python - 从 L1 正则化逻辑回归中恢复命名特征

Question

我有以下管道:

sg = Pipeline([('tfidf', TfidfVectorizer()),                                                                                                                                            
               ('normalize', Normalizer()),                                                                                                                                                 
               ('l1', LogisticRegression(penalty="l1", dual=False))])

执行拟合后，我想提取对应的标记到非零权重。

我该怎么做？

Answer 1

features = pipeline.named_steps['tfidf'].get_feature_names()
print(features[pipeline.named_steps['l1'].coef_ != 0])

参见TfidfTransformer docs , LogisticRegression docs和 unmerged improved pipeline docs here