building such system possible ?

[lschneidpro]

I checked the documentation, but i could not figure out if i could build such system ?

thank you for your help

[flennerhag]

Yes that's entirely possible. A simple version could look like this:

from mlens.ensemble import SuperLearner

# First, create a transform class for the target variable

class MyTrans:

    def __init__(self, *args, **kwargs):
        # up to you

    def fit(self, X, y):
        # up to you
        return self

    def transform(self, X, y):
        y_transformed = # up to you
        return X, y_transformed

    def fit_transform(self, X, y):
        return self.fit(X, y).transform(X, y)

# Next, create the ensemble

ens = SuperLearner(proba=True)

ens.add(estimator=classifier_1(),
        propagate_features=range(num_features))

ens.add_meta(estimator=classifier_2(),
             preprocessing=MyTrans())

Hope that helps, otherwise take a look at the docs of the ensembles, they contain a wealth of information about how to create layers!