deepAI Implementation of a Deep Active Inference agent, accompanying the article “Deep Active Inference” (Biological Cybernetics, 2018; preprint). The code depends on Python 3 and Theano for tensor operations.
To install the required components on Ubuntu, you can follow: http://deeplearning.net/software/theano/install_ubuntu.html
To sample from the trained model saved in deepAI_demo_best.pkl, just run python sample_deepAI_paper.py
To train a new agent, run python deepAI_paper.py
To plot the progress, run python plot_log_deepAI_paper.py
The scripts to directly reproduce all figures in the paper are in the ./figures/ subfolder.