gpx.Dataset how to handle mixed types? Secifically categorical encodings?

[cottrell]

So far I don't see anything in the code that suggests anyone is handling high-arity categoricals (I see one hot encodings).

Is there a pattern for this?

I typically like X be a dict of features to generalize things. Am wondering if there is a clear way to work around this pattern here besides casting ints to the array dtype and downcasting at embedding time.