To generate *data.csv*, the Matlab file *RL_data.mat*, available at https://github.com/sjgershm/RL-models/tree/master/jmp, has to be exported as a .csv file. These data were collected and originally published by Gershman (Gershman, S. J., 2015. Do learning rates adapt to the distribution of rewards? *Psychonomic Bulletin & Review, 22*, 1320–1327. doi:10.3758/s13423-014-0790-3). *simulated_dataset.csv* contains simulated experiments (4 blocks of 25 trials each) for 1000 virtual participants. Columns code, in that order, *participant id, outcome of the left option, outcome of the right option, block ID, probability of reward of left option, probability of reward of the right option*. Files *complete_mX.R* (where *X* represents the model number) contain analysis scripts for the first part of the paper for each of the models. Included are the following steps: - Fit the corresponding models to the participant data (parameters saved in *param_mX.csv*) - Fit Gaussian mixture models with 1 to 10 mixture components to the resulting parameters (500 times). The best-fitting mixture (according to BIC) is then saved to *RLprior_mX.RData*) - Simulate the responses of 1000 virtual participants for the uniform and the empirical-prior populations, respectively. These responses are then re-fit using MLE and MAP (with empirical priors). Responses, true parameters, re-estimated parameters, and goodnesses of fit are then stored in *sim_mX.RData*. - Estimation of distribution recovery. *recoverability.ipynb* is the Jupyter Notebook (http://jupyter.org/) that contains the analyses for the second part of the paper for model **M1**. Included are the following steps: - Baseline assessment (results are stored in *baseline.csv*) - Variant 1, which increases the number of trials (results are stored in *trials.csv* and *trials_X.csv*; robustness check) - Variant 2, which increases the number of options to choose from (results are stored in *options.csv*) - Variant 3, which provides full feedback (results are stored in *full_feedback.csv*)