This repo is based on the code from the benchmark and techniques introduced in the paper Unsupervised State Representation Learning in Atari. Please visit https://github.com/mila-iqia/atari-representation-learning for detailed instructions on the benchmark.
To run the script:
python run_probe.pyAn example of setting the environment, pretrain with masked images and masking ratio 0.8, seed 2:
run_probe.py --env-name VideoPinballNoFrameskip-v4 --pretrain-masks --mask-ratio 0.8 --seed 2