scikit learn:未选择所需的最佳功能(k)

Question

我试图使用卡方(scikit-learn 0.10)选择最佳功能.从总共80个培训文档中我首先提取了227个特征,并从这227个特征中我想选择前10个特征.

my_vectorizer = CountVectorizer(analyzer=MyAnalyzer())      
X_train = my_vectorizer.fit_transform(train_data)
X_test = my_vectorizer.transform(test_data)
Y_train = np.array(train_labels)
Y_test = np.array(test_labels)
X_train = np.clip(X_train.toarray(), 0, 1)
X_test = np.clip(X_test.toarray(), 0, 1)    
ch2 = SelectKBest(chi2, k=10)
print X_train.shape
X_train = ch2.fit_transform(X_train, Y_train)
print X_train.shape

结果如下.

(80, 227)
(80, 14)

如果我设置k相等,它们是相似的100.

(80, 227)
(80, 227)

为什么会这样？

*编辑:一个完整​​的输出示例,现在没有剪切,我请求30并获得32:

Train instances: 9 Test instances: 1
Feature extraction...
X_train:
[[0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
 [0 0 1 0 1 1 0 1 1 0 0 0 1 0 1 0 0 0 0 1 1 1 0 0 1 0 0 1 0 0 0 0]
 [0 0 2 1 0 0 0 0 0 1 0 0 0 1 1 0 0 0 1 0 0 0 0 1 0 1 1 0 0 1 0 1]
 [1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
 [1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 1 0]
 [0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 1 0 0 0]]
Y_train:
[0 0 0 0 0 0 0 0 1]
32 features extracted from 9 training documents.
Feature selection...
(9, 32)
(9, 32)
Using 32(requested:30) best features from 9 training documents
get support:
[ True  True  True  True  True  True  True  True  True  True  True  True
  True  True  True  True  True  True  True  True  True  True  True  True
  True  True  True  True  True  True  True  True]
get support with vocabulary :
[ 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
 25 26 27 28 29 30 31]
Training...
/usr/local/lib/python2.6/dist-packages/scikit_learn-0.10-py2.6-linux-x86_64.egg/sklearn/svm/sparse/base.py:23: FutureWarning: SVM: scale_C will be True by default in scikit-learn 0.11
  scale_C)
Classifying...

另一个没有剪辑的例子,我请求10并获得11:

Train instances: 9 Test instances: 1
Feature extraction...
X_train:
[[0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
 [0 0 1 0 1 1 0 1 1 0 0 0 1 0 1 0 0 0 0 1 1 1 0 0 1 0 0 1 0 0 0 0]
 [0 0 2 1 0 0 0 0 0 1 0 0 0 1 1 0 0 0 1 0 0 0 0 1 0 1 1 0 0 1 0 1]
 [1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0]
 [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
 [1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 1 0]
 [0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 1 0 0 0]]
Y_train:
[0 0 0 0 0 0 0 0 1]
32 features extracted from 9 training documents.
Feature selection...
(9, 32)
(9, 11)
Using 11(requested:10) best features from 9 training documents
get support:
[ True  True  True False False  True False False False False  True False
 False False  True False False False  True False  True False  True  True
 False False False False  True False False False]
get support with vocabulary :
[ 0  1  2  5 10 14 18 20 22 23 28]
Training...
/usr/local/lib/python2.6/dist-packages/scikit_learn-0.10-py2.6-linux-x86_64.egg/sklearn/svm/sparse/base.py:23: FutureWarning: SVM: scale_C will be True by default in scikit-learn 0.11
  scale_C)
Classifying...

Answer 1

你有没有检查过从get_support()函数返回的内容(ch2应该有这个成员函数)？这返回在最佳k中选择的索引.

我的猜想是,由于你正在做的数据剪辑(或者由于重复的特征向量,如果你的特征向量是分类的并且可能有重复)存在关系,并且scikits函数返回所有被绑定的条目对于前k个点.您设置的额外示例k = 100对此猜想产生了一些疑问,但值得一看.

查看get_support()返回的内容,并查看X_train这些索引上的内容,查看裁剪是否会导致大量特征重叠,从而在SelectKBest正在使用的chi ^ 2 p值排名中创建关联.

如果情况确实如此,那么您应该向scikits.learn提交错误/问题,因为目前他们的文档没有说明SelectKBest在发生关系时会做什么.显然,它不仅可以采取一些并列指数而不是其他指数,但至少应该警告用户关系可能导致意外的特征维数减少.