reward-weighted-regression

This repository contains the source code for the experiments presented in Reward-Weighted Regression Converges to a Global Optimum by Miroslav Štrupl, Francesco Faccio, Dylan R. Ashley, Rupesh Kumar Srivastava, and Jürgen Schmidhuber.

To produce the plots shown in the paper, first ensure you have python 3.8.0 installed and execute the following in a bash shell:

pip install -r requirements.txt
./build.sh
./tasks_0.sh
./plot.py results.pdf

After execution has completed, a results.pdf file should have been generated.