Add TabPFN support as an candidate model.

[Ben-FCC]

Hi there,

First off, I want to say that neighbors is a fantastic tool for recovering missing data! The framework is intuitive and incredibly useful for various applications.

I’d like to request the addition of TabPFN (repo, paper) to neighbors. TabPFN is a transformer-based model that performs tabular classification with a single forward pass, without requiring per-dataset training. It has been shown to be highly effective, especially for small datasets.

Would it be possible to integrate TabPFN as an additional model within neighbors? I believe this would greatly enhance the functionality for users working with tabular data.

Looking forward to your thoughts!

Thanks!