rates for positive and negative prediction error (PE), without alpha (indecision point) [Glascher2009], [Ouden2013] with the following parameters: “eta_pos” (learning rate, +PE), “eta_neg” (learning rate, -PE), “beta” (inverse temperature).

[Glascher2009]Glascher, J., Hampton, A. N., & O’Doherty, J. P. (2009). Determining a Role for Ventromedial Prefrontal Cortex in Encoding Action-Based Value Signals During Reward-Related Decision Making. Cerebral Cortex, 19(2), 483-495. http://doi.org/10.1093/cercor/bhn098
[Ouden2013]Ouden, den, H. E. M., Daw, N. D., Fernandez, G., Elshout, J. A., Rijpkema, M., Hoogman, M., et al. (2013). Dissociable Effects of Dopamine and Serotonin on Reversal Learning. Neuron, 80(4), 1090-1100. http://doi.org/10.1016/j.neuron.2013.08.030

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Probabilistic Reversal Learning Task, there should be 3 columns of data with the labels “subjID”, “choice”, “outcome”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer value representing the option chosen on that trial: 1 or 2.
  • “outcome”: Integer value representing the outcome of that trial (where reward == 1, and loss == -1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “outcome”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. For this model they are: “ev_c”, “ev_nc”, “pe_c”, “pe_nc”, “dv”.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘prl_fictitious_rp_woa’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.
  • model_regressor: Dict holding the extracted model-based regressors.

Examples

# Run the model and store results in "output"
output <- prl_fictitious_rp_woa(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.prl_fictitious_woa(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Probabilistic Reversal Learning Task - Fictitious Update Model, without alpha (indecision point)

Hierarchical Bayesian Modeling of the Probabilistic Reversal Learning Task using Fictitious Update Model, without alpha (indecision point) [Glascher2009] with the following parameters: “eta” (learning rate), “beta” (inverse temperature).

[Glascher2009]Glascher, J., Hampton, A. N., & O’Doherty, J. P. (2009). Determining a Role for Ventromedial Prefrontal Cortex in Encoding Action-Based Value Signals During Reward-Related Decision Making. Cerebral Cortex, 19(2), 483-495. http://doi.org/10.1093/cercor/bhn098

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Probabilistic Reversal Learning Task, there should be 3 columns of data with the labels “subjID”, “choice”, “outcome”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer value representing the option chosen on that trial: 1 or 2.
  • “outcome”: Integer value representing the outcome of that trial (where reward == 1, and loss == -1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “outcome”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. For this model they are: “ev_c”, “ev_nc”, “pe_c”, “pe_nc”, “dv”.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘prl_fictitious_woa’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.
  • model_regressor: Dict holding the extracted model-based regressors.

Examples

# Run the model and store results in "output"
output <- prl_fictitious_woa(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.prl_rp(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Probabilistic Reversal Learning Task - Reward-Punishment Model

Hierarchical Bayesian Modeling of the Probabilistic Reversal Learning Task using Reward-Punishment Model [Ouden2013] with the following parameters: “Apun” (punishment learning rate), “Arew” (reward learning rate), “beta” (inverse temperature).

[Ouden2013]Ouden, den, H. E. M., Daw, N. D., Fernandez, G., Elshout, J. A., Rijpkema, M., Hoogman, M., et al. (2013). Dissociable Effects of Dopamine and Serotonin on Reversal Learning. Neuron, 80(4), 1090-1100. http://doi.org/10.1016/j.neuron.2013.08.030

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Probabilistic Reversal Learning Task, there should be 3 columns of data with the labels “subjID”, “choice”, “outcome”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer value representing the option chosen on that trial: 1 or 2.
  • “outcome”: Integer value representing the outcome of that trial (where reward == 1, and loss == -1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “outcome”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. For this model they are: “ev_c”, “ev_nc”, “pe”.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘prl_rp’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.
  • model_regressor: Dict holding the extracted model-based regressors.

Examples

# Run the model and store results in "output"
output <- prl_rp(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.prl_rp_multipleB(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Probabilistic Reversal Learning Task - Reward-Punishment Model

Multiple-Block Hierarchical Bayesian Modeling of the Probabilistic Reversal Learning Task using Reward-Punishment Model [Ouden2013] with the following parameters: “Apun” (punishment learning rate), “Arew” (reward learning rate), “beta” (inverse temperature).

