This is the code for the publication:

V. Liakoni *, M. Lehmann *, A. Modirshanechi, J. Brea, A. Lutti, W. Gerstner ** , K. Preuschoff ** Brain signals of a Surprise-Actor-Critic model: Evidence for multiple learning modules in human decision making, NeuroImage 246, 1053-8119 (2022)

* V.L. and M.L. made equal contribution to this article.
** W.G. and K.P. made equal contribution to this article.

Contact: vasiliki.liakoni@gmail.com, marco.lehmann@gmail.com

Installation

Dependencies:

• Mac or Linux
• Julia (1.2)
• MATLAB R2015a

For the julia code

Navigate into the src folder.
Open a julia terminal, press "]" to enter the package management mode and type

(v1.2) pkg> activate ..
(SinergiafMRI_datafit) pkg> instantiate

All (julia) packages and dependencies will be installed automatically within this environment.

For the MATLAB code

Navigate into the MATLAB folder and add all subfolders to your MATLAB path.

Usage

For the julia code

julia> using SinergiafMRI_datafit

To run the model fitting and model comparison procedures of the paper, type

julia> SinergiafMRI_datafit.runner_crossval_multipletimes_fit()

Note that these procedures would take long, so the settings are different from the ones used in the paper.
Uncomment the marked lines in the fmri_run_multipletimes.jl file to perform the analysis using the paper's settings.

A good place to start playing with the data is the simpleFittingTest.jl file.

For more information, please see the src/readme.txt.

For the MATLAB code

To run the experiment run the script start_chuv_7S_G1_LinkFlip.m (for graph 1 of paper - left column of Fig. A.1) and the script start_chuv_7S_G3_LinkFlip.m (for graph 2 of paper - right column of Fig. A.1).

For more information, please see the MATLAB/readme.txt.

Code

• MATLAB/experiment: the paper's experiment, participants' data recording and some data preprocessing.
• MATLAB/recovery: the paper's experiment (same as above, but without visualizations) with simulated Surprise Actor-critic agents.
• src/mcmc_rl_fit/src: RL algorithms, model fitting and model comparison (via crossvalidation).
• src/mcmc_rl_fit/fmri: high level runners, data analysis, plotting. Please refer to src/readme.txt and to MATLAB/readme.txt for more information.

Data and Figures

• /MATLAB/experiment/ParticipantsData: the participants's (raw) data.
• /src/mcmc_rl_fit/projects/fmri/data: all participants' data concatenated. The file SARSPEICZVG_all_fMRI.csv is used for all analyses within the julia code.
• /src/mcmc_rl_fit/projects/fmri/data_sim: all simulated agents' data concatenated.
• data: final results used for the paper's figures.
• figs: latex source code to reproduce the paper's figures.