This page lists answers to some frequently asked questions about TableShift.

How can I access {dataset}? Can you help me access it?

Instructions for accessing datasets are on the datasets page.

If you have questions about an individual dataset, first check the instructions linked for that dataset on accessing the datasets. Then, the best way to get help would be by filing a github issue here (please make sure to use the "Dataset Access" issue tag!).

Can you provide preprocessed versions of the datasets?

No. In order to comply with the data access restrictions for all datasets while also providing a unified interface, we do not make preprocessed versions of the datasets available. However, TableShift is fully capable of building the preprocessed datasets from source! See the instructions in the quisktart for more info.

Can I contribute my dataset to the benchmark?

We are open to contributions of high-quality datasets that meet the criteria listed in our paper. If you are interested in contributing your dataset to the benchmark, please contact the team directly.

Where can I find more information about a data source?

  • Check the entry for that dataset on the datasets page.
  • Follow the links on the datasets page to the source documentation for that dataset.
  • Read the TableShift paper.

How can I cite the TableShift benchmark?

Please cite our paper:

  title={Benchmarking Distribution Shift in Tabular Data with TableShift},
  author={Gardner, Josh and Popovic, Zoran and Schmidt, Ludwig},
  journal={Advances in Neural Information Processing Systems},