Support floating-point targets in Khiops classifiers

[popescu-v]

Description

Reported by M.Arcadias

Currently, KhiopsClassifier's fit method accepts targets of type string, integer or categorical.

However the Scikit-learn/XGBoost Python API also accept float target types.

The pyKhiops KhiopsClassifier's API should also support this.

Questions / Ideas

• Instead of converting to a currently-supported type (string, categorial, or integer, if targets are convertible to integer):

clf = KhiopsClassifier()
clf.fit(X_train, y_train.astype("str"))

one should be able to just write:

clf = KhiopsClassifier()
clf.fit(X_train, y_train)

where y_train.dtype is dtype('float64') (if y_train is a pandas series).

• The predict method should also return float64 ?

[folmos-at-orange]

At the moment the train works but the predict fails.