The dataset link is expired.

[jayden222]

Could you please renew it ?

[puhsu]

Sory for not answering that long.

I'm unable to get the exact set of datasets, used in this paper. But you can use the most recent version of our benchmark at here. The only difference is that there is no Gesture Phase (GE) dataset and two additional ones in Diamonds (DI) and Black Friday (BL). The results in the TabR repo are also closer to our current setup (e.g. better tuned GBDTs).

Note

Also check new minimal (pip packaged) implementations of the baselines MLP, ResNet, FT-Transformer at (https://github.com/yandex-research/rtdl-revisiting-models) and MLP with embeddings at https://github.com/yandex-research/rtdl-num-embeddings)