xionghuichen / GALILEO

The official code of "Adversarial Counterfactual Environment Model Learning" (NeurIPS'23 spotlight)

GALILEO (NeurIPS'23 Spotlight)

The official code of "Adversarial Counterfactual Environment Model Learning".

We provide a faithful offline dynamics model learning technique based on the adversarial model learning paradigm.

The optimization pipeline of the proposed algorithm is as follows:

Also, you can see this Twitter thread for a brief discussion of this algorithm: https://x.com/xiong_hui_chen/status/1737597369514860803?s=20

quickstart

install

pip install -e .
pip install -r requirements.txt

install RLAssistant for experiment management

git clone https://github.com/polixir/RLAssistant.git
cd RLAssistant
pip install -e .

run

cd run_scripts
python main.py --data_type d4rl --env_name hopper --data_train_type medium

view your results

1. the tensorboard logs are in ./RLA_LOG/log folder;
2. you can manager your experiment result via RLAssistant (see: https://github.com/polixir/RLAssistant)