HashingTfIdfVectorizer

class HashingTfIdfVectorizer:
    """Difference with HashingVectorizer: non_negative=True, norm=None, dtype=np.float32"""

    def __init__(self, ngram_range=(1, 1), analyzer=u'word', n_features=1 << 21, min_df=1, sublinear_tf=False):
        self.min_df = min_df
        self.hasher = HashingVectorizer(non_negative=True, norm=None, dtype=np.float32,
                                        ngram_range=ngram_range, analyzer=analyzer, n_features=n_features)
        self.tfidf = TfidfTransformer(sublinear_tf=sublinear_tf)

    def fit_transform(self, X, y=None):
        X_hashed = self.hasher.fit_transform(X)
        self.mask = np.array((X_hashed != 0).sum(axis=0)).flatten() >= self.min_df
        X_masked = X_hashed[:, self.mask]
        return self.tfidf.fit_transform(X_masked)

    def fit(self, X, y=None):
        self.fit_transform(X, y)
        return self

    def transform(self, X):
        X_hashed = self.hasher.transform(X)
        X_masked = X_hashed[:, self.mask]
        return self.tfidf.transform(X_masked)