Transformers fail without categorical features

[dholzmueller]

The models SAINT, FTTransformer and TabFastFormer contain the code

        if self.n_cont and not self.n_cat and not self.embed_continuous:
            raise ValueError(
                "If only continuous features are used 'embed_continuous' must be set to 'True'"
            )

The variable self.embed_continuous is not set anywhere, probably it has been used in an older version of the code and was then replaced by True later. For data sets without categorical features, this code raises an error. Removing this code should resolve the issue.

[jrzaurin]

hi @dholzmueller

thanks, I will review this asap

[jrzaurin]

The issue is now fixed and merged into master.

I will publish to pypi asap, in the meantime please install via github

[jrzaurin]

It is now available on pypi (v.1.0.10)

Thanks for the issue. Closing it now