scikit学习矢量化器词汇，具有映射到同一索引的多个术语

Question

skikit-learn's TfidfVectorizer correctly maps vocabulary terms with the same dictionary value to the same index, however, it creates as many columns in the output as there are entries in the vocabulary dictionary. Is there a better way to get around this than to strip off the extra columns after the transformation? That is, in the example below, I don't want the third column because it will always be zero.

from sklearn.feature_extraction.text import TfidfVectorizer
vectorizer=TfidfVectorizer(vocabulary={'surgery':0, 'sx':0, 'radiology':1})
text=['i had surgery','patient sx went well','radiology department']
vectorizer.fit(text)
vectorizer.transform(text).todense()

>>> matrix([[ 1.,  0.,  0.],
            [ 1.,  0.,  0.],
            [ 0.,  1.,  0.]])

Answer 1

A sklearn.feature_selection.VarianceThreshold (scikit-learn >= 0.15) will remove all-zero features (and constant features more generally).

>>> X = np.array([[1, 0, 0], [1, 0, 0], [0, 1, 0]])
>>> VarianceThreshold().fit_transform(X)
array([[1, 0],
       [1, 0],
       [0, 1]])