Dirichlet Mixture Model for Probabilistic Reversal Learning

Description

An application of the model from "A generative framework for the study of delusions" developed by Tore Erdmann and Christoph Mathys. The model is used for simulating probabilistic reversal learning tasks performed by delusional and non-delusional agents. The implemented probabilistic reversal learning task is based on "Defining the neural mechanisms of probabilistic reversal learning using event-related functional magnetic resonance imaging".

Requirements

The code was developed and tested using Julia Version 1.7.0-rc1 and requires the following packages:

• DataFrames
• Distributions
• FileIO
• JLD2
• Plots
• Random
• StatsBase
• StatsFuns
• StatsPlots

For convenience, we include a Julia environment with the above packages in the repository.
Additionally, we use Make to generate the directories for the output of the code, however, it is not required as the directories can be manualy created.

Setting up the project

To setup the required directories run the following command in the commandline

make init

The directories can also be setup manually in the root directory of the repository. The following directories should be created:

• io
• io/plots
• io/results

Afterwards, start the Julia REPL and run the following command to go into pkg mode

]

followed by

activate .

to activate the Julia envrionment.

Usage

To generate simulation results run the following command in Julia REPL:

include("main.jl")

To run simulations with different parameters, change the parameters in the main.jl file. The following parameters can be changed:

• method: Which experiment to run. Can be either "3p_10t" for 3 phases with 10 trials or "cools" for 3 phases with at most 50 trials incremented in steps of 10
• mu_tau_c: Expected precision of the control agent
• mu_tau_p: Expected precision of the patient agent
• n_runs: How many runs to simulate per agent
• n_history: Models the memory of an agent. The agent will remember the last n_history trials.
• belief_strength: How many samples to use for reinforcing the current hypothesis.

Support

For questions concerning the project, please contact us at:
Wiona Glänzer, wglaenzer@ethz.ch
Katarzyna Kransapolska, krasnopk@ethz.ch
Timofey Shpakov, tshpakov@ethz.ch

Authors and acknowledgment

We want to thank Tore Erdmann, Christoph Mathys and Alexander Hess for their support and guidance during this projects.