Using recurrent neural network to estimate irreducible stochasticity in human choice-behavior

git repo accompanying the project https://psyarxiv.com/ve4rg

Background

Theory-driven computational modeling allows estimation of latent cognitive variables.
Nonetheless, studies have found that computational models can systematically fail to describe some individuals.
Presumably two factors contribute to this shortcoming:

1. Higher internal noise (stochasticity in behavior).
2. Model miss-specification.

However, when measuring behavior of individuals on cognitive tasks, these two factors are entangled and therefore hard to dissociate.
Here we examine the use of RNNs to disentangled this two factors.

study_1_simulation

• simulating agents, fitting the theoretical models, RNN and Logistic regression model run the following notebook:

study_1_simulation/code/sim_fit_predict.ipynb

• plotting Figures of study 1

study_1_simulation/code/plots.ipynb

study_2_emprical

• Fitting the theoretical hybrid model and RNN

study_2_emprical/code/fit_predict.ipynb

• To run the bayesian analysis of study 2

study_2_emprical/code/bayesian.ipynb

• plotting Figures of study 2

study_2_emprical/code/plots.ipynb

Dependencie

• numpy
• pandas
• matplolib
• seaborn
• sklearn
• scipy
• tqdm
• torch
• pymc
• bambi