Pipeline object does not have a get_feature_names method - intentional? #6421
Description
The Pipeline object does not have a get_feature_names method. Is this intentional?
A get_feature_names method is useful when dealing with parallel feature extraction like in this blog post or in the short example below:
from sklearn.feature_extraction import DictVectorizer
from sklearn.pipeline import FeatureUnion
from sklearn.pipeline import make_pipeline
from sklearn.base import TransformerMixin
class WordFeatureExtractor(TransformerMixin):
def transform(self, X, **transform_params):
return [dict(self._extract_features(data.items())) for data in X]
def fit(self, X, y=None, **fit_params):
return self
def _extract_features(self, key_value_pair):
for key, value in key_value_pair:
if isinstance(value, basestring):
for word in value.split():
yield '%s.word=%s' % (key, word), True
feature_extractor = FeatureUnion([
('dict', DictVectorizer()),
('word', make_pipeline(
WordFeatureExtractor(),
DictVectorizer())
)
])
feature_extractor.fit_transform([{'number': 123, 'text': 'foo bar'}, {'number': 321}])
feature_extractor.get_feature_names()
The above code example results in AttributeError: Transformer word does not provide get_feature_names.
I would be happy to create a pullrequest is someone who has an overview of the project could verify that this functionality is actually wanted. Looking at Pipeline source, it looks like we only need 3 lines of code:
@if_delegate_has_method(delegate='_final_estimator')
def get_feature_names(self):
return self.steps[-1][-1].get_feature_names()