fit_params in conjunction with FeatureUnion #7136
Description
Description
Using fit_params in conjunction with FeatureUnion may not work as expected. It would be helpful if fit_params names would be resolved to fit the estimators in transformer_list.
Steps/Code to Reproduce
from sklearn.pipeline import Pipeline, FeatureUnion
from sklearn.preprocessing import FunctionTransformer
from sklearn.datasets import make_classification
X, y = make_classification()
class MyTransformer(FunctionTransformer):
def fit(self, X, y=None, **fit_params):
print("Fit params are: ", fit_params)
return super().fit(X, y)
pipe = Pipeline([
('step0', FunctionTransformer()),
('step1', FeatureUnion([
('feature0', MyTransformer()),
('feature1', MyTransformer()),
])),
('step2', FunctionTransformer()),
])
pipe.fit(X, y, step1__feature1__someparam=123)
Expected Results
# prints
Fit params are: {'someparam': 123}
Actual Results
# prints
Fit params are: {'feature1__someparam': 123} # by feature0
Fit params are: {'feature1__someparam': 123} # by feature1
Comment
Maybe the actual outcome is what it is supposed to be, but my expectation would be that FeatureUnion resolves the estimator name and only passes the fit_params to the corresponding estimator. At least it would be very useful if it did.
Versions
Python 3.4.5
NumPy 1.10.4
SciPy 0.17.0
Scikit-Learn 0.17.1