Reinforcement Learning via. Sequence Modeling using Decision Transformer

Introduction

This repository contains experiments to reproduce some of the results of the paper Decision Transformer: Reinforcement Learning via Sequence Modeling (Chen et al.) and Online Decision Transformer (Zheng et al.) and study the effect of context length on the variances of the expected returns of the Decision Transformer model in OpenAI Gym environments and Atari games. For more details, check out this report / slides.

Running Experiments

Run the notebook file HuggingFaceDecisionTransformer.ipynb and follow the instructions in the file to experiment with the pre-trained models using HuggingFace models.

Open in Colab

Run the notebook file decision_transformer_gym.ipynb and follow the instructions in the file. Note that Google Drive has been linked to the notebook for data and output storage. This notebook file is used to run Decision Transformer training on OpenAI Gym environments.

Open in Colab

Run the notebook file decision_transformer_atari.ipynb and follow the instructions in the file. Note that Google Drive has been linked to the notebook for data and output storage. This notebook file is used to run Decision Transformer training on on Atari games.

Open in Colab

Run the notebook file online_dt.ipynb and follow the instructions in the file. Note that Google Drive has been linked to the notebook for data and output storage. This notebook file is used to run Online Decision Transformer training on both OpenAI Gym environments and Atari games.

Open in Colab

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

License

Apache License 2.0