---
pretty_name: Amazon-WILDS
license: other
viewer: false
tags:
- wilds
- distribution-shift
- benchmark
---

# Amazon-WILDS

This repository contains the Amazon dataset used in the [WILDS benchmark](https://wilds.stanford.edu). It mirrors the released data from the [WILDS CodaLab worksheet](https://worksheets.codalab.org/worksheets/0xb44731cc8e8a4265a20146c3887b6b90).

For dataset descriptions and usage instructions, see the [WILDS Python package](https://github.com/p-lambda/wilds) and the papers:

- [WILDS: A Benchmark of in-the-Wild Distribution Shifts](https://arxiv.org/abs/2012.07421)
- [Extending the WILDS Benchmark for Unsupervised Adaptation](https://arxiv.org/abs/2112.05090)

Download the archive files from this repository and extract them into the corresponding dataset folders expected by WILDS. Release information and checksums are in `manifest.json` and `SHA256SUMS`.

## License

None. However, the original authors request that the data be used for research purposes only.