[Ouden2013]Ouden, den, H. E. M., Daw, N. D., Fernandez, G., Elshout, J. A., Rijpkema, M., Hoogman, M., et al. (2013). Dissociable Effects of Dopamine and Serotonin on Reversal Learning. Neuron, 80(4), 1090-1100. http://doi.org/10.1016/j.neuron.2013.08.030

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Probabilistic Reversal Learning Task, there should be 4 columns of data with the labels “subjID”, “block”, “choice”, “outcome”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “block”: A unique identifier for each of the multiple blocks within each subject.
  • “choice”: Integer value representing the option chosen on that trial: 1 or 2.
  • “outcome”: Integer value representing the outcome of that trial (where reward == 1, and loss == -1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “block”, “choice”, “outcome”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. For this model they are: “ev_c”, “ev_nc”, “pe”.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘prl_rp_multipleB’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.
  • model_regressor: Dict holding the extracted model-based regressors.

Examples

# Run the model and store results in "output"
output <- prl_rp_multipleB(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.pst_gainloss_Q(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Probabilistic Selection Task - Gain-Loss Q Learning Model

Hierarchical Bayesian Modeling of the Probabilistic Selection Task using Gain-Loss Q Learning Model [Frank2007] with the following parameters: “alpha_pos” (learning rate for positive feedbacks), “alpha_neg” (learning rate for negative feedbacks), “beta” (inverse temperature).

[Frank2007]Frank, M. J., Moustafa, A. A., Haughey, H. M., Curran, T., & Hutchison, K. E. (2007). Genetic triple dissociation reveals multiple roles for dopamine in reinforcement learning. Proceedings of the National Academy of Sciences, 104(41), 16311-16316.

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Probabilistic Selection Task, there should be 4 columns of data with the labels “subjID”, “type”, “choice”, “reward”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “type”: Two-digit number indicating which pair of stimuli were presented for that trial, e.g. 12, 34, or 56. The digit on the left (tens-digit) indicates the presented stimulus for option1, while the digit on the right (ones-digit) indicates that for option2. Code for each stimulus type (1~6) is defined as for 80% (type 1), 20% (type 2), 70% (type 3), 30% (type 4), 60% (type 5), 40% (type 6). The modeling will still work even if different probabilities are used for the stimuli; however, the total number of stimuli should be less than or equal to 6.
  • “choice”: Whether the subject chose the left option (option1) out of the given two options (i.e. if option1 was chosen, 1; if option2 was chosen, 0).
  • “reward”: Amount of reward earned as a result of the trial.

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “type”, “choice”, “reward”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘pst_gainloss_Q’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- pst_gainloss_Q(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.ra_noLA(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Risk Aversion Task - Prospect Theory, without loss aversion (LA) parameter

Hierarchical Bayesian Modeling of the Risk Aversion Task using Prospect Theory, without loss aversion (LA) parameter [Sokol-Hessner2009] with the following parameters: “rho” (risk aversion), “tau” (inverse temperature).

[Sokol-Hessner2009]Sokol-Hessner, P., Hsu, M., Curley, N. G., Delgado, M. R., Camerer, C. F., Phelps, E. A., & Smith, E. E. (2009). Thinking like a Trader Selectively Reduces Individuals’ Loss Aversion. Proceedings of the National Academy of Sciences of the United States of America, 106(13), 5035-5040. http://www.pnas.org/content/106/13/5035

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Risk Aversion Task, there should be 5 columns of data with the labels “subjID”, “gain”, “loss”, “cert”, “gamble”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “gain”: Possible (50%) gain outcome of a risky option (e.g. 9).
  • “loss”: Possible (50%) loss outcome of a risky option (e.g. 5, or -5).
  • “cert”: Guaranteed amount of a safe option. “cert” is assumed to be zero or greater than zero.
  • “gamble”: If gamble was taken, gamble == 1; else gamble == 0.

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “gain”, “loss”, “cert”, “gamble”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘ra_noLA’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- ra_noLA(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.ra_noRA(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Risk Aversion Task - Prospect Theory, without risk aversion (RA) parameter

Hierarchical Bayesian Modeling of the Risk Aversion Task using Prospect Theory, without risk aversion (RA) parameter [Sokol-Hessner2009] with the following parameters: “lambda” (loss aversion), “tau” (inverse temperature).

