Difference between standalone implementation and full package

[albertcthomas]

If I want to use Tabmini in one of my projects I think I can only use the standalone implementation. Is the full package mostly needed if I want to reproduce the results of the papers and having hyperopt? Or are there other benefits in installing the full package?

[kegl]

I'm a colleague of Albert. What we'd need is a classical fit/predict API, do you guys think it would be possible?

[Yura52]

@albertcthomas the "full package" (i.e. bin, lib etc.) is needed only to reproduce our setup used the paper, with hyperparameter tuning etc. So if you simply want to import TabM without using our "infrastructure", the standalone implementation is recommended: it is the exact full-featured implementation of TabM.

[Yura52]

@kegl I don't expect the Scikit-learn API to be implemented within this repository. However, I totally understand the motivation behind the request, and we are currently thinking of how to make the training easier for users.