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.
To install all dependencies, run bash install.sh.
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 .After installing all dependencies run following command for training.
bash run_train.sh