[Sokol-Hessner2009]Sokol-Hessner, P., Hsu, M., Curley, N. G., Delgado, M. R., Camerer, C. F., Phelps, E. A., & Smith, E. E. (2009). Thinking like a Trader Selectively Reduces Individuals’ Loss Aversion. Proceedings of the National Academy of Sciences of the United States of America, 106(13), 5035-5040. http://www.pnas.org/content/106/13/5035

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Risk Aversion Task, there should be 5 columns of data with the labels “subjID”, “gain”, “loss”, “cert”, “gamble”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “gain”: Possible (50%) gain outcome of a risky option (e.g. 9).
  • “loss”: Possible (50%) loss outcome of a risky option (e.g. 5, or -5).
  • “cert”: Guaranteed amount of a safe option. “cert” is assumed to be zero or greater than zero.
  • “gamble”: If gamble was taken, gamble == 1; else gamble == 0.

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “gain”, “loss”, “cert”, “gamble”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘ra_noRA’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- ra_noRA(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.ra_prospect(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Risk Aversion Task - Prospect Theory

Hierarchical Bayesian Modeling of the Risk Aversion Task using Prospect Theory [Sokol-Hessner2009] with the following parameters: “rho” (risk aversion), “lambda” (loss aversion), “tau” (inverse temperature).

[Sokol-Hessner2009]Sokol-Hessner, P., Hsu, M., Curley, N. G., Delgado, M. R., Camerer, C. F., Phelps, E. A., & Smith, E. E. (2009). Thinking like a Trader Selectively Reduces Individuals’ Loss Aversion. Proceedings of the National Academy of Sciences of the United States of America, 106(13), 5035-5040. http://www.pnas.org/content/106/13/5035

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Risk Aversion Task, there should be 5 columns of data with the labels “subjID”, “gain”, “loss”, “cert”, “gamble”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “gain”: Possible (50%) gain outcome of a risky option (e.g. 9).
  • “loss”: Possible (50%) loss outcome of a risky option (e.g. 5, or -5).
  • “cert”: Guaranteed amount of a safe option. “cert” is assumed to be zero or greater than zero.
  • “gamble”: If gamble was taken, gamble == 1; else gamble == 0.

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “gain”, “loss”, “cert”, “gamble”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘ra_prospect’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- ra_prospect(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.rdt_happiness(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Risky Decision Task - Happiness Computational Model

Hierarchical Bayesian Modeling of the Risky Decision Task using Happiness Computational Model [Rutledge2014] with the following parameters: “w0” (baseline), “w1” (weight of certain rewards), “w2” (weight of expected values), “w3” (weight of reward prediction errors), “gam” (forgetting factor), “sig” (standard deviation of error).

[Rutledge2014]Rutledge, R. B., Skandali, N., Dayan, P., & Dolan, R. J. (2014). A computational and neural model of momentary subjective well-being. Proceedings of the National Academy of Sciences, 111(33), 12252-12257.

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Risky Decision Task, there should be 9 columns of data with the labels “subjID”, “gain”, “loss”, “cert”, “type”, “gamble”, “outcome”, “happy”, “RT_happy”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “gain”: Possible (50%) gain outcome of a risky option (e.g. 9).
  • “loss”: Possible (50%) loss outcome of a risky option (e.g. 5, or -5).
  • “cert”: Guaranteed amount of a safe option.
  • “type”: loss == -1, mixed == 0, gain == 1
  • “gamble”: If gamble was taken, gamble == 1; else gamble == 0.
  • “outcome”: Result of the trial.
  • “happy”: Happiness score.
  • “RT_happy”: Reaction time for answering the happiness score.

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “gain”, “loss”, “cert”, “type”, “gamble”, “outcome”, “happy”, “RT_happy”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘rdt_happiness’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- rdt_happiness(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.ts_par4(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Two-Step Task - Hybrid Model, with 4 parameters

Hierarchical Bayesian Modeling of the Two-Step Task [Daw2011] using Hybrid Model, with 4 parameters [Daw2011], [Wunderlich2012] with the following parameters: “a” (learning rate for both stages 1 & 2), “beta” (inverse temperature for both stages 1 & 2), “pi” (perseverance), “w” (model-based weight).

[Daw2011]Daw, N. D., Gershman, S. J., Seymour, B., Ben Seymour, Dayan, P., & Dolan, R. J. (2011). Model-Based Influences on Humans’ Choices and Striatal Prediction Errors. Neuron, 69(6), 1204-1215. http://doi.org/10.1016/j.neuron.2011.02.027
[Wunderlich2012]Wunderlich, K., Smittenaar, P., & Dolan, R. J. (2012). Dopamine enhances model-based over model-free choice behavior. Neuron, 75(3), 418-424.

