This repository will include the original implementation and experiment codes of ExcelFormer. TabFormer is a pioneering neural network can surpass extensively-tuned XGboost, Catboost, and most tuned previous deep learning approaches on most of tabular data prediction tasks, in the supervised learning manner. It can be a go-to choice on tabualr dataset prediction competitions (e.g., Kaggle).

Even without hyper-parameter tuning, TabFormer performs comparable to tuned models. After hyper-parameter tuning, TabFormer typically outperforms them.

The implementation of TabFormer in the original paper is bin/excel_former.py.

You can test your models by adding them to bin directory and bin/__init__.py. Keep the same API we used in other models, and write your own evaluation script (run_default_config_excel.py as a reference).

The datasets (96 small tabular datasets + 21 large tabular datasets) are available at: https://huggingface.co/datasets/jyansir/excelformer.

We will organize our previous works on tabular prediction into Tabular AI Research group for industrial use (e.g. further architecture optimization or acceleration / compilation). If you want to include our model as a baseline in your paper, please use the version in this repository rather than the industrial one in the group repository.