python - 在 Sklearn Pipeline 中将 VotingClassifier 与其他分类器一起使用

Question

我想在 sklearn Pipeline 中使用 VotingClassifier，我在其中定义了一组分类器..

我从这个问题中得到了一些直觉:Using VotingClassifier in Sklearn Pipeline构建下面的代码，但在这个问题中，每个分类器都在独立的管道中定义。我不想以这种方式使用它，我之前准备了一组功能，这不是一个好主意使用不同的分类器在多个管道中重复生成这些特征(耗时的过程)!

我怎样才能实现这个目标？!

model = Pipeline([
        ('feat', FeatureUnion([
            ('tfidf', TfidfVectorizer(analyzer='char', ngram_range=(3, 5), min_df=0.01, lowercase=True, tokenizer=tokenizeTfidf)),    
        ])),


        ('pip1', Pipeline([('clf1', GradientBoostingClassifier(n_estimators=1000, random_state=7))])),
        ('pip2', Pipeline([('clf2', SVC())])),
        ('pip3', Pipeline([('clf3', RandomForestClassifier())])),
        ('clf', VotingClassifier(estimators=["pip1", "pip2", "pip3"]))
    ])

clf = model.fit(X_train, y_train)

但是我收到了这个错误:

 ('clf', VotingClassifier(estimators=["pip1", "pip2", "pip3"])),
  File "C:\Python35\lib\site-packages\imblearn\pipeline.py", line 115, in __init__
    self._validate_steps()
  File "C:\Python35\lib\site-packages\imblearn\pipeline.py", line 139, in _validate_steps
    "(but not both) '%s' (type %s) doesn't)" % (t, type(t)))
TypeError: All intermediate steps of the chain should be estimators that implement fit and transform or sample (but not both) 'Pipeline(memory=None,
     steps=[('clf1', GradientBoostingClassifier(criterion='friedman_mse', init=None,
              learning_rate=0.1, loss='deviance', max_depth=3,
              max_features=None, max_leaf_nodes=None,
              min_impurity_decrease=0.0, min_impurity_split=None,
              min_samples_leaf=1, min_samples_split=2,
              min_weight_fraction_leaf=0.0, n_estimators=1000,
              presort='auto', random_state=7, subsample=1.0, verbose=0,
              warm_start=False))])' (type <class 'imblearn.pipeline.Pipeline'>) doesn't)

Answer 1

我假设你想做这样的事情:

1) 使用 TfidfVectorizer 将文本数据转换为 tfidf2)将转换后的数据发送到3个估计器(GradientBoostingClassifier、SVC、RandomForestClassifier)，然后使用投票来获得预测。

如果是这种情况，这就是您所需要的。

model = Pipeline([
        ('feat', FeatureUnion([
            ('tfidf', TfidfVectorizer(analyzer='char',  
                                      ngram_range=(3, 5), 
                                      min_df=0.01, 
                                      lowercase=True, 
                                      tokenizer=tokenizeTfidf)),    
        ])),
        ('clf', VotingClassifier(estimators=[("pip1", GradientBoostingClassifier(n_estimators=1000, 
                                                                                 random_state=7)), 
                                             ("pip2", SVC()), 
                                             ("pip3", RandomForestClassifier())]))
    ])

此外，如果您仅使用单个 TfidfVectorizer 并且未将任何其他功能与其组合，则甚至不需要 FeatureUnion:

model = Pipeline([
        ('tfidf', TfidfVectorizer(analyzer='char',  
                                  ngram_range=(3, 5), 
                                  min_df=0.01, 
                                  lowercase=True, 
                                  tokenizer=tokenizeTfidf)),    
        ('clf', VotingClassifier(estimators=[("pip1", GradientBoostingClassifier(n_estimators=1000, 
                                                                                 random_state=7)), 
                                             ("pip2", SVC()), 
                                             ("pip3", RandomForestClassifier())]))
    ])