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Two-Step Task, there should be 4 columns of data with the labels “subjID”, “level1_choice”, “level2_choice”, “reward”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “level1_choice”: Choice made for Level (Stage) 1 (1: stimulus 1, 2: stimulus 2).
  • “level2_choice”: Choice made for Level (Stage) 2 (1: stimulus 3, 2: stimulus 4, 3: stimulus 5, 4: stimulus 6).
    Note that, in our notation, choosing stimulus 1 in Level 1 leads to stimulus 3 & 4 in Level 2 with a common (0.7 by default) transition. Similarly, choosing stimulus 2 in Level 1 leads to stimulus 5 & 6 in Level 2 with a common (0.7 by default) transition. To change this default transition probability, set the function argument trans_prob to your preferred value.
  • “reward”: Reward after Level 2 (0 or 1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “level1_choice”, “level2_choice”, “reward”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – For this model, it’s possible to set the following model-specific argument to a value that you may prefer.

    • trans_prob: Common state transition probability from Stage (Level) 1 to Stage (Level) 2. Defaults to 0.7.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘ts_par4’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- ts_par4(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.ts_par6(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Two-Step Task - Hybrid Model, with 6 parameters

Hierarchical Bayesian Modeling of the Two-Step Task [Daw2011] using Hybrid Model, with 6 parameters [Daw2011] with the following parameters: “a1” (learning rate in stage 1), “beta1” (inverse temperature in stage 1), “a2” (learning rate in stage 2), “beta2” (inverse temperature in stage 2), “pi” (perseverance), “w” (model-based weight).

[Daw2011]Daw, N. D., Gershman, S. J., Seymour, B., Ben Seymour, Dayan, P., & Dolan, R. J. (2011). Model-Based Influences on Humans’ Choices and Striatal Prediction Errors. Neuron, 69(6), 1204-1215. http://doi.org/10.1016/j.neuron.2011.02.027

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Two-Step Task, there should be 4 columns of data with the labels “subjID”, “level1_choice”, “level2_choice”, “reward”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “level1_choice”: Choice made for Level (Stage) 1 (1: stimulus 1, 2: stimulus 2).
  • “level2_choice”: Choice made for Level (Stage) 2 (1: stimulus 3, 2: stimulus 4, 3: stimulus 5, 4: stimulus 6).
    Note that, in our notation, choosing stimulus 1 in Level 1 leads to stimulus 3 & 4 in Level 2 with a common (0.7 by default) transition. Similarly, choosing stimulus 2 in Level 1 leads to stimulus 5 & 6 in Level 2 with a common (0.7 by default) transition. To change this default transition probability, set the function argument trans_prob to your preferred value.
  • “reward”: Reward after Level 2 (0 or 1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “level1_choice”, “level2_choice”, “reward”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – For this model, it’s possible to set the following model-specific argument to a value that you may prefer.

    • trans_prob: Common state transition probability from Stage (Level) 1 to Stage (Level) 2. Defaults to 0.7.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘ts_par6’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- ts_par6(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.ts_par7(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Two-Step Task - Hybrid Model, with 7 parameters (original model)

Hierarchical Bayesian Modeling of the Two-Step Task [Daw2011] using Hybrid Model, with 7 parameters (original model) [Daw2011] with the following parameters: “a1” (learning rate in stage 1), “beta1” (inverse temperature in stage 1), “a2” (learning rate in stage 2), “beta2” (inverse temperature in stage 2), “pi” (perseverance), “w” (model-based weight), “lambda” (eligibility trace).

[Daw2011]Daw, N. D., Gershman, S. J., Seymour, B., Ben Seymour, Dayan, P., & Dolan, R. J. (2011). Model-Based Influences on Humans’ Choices and Striatal Prediction Errors. Neuron, 69(6), 1204-1215. http://doi.org/10.1016/j.neuron.2011.02.027

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Two-Step Task, there should be 4 columns of data with the labels “subjID”, “level1_choice”, “level2_choice”, “reward”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “level1_choice”: Choice made for Level (Stage) 1 (1: stimulus 1, 2: stimulus 2).
  • “level2_choice”: Choice made for Level (Stage) 2 (1: stimulus 3, 2: stimulus 4, 3: stimulus 5, 4: stimulus 6).
    Note that, in our notation, choosing stimulus 1 in Level 1 leads to stimulus 3 & 4 in Level 2 with a common (0.7 by default) transition. Similarly, choosing stimulus 2 in Level 1 leads to stimulus 5 & 6 in Level 2 with a common (0.7 by default) transition. To change this default transition probability, set the function argument trans_prob to your preferred value.
  • “reward”: Reward after Level 2 (0 or 1).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “level1_choice”, “level2_choice”, “reward”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – For this model, it’s possible to set the following model-specific argument to a value that you may prefer.

