ml-jku/dark_room

---
license: mit
task_categories:
- reinforcement-learning
---

# Dark-Room
This repository contains the Dark-Room datasets as used in ["Retrieval-augmented Decision Transformer: External Memory for In-context RL"](https://huggingface.co/papers/2410.07071):
- Datasets for 3 grid-sizes: 10x10, 20x20, 40x20.
- Every split contains 100 `.npz` files, corresponding to 100 start-goal position combinations.
- Each file contains 100K transitions.
- Filenames have the format `START_GOAL.npz`. E.g., for `[0,0]_[0,5].npz`, `[0.0]` is the start position and the goal is located at `[0,5]`.

Download the dataset using:
```
huggingface-cli download ml-jku/dark_room --local-dir=./dark_room --repo-type dataset
```
For dataloading we refer to our Github repository: https://github.com/ml-jku/RA-DT

Citation:
```
@article{schmied2024retrieval,
title={Retrieval-Augmented Decision Transformer: External Memory for In-context RL},
author={Schmied, Thomas and Paischer, Fabian and Patil, Vihang and Hofmarcher, Markus and Pascanu, Razvan and Hochreiter, Sepp},
journal={arXiv preprint arXiv:2410.07071},
year={2024}
}
```

---
许可证：MIT
任务类别：
- 强化学习（Reinforcement Learning）
---

# 暗室（Dark-Room）数据集

本仓库包含用于论文《检索增强决策Transformer（Retrieval-augmented Decision Transformer）：面向上下文强化学习的外部记忆》的暗室数据集，相关论文链接为：https://huggingface.co/papers/2410.07071，具体信息如下：
- 该数据集涵盖3种网格尺寸：10×10、20×20、40×20。
- 每个数据拆分包含100个`.npz`格式文件，对应100组起始-目标位置组合。
- 每个文件包含10万条环境交互转移数据。
- 文件名格式为`START_GOAL.npz`。例如`[0,0]_[0,5].npz`中，`[0,0]`代表起始位置，目标位置为`[0,5]`。

可通过以下命令下载该数据集：

huggingface-cli download ml-jku/dark_room --local-dir=./dark_room --repo-type dataset

关于数据集加载方式，请参阅我们的GitHub仓库：https://github.com/ml-jku/RA-DT

引用格式：

@article{schmied2024retrieval,
title={Retrieval-Augmented Decision Transformer: External Memory for In-context RL},
author={Schmied, Thomas and Paischer, Fabian and Patil, Vihang and Hofmarcher, Markus and Pascanu, Razvan and Hochreiter, Sepp},
journal={arXiv preprint arXiv:2410.07071},
year={2024}
}