Cooperation Task Model Fitting

Description

This repository contains MATLAB scripts for fitting computational models to cooperation task behavioral data. There are three folders: The AInf_scripts folder contains the most up-to-date Active Inference scripts. The AInf_scripts_used_to_compare_VB_with_MCMC folder contains slightly out-of-date active inference scripts to compare fitting the same model with Variational Bayes (here) vs. Markov Chain Monte Carlo Sampling (https://github.com/cgoldman123/Active_Inference_Model_MCMC_Cooperation_Task) The RL_heuristic_scripts contains Reinforcement Learning scripts.

Each of these pipelines implement model fitting using variational Bayes, simulate fitted parameters, and generate model-free analyses. The code can operate in two experimental modes: local (for in-house experiments) and prolific (for online data collection). The results are saved in .csv and .mat formats for further analysis.

---

Scripts used in each pipeline:

main_script.m

• Purpose: The primary script for fitting models to cooperation task data.
• Key Features:
  • Fits models using variational Bayes for each subject listed in fit_list.
  • Supports parameter simulation using pre-defined values (SIM_PASSED_PARAMETERS).
  • Saves model-free analysis results if DO_MODEL_FREE is set to true.
  • Allows for saving prediction errors and fitting results in the specified result_dir.

TAB_fit_simple_prolific.m

• Purpose: Fits cooperation task data for Prolific subjects.
• Workflow:
  • Reads behavioral data files.
  • Parses observations and actions.
  • Sets up the MDP (Markov Decision Process) model structure.
  • Inverts the model using variational Bayes via TAB_inversion_simple.

TAB_inversion_simple.m

• Purpose: Inverts the MDP model using variational Bayes.
• Inputs:
  • DCM: Contains the MDP structure, parameters to optimize, outcomes (DCM.U), and responses (DCM.Y).
• Outputs:
  • DCM.M: The generative model.
  • DCM.Ep: Posterior means of parameters.
  • DCM.Cp: Posterior covariances.
  • DCM.F: Variational free energy (used as a log evidence bound).

Model scripts: Simple_TAB_model_v3.m, RW_model.m, Simple_TAB_model_v2.m, etc

• Purpose: Implements the computational models for simulating and fitting cooperation task data.
• Functionality:
  • Computes action probabilities by modeling subjects' learning and decision-making.
  • Supports both simulation and fitting modes.
  • Incorporates learning rates and forgetting rates that can be split based on outcome type (wins, losses, neutral).
  • Outputs key model variables like chosen actions, outcomes, action probabilities, and model parameters.