    • trans_prob: Common state transition probability from Stage (Level) 1 to Stage (Level) 2. Defaults to 0.7.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘ts_par7’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- ts_par7(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.ug_bayes(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Norm-Training Ultimatum Game - Ideal Observer Model

Hierarchical Bayesian Modeling of the Norm-Training Ultimatum Game using Ideal Observer Model [Xiang2013] with the following parameters: “alpha” (envy), “beta” (guilt), “tau” (inverse temperature).

[Xiang2013]Xiang, T., Lohrenz, T., & Montague, P. R. (2013). Computational Substrates of Norms and Their Violations during Social Exchange. Journal of Neuroscience, 33(3), 1099-1108. http://doi.org/10.1523/JNEUROSCI.1642-12.2013

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Norm-Training Ultimatum Game, there should be 3 columns of data with the labels “subjID”, “offer”, “accept”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “offer”: Floating point value representing the offer made in that trial (e.g. 4, 10, 11).
  • “accept”: 1 or 0, indicating whether the offer was accepted in that trial (where accepted == 1, rejected == 0).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “offer”, “accept”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘ug_bayes’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- ug_bayes(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.ug_delta(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Norm-Training Ultimatum Game - Rescorla-Wagner (Delta) Model

Hierarchical Bayesian Modeling of the Norm-Training Ultimatum Game using Rescorla-Wagner (Delta) Model [Gu2015] with the following parameters: “alpha” (envy), “tau” (inverse temperature), “ep” (norm adaptation rate).

[Gu2015]Gu, X., Wang, X., Hula, A., Wang, S., Xu, S., Lohrenz, T. M., et al. (2015). Necessary, Yet Dissociable Contributions of the Insular and Ventromedial Prefrontal Cortices to Norm Adaptation: Computational and Lesion Evidence in Humans. Journal of Neuroscience, 35(2), 467-473. http://doi.org/10.1523/JNEUROSCI.2906-14.2015

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Norm-Training Ultimatum Game, there should be 3 columns of data with the labels “subjID”, “offer”, “accept”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “offer”: Floating point value representing the offer made in that trial (e.g. 4, 10, 11).
  • “accept”: 1 or 0, indicating whether the offer was accepted in that trial (where accepted == 1, rejected == 0).

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “offer”, “accept”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘ug_delta’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- ug_delta(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)
hbayesdm.models.wcs_sql(data: Union[pandas.core.frame.DataFrame, str, None] = None, niter: int = 4000, nwarmup: int = 1000, nchain: int = 4, ncore: int = 1, nthin: int = 1, inits: Union[str, Sequence[float]] = 'vb', ind_pars: str = 'mean', model_regressor: bool = False, vb: bool = False, inc_postpred: bool = False, adapt_delta: float = 0.95, stepsize: float = 1, max_treedepth: int = 10, **additional_args) → hbayesdm.base.TaskModel

Wisconsin Card Sorting Task - Sequential Learning Model

Hierarchical Bayesian Modeling of the Wisconsin Card Sorting Task using Sequential Learning Model [Bishara2010] with the following parameters: “r” (reward sensitivity), “p” (punishment sensitivity), “d” (decision consistency or inverse temperature).

[Bishara2010]Bishara, A. J., Kruschke, J. K., Stout, J. C., Bechara, A., McCabe, D. P., & Busemeyer, J. R. (2010). Sequential learning models for the Wisconsin card sort task: Assessing processes in substance dependent individuals. Journal of Mathematical Psychology, 54(1), 5-13.

User data should contain the behavioral data-set of all subjects of interest for the current analysis. When loading from a file, the datafile should be a tab-delimited text file, whose rows represent trial-by-trial observations and columns represent variables.

For the Wisconsin Card Sorting Task, there should be 3 columns of data with the labels “subjID”, “choice”, “outcome”. It is not necessary for the columns to be in this particular order; however, it is necessary that they be labeled correctly and contain the information below:

  • “subjID”: A unique identifier for each subject in the data-set.
  • “choice”: Integer value indicating which deck was chosen on that trial: 1, 2, 3, or 4.
  • “outcome”: 1 or 0, indicating the outcome of that trial: correct == 1, wrong == 0.

