tab-transformer-pytorch

Why we included this project

Most tabular modeling still runs on gradient boosting, and this repo is for teams that want to test what attention can do on the same kind of data without changing their whole stack. It implements the original TabTransformer from the 2020 paper, which the author says came close to matching GBDT performance, and it also bundles Yandex's FT-Transformer, a follow-up that embeds continuous values with a simpler scheme. Having both in one install makes it easy to run them head to head against your existing pipeline. The code is a plain PyTorch module, small and pip-installable, so you can drop it into a classification or regression task without adopting a larger framework. If you are unsure whether deep learning on structured data beats what you already have, this is a cheap way to find out on your own data.