Skip to content

Official PyTorch implementation of "Discovering Hierarchical Achievements in Reinforcement Learning via Contrastive Learning" (NeurIPS 2023)

License

Notifications You must be signed in to change notification settings

snu-mllab/Achievement-Distillation

Repository files navigation

Achievement Distillation

This is the code for the paper Discovering Hierarchical Achievements in Reinforcement Learning via Contrastive Learning accepted to NeurIPS 2023.

Installation

conda create --name ad-crafter python=3.10
conda activate ad-crafter

pip install --upgrade "setuptools==65.7.0" "wheel==0.38.4"
pip install -r requirements.txt
pip install -e .

Usage

To execute train.py in a Bash environment, you can use the following commands. By default, the system will assign the timestamp debug.

PPO

python train.py --exp_name ppo --log_stats

PPO + Achievement Distillation (ours)

python train.py --exp_name ppo_ad --log_stats

If you are working in an environment that uses the Slurm Workload Manager, you can submit your job using slurm.py. In this setup, the system automatically assigns a timestamp that corresponds to the actual start time of your job.

Citation

If you find this code useful, please cite this work.

@inproceedings{moon2023ad,
    title={Discovering Hierarchical Achievements in Reinforcement Learning via Contrastive Learning},
    author={Seungyong Moon and Junyoung Yeom and Bumsoo Park and Hyun Oh Song},
    booktitle={Neural Information Processing Systems},
    year={2023}
}

Credit

About

Official PyTorch implementation of "Discovering Hierarchical Achievements in Reinforcement Learning via Contrastive Learning" (NeurIPS 2023)

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages