cross_val_score和估算器得分之间的差异?

时间:2014-03-05 21:43:30

标签: python scikit-learn

我正在使用Scikit-learn。我尝试使用纯交叉验证程序和快速cross_validation.cross_val_score进行交叉验证。但我发现我得到了不同的数字。为什么呢?

import numpy as np
from sklearn import cross_validation, datasets

digits = datasets.load_digits()
X = digits.data
y = digits.target

svc = svm.SVC(kernel='linear')

kfold = cross_validation.KFold(len(X))
scores = [svc.fit(X[train], y[train]).score(X[test], y[test]) for train, test in kfold]

#scores output: [0.93489148580968284, 0.95659432387312182, 0.93989983305509184]

cross_validation.cross_val_score(svc, X, y)
#output: array([ 0.98 ,  0.982,  0.983])

1 个答案:

答案 0 :(得分:1)

正如cross_val_score的文档字符串告诉你的那样,当给出类标签(整数)的目标向量时,它会进行分层交叉验证。

>>> kfold = cross_validation.StratifiedKFold(y)
>>> [svc.fit(X[train], y[train]).score(X[test], y[test])
...  for train, test in kfold]
[0.93521594684385378, 0.95826377295492482, 0.93791946308724827]
>>> cross_validation.cross_val_score(svc, X, y)
array([ 0.93521595,  0.95826377,  0.93791946])