implicit_q_learning's Introduction

Offline Reinforcement Learning with Implicit Q-Learning

This repository contains the official implementation of Offline Reinforcement Learning with Implicit Q-Learning by Ilya Kostrikov, Ashvin Nair, and Sergey Levine.

If you use this code for your research, please consider citing the paper:

@article{kostrikov2021iql,
    title={Offline Reinforcement Learning with Implicit Q-Learning},
    author={Ilya Kostrikov and Ashvin Nair and Sergey Levine},
    year={2021},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

For a PyTorch reimplementation see https://github.com/rail-berkeley/rlkit/tree/master/examples/iql

How to run the code

Install dependencies

pip install --upgrade pip

pip install -r requirements.txt

# Installs the wheel compatible with Cuda 11 and cudnn 8.
pip install --upgrade "jax[cuda]>=0.2.27" -f https://storage.googleapis.com/jax-releases/jax_releases.html

Also, see other configurations for CUDA here.

Run training

Locomotion

python train_offline.py --env_name=halfcheetah-medium-expert-v2 --config=configs/mujoco_config.py

AntMaze

XLA_PYTHON_CLIENT_PREALLOCATE=false  CUDA_VISIBLE_DEVICES=7 python train_offline.py --env_name=antmaze-medium-play-v0 --config=configs/antmaze_config.py --eval_episodes=100 --eval_interval=100000 --seed 43

Kitchen and Adroit

python train_offline.py --env_name=pen-human-v0 --config=configs/kitchen_config.py

Finetuning on AntMaze tasks

python train_finetune.py --env_name=antmaze-large-play-v0 --config=configs/antmaze_finetune_config.py --eval_episodes=100 --eval_interval=100000 --replay_buffer_size 2000000