Note

User data may contain other columns of data (e.g. ReactionTime, trial_number, etc.), but only the data within the column names listed above will be used during the modeling. As long as the necessary columns mentioned above are present and labeled correctly, there is no need to remove other miscellaneous data columns.

Parameters:

  • data – Data to be modeled. It should be given as a Pandas DataFrame object, a filepath for a data file, or "example" for example data. Data columns should be labeled as: “subjID”, “choice”, “outcome”.

  • niter – Number of iterations, including warm-up. Defaults to 4000.

  • nwarmup – Number of iterations used for warm-up only. Defaults to 1000.

    nwarmup is a numerical value that specifies how many MCMC samples should not be stored upon the beginning of each chain. For those familiar with Bayesian methods, this is equivalent to burn-in samples. Due to the nature of the MCMC algorithm, initial values (i.e., where the sampling chains begin) can have a heavy influence on the generated posterior distributions. The nwarmup argument can be set to a higher number in order to curb the effects that initial values have on the resulting posteriors.

  • nchain – Number of Markov chains to run. Defaults to 4.

    nchain is a numerical value that specifies how many chains (i.e., independent sampling sequences) should be used to draw samples from the posterior distribution. Since the posteriors are generated from a sampling process, it is good practice to run multiple chains to ensure that a reasonably representative posterior is attained. When the sampling is complete, it is possible to check the multiple chains for convergence by running the following line of code:

    output.plot(type='trace')

  • ncore – Number of CPUs to be used for running. Defaults to 1.

  • nthin – Every nthin-th sample will be used to generate the posterior distribution. Defaults to 1. A higher number can be used when auto-correlation within the MCMC sampling is high.

    nthin is a numerical value that specifies the “skipping” behavior of the MCMC sampler. That is, only every nthin-th sample is used to generate posterior distributions. By default, nthin is equal to 1, meaning that every sample is used to generate the posterior.

  • inits – String or list specifying how the initial values should be generated. Options are 'fixed' or 'random', or your own initial values.

  • ind_pars – String specifying how to summarize the individual parameters. Current options are: 'mean', 'median', or 'mode'.

  • model_regressor – Whether to export model-based regressors. Currently not available for this model.

  • vb – Whether to use variational inference to approximately draw from a posterior distribution. Defaults to False.

  • inc_postpred – Include trial-level posterior predictive simulations in model output (may greatly increase file size). Defaults to False.

  • adapt_delta – Floating point value representing the target acceptance probability of a new sample in the MCMC chain. Must be between 0 and 1. See note below.

  • stepsize – Integer value specifying the size of each leapfrog step that the MCMC sampler can take on each new iteration. See note below.

  • max_treedepth – Integer value specifying how many leapfrog steps the MCMC sampler can take on each new iteration. See note below.

  • **additional_args – Not used for this model.

  • .. note:: – adapt_delta, stepsize, and max_treedepth are advanced options that give the user more control over Stan’s MCMC sampler. It is recommended that only advanced users change the default values, as alterations can profoundly change the sampler’s behavior. See [Hoffman2014] for more information on the sampler control parameters. One can also refer to ‘Section 34.2. HMC Algorithm Parameters’ of the Stan User’s Guide and Reference Manual.

    [Hoffman2014]

    Hoffman, M. D., & Gelman, A. (2014). The No-U-Turn sampler: adaptively setting path lengths in Hamiltonian Monte Carlo. Journal of Machine Learning Research, 15(1), 1593-1623.
Returns:

model_data – An hbayesdm.TaskModel instance with the following components:

  • model: String value that is the name of the model (‘wcs_sql’).
  • all_ind_pars: Pandas DataFrame containing the summarized parameter values
    (as specified by ind_pars) for each subject.
  • par_vals: OrderedDict holding the posterior samples over different parameters.
  • fit: A PyStan StanFit object that contains the fitted Stan model.
  • raw_data: Pandas DataFrame containing the raw data used to fit the model,
    as specified by the user.

Examples

# Run the model and store results in "output"
output <- wcs_sql(data='example', niter=2000, nwarmup=1000, nchain=4, ncore=4)

# Visually check convergence of the sampling chains (should look like "hairy caterpillars")
output.plot(type='trace')

# Check Rhat values (all Rhat values should be less than or equal to 1.1)
rhat(output, less=1.1)

# Plot posterior distributions of the hyper-parameters (distributions should be unimodal)
output.plot()

# Show the LOOIC and WAIC model fit estimates
print_fit(output)