Can Wikipedia Help Offline RL?
Machel Reid, Yutaro Yamada and Shixiang Shane Gu.
Our paper is up on arXiv.
Overview
Official codebase for Can Wikipedia Help Offline Reinforcement Learning?.
Contains scripts to reproduce experiments. (This codebase is based on that of https://github.com/kzl/decision-transformer)
Instructions
We provide code our code directory containing code for our experiments.
Installation
Experiments require MuJoCo.
Follow the instructions in the mujoco-py repo to install.
Then, dependencies can be installed with the following command:
conda env create -f conda_env.yml
Downloading datasets
Datasets are stored in the data directory. LM co-training and vision experiments can be found in lm_cotraining and vision directories respectively.
Install the D4RL repo, following the instructions