CODEWITHSUNDEEP
scikit-learnpytorch
natasha_667
natasha_667

Reputation: 57

sklearn pipeline with multiple inputs/outputs

How can I build a sklearn pipeline to do the following?

What I have:

A, B = getAB(X_train)
X_train = transform(X_train)
model(A, B, X_train)

What I want:

pipe = Pipeline([
(‘ab’, getAB),
(‘tranf’, transform),
(‘net’, net)
]
pipe.fit(X_train, y_train)

Please help!

Upvotes: 2

Views: 2458

Answers (1)

Coderji
Coderji

Reputation: 7765

Yes it is doable by writing a custom transformer that has a fit/transform function. This can be your class:

from sklearn.base import BaseEstimator, TransformerMixin

def getABTransformer(BaseEstimator, TransformerMixin):
    def __init__(self): # no *args or **kargs
        pass

    def fit(self, X, y=None):
        return self # nothing else to do

    def transform(self, X, y=None):
        return getAB(X)

Then you can create your ColumnTransformer as following:

from sklearn.compose import ColumnTransformer

clm_pipe = ColumnTransformer([
(‘ab’, getABTransformer, np.arange(0, len(X_train)),  # list of columns indices
(‘tranf’, transform, np.arange(0, len(X_train))),  # list of columns indices
]

and a final pipeline with the model:

pipe = Pipeline([
(‘clm_pipe’, clm_pipe),
(‘net’, net)
]

You can read more about ColumnTransformer

Upvotes: 1

Related Questions