BarlowRL: Barlow Twins for Data-Efficient Reinforcement Learning

Status: Archive (code is provided as-is, no updates expected)

This is an implementation of BarlowRL: Barlow Twins for Data-Efficient Reinforcement Learning. Unlike CURL, BarlowRL replaces contrastive loss with the Barlow Twins objective. BarlowRLhas 0.378 Mean Human-Normalized Score and 0.296 Median Human-Normalized Score, 0.278 IQM and 0.651 Optimality Gap in Atari 100k Benchmark. This repository is mostly built from the CURL repository.

Dependencies

To install all dependencies, run bash install.sh.

Atari-py

Download roms into atari_py package and prepare roms via pyton script in it.

wget http://www.atarimania.com/roms/Roms.rar
unrar x Roms.rar
python -m atari_py.import_roms .

Training

After installing all dependencies run following command for training.

bash run_train.sh