Awesome
<p align="center"> <img width="80%" src="https://raw.githubusercontent.com/MichaelTMatthews/Craftax_Baselines/main/images/logo.png" /> </p>Craftax Baselines
This repository contains the code for running the baselines from the Craftax paper. For packaging reasons, this is separate to the main repository.
Installation
git clone https://github.com/MichaelTMatthews/Craftax_Baselines.git
cd Craftax_Baselines
pip install -r requirements.txt -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
pre-commit install
Run Experiments
PPO
python ppo.py
PPO-RNN
python ppo_rnn.py
ICM
python ppo.py --train_icm
E3B
python ppo.py --train_icm --use_e3b --icm_reward_coeff 0
RND
python ppo_rnd.py
Visualisation
You can save trained policies with the --save_policy flag. These can then be viewed with the view_ppo_agent script (pass in the path up to the files directory).