Sklearn compatible multi feature transformer

[lcrmorin]

I have found myself in a position to need a practical way to apply binning + WoE encoding on all features of a dataset. As I was not able to find the solution within the package I have implemented a custom sklearn compatible version myself. Typically it works as follows:

OBT = OptBinningTransformer()
X_WoE = OBT.fit_transform(X, y)

I have made some effort to comment / sanitise / make it customisable / handle types / define some tests. Would it be a good idea to integrate this ? Where ?

[bmreiniger]

Doesn't BinningProcess do this, since the default transform method for a binary classification is WoE?

[lcrmorin]

It seems to be working after inputing column names... I somehow confused it with something else.
Still a bit weird to have to specify column names.

[bmreiniger]

Oh, yes, I would advocate for maybe defaulting the list of columns to None which then applies to all columns (using names if pandas and x{i} for i in range(m) else).

(I also have a thin wrapper in my company's packaging for this sort of thing. I think I recall the priority of the transform metric specifications (in constructor and in transform) seemed backwards from what I'd expect too?)

[jnsofini]

I would advocate for setting the name to null or dummy values, such as C1, C2, etc., so that one can pass a numpy array in it.