nam-trinh / rnn_beh

Recurrent neural networks for the analysis of decision-making data

ActionFlow: Recurrent neural networks for the analysis of decision-making data

This repository contains the code for the model introduced in the following paper:

Dezfouli A, Griffiths K, Ramos F, Dayan P, Balleine BW (2019). Models that learn how humans learn: The case of decision-making and its disorders. PLoS Comput Biol 15(6): e1006903

https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1006903

The model is called ActionFlow and describes how to fit a recurrent neural network to behavioural data, and how to interpret it to gain insights into the underlying decision-making processes in the brain.

See also the following paper which uses a similar approach to fit a recurrent neural network to both behavioural and neural data:

Dezfouli, A., Morris, R., Ramos, F. T., Dayan, P., & Balleine, B. (2018). Integrated accounts of behavioral and neuroimaging data using flexible recurrent neural network models. In Advances in Neural Information Processing Systems

https://www.biorxiv.org/content/biorxiv/early/2018/12/20/328849.full.pdf

(Note this repository does not contain the code for this paper).

Installation

The required packages are in src/requirements.txt. Using Virtual Environments, they can be installed as follows:

pip install virtualenv
virtualenv venv
source venv/bin/activate
cd src
pip install -r requirements.txt

Content

The repository contains three main packages:

• actionflow: This is the main package, which can be used for training, testing and simulating the model.

• BD: This contains the analysis of the dataset reported in the paper. It uses actionflow for training, testing and simulating the model. There are two packages in BD: one is fit and the other one is sim. fit contains the files used for fitting the models to data, and sim contains the files used for simulating and testing the model. A brief comment is presented at the beginning of each file in this package to summarise its content.

  For example, for running cross-validation experiments, the following command can be used (current directory should be src):

  python -m BD.fit.rnn_cv n_proc

  n_proc number of parallel processes

  The package also contains an R folder which contains the codes used for data analysis and generating graphs in the paper.

• examples: This includes an IPython notebook to demonstrate the basic functionalities of actionflow.