sklearn中的交叉验证的实现(Cross-Validation) sklearn中的交叉验证的实现(Cr

sklearn是利用python进行机器学习中一个非常全面和好用的第三方库，用过的都说好。今天主要记录一下sklearn中关于交叉验证的各种用法，主要是对sklearn官方文档 Cross-validation: evaluating estimator performance进行讲解，英文水平好的建议读官方文档，里面的知识点很详细。
先导入需要的库及数据集

In [1]: import numpy as npIn [2]: from sklearn.model_selection import train_test_splitIn [3]: from sklearn.datasets import load_irisIn [4]: from sklearn import svmIn [5]: iris = load_iris()In [6]: iris.data.shape, iris.target.shapeOut[6]: ((150, 4), (150,))

1.train_test_split 对数据集进行快速打乱（分为训练集和测试集）
这里相当于对数据集进行了shuffle后按照给定的test_size 进行数据集划分。

In [7]: X_train, X_test, y_train, y_test = train_test_split(...:iris.data, iris.target, test_size=.4, random_state=0)#这里是按照6:4对训练集测试集进行划分In [8]: X_train.shape, y_train.shapeOut[8]: ((90, 4), (90,))In [9]: X_test.shape, y_test.shapeOut[9]: ((60, 4), (60,))In [10]: iris.data[:5]Out[10]: array([[ 5.1, 3.5, 1.4, 0.2],[ 4.9, 3. , 1.4, 0.2],[ 4.7, 3.2, 1.3, 0.2],[ 4.6, 3.1, 1.5, 0.2],[ 5. , 3.6, 1.4, 0.2]])In [11]: X_train[:5]Out[11]: array([[ 6. , 3.4, 4.5, 1.6],[ 4.8, 3.1, 1.6, 0.2],[ 5.8, 2.7, 5.1, 1.9],[ 5.6, 2.7, 4.2, 1.3],[ 5.6, 2.9, 3.6, 1.3]])In [12]: clf = svm.SVC(kernel='linear', C=1).fit(X_train, y_train)In [13]: clf.score(X_test, y_test)Out[13]: 0.96666666666666667

2.cross_val_score 对数据集进行指定次数的交叉验证并为每次验证效果评测
【sklearn中的交叉验证的实现(Cross-Validation)】其中，score 默认是以 scoring='f1_macro'进行评测的，余外针对分类或回归还有：

文章图片

这需要from　sklearn import metrics ,通过在cross_val_score 指定参数来设定评测标准；
当cv 指定为int 类型时，默认使用KFold 或StratifiedKFold 进行数据集打乱，下面会对KFold 和StratifiedKFold 进行介绍。

In [15]: from sklearn.model_selection import cross_val_scoreIn [16]: clf = svm.SVC(kernel='linear', C=1)In [17]: scores = cross_val_score(clf, iris.data, iris.target, cv=5)In [18]: scoresOut[18]: array([ 0.96666667, 1., 0.96666667, 0.96666667, 1.])In [19]: scores.mean()Out[19]: 0.98000000000000009

除使用默认交叉验证方式外，可以对交叉验证方式进行指定，如验证次数，训练集测试集划分比例等

In [20]: from sklearn.model_selection import ShuffleSplitIn [21]: n_samples = iris.data.shape[0]In [22]: cv = ShuffleSplit(n_splits=3, test_size=.3, random_state=0)In [23]: cross_val_score(clf, iris.data, iris.target, cv=cv)Out[23]: array([ 0.97777778, 0.97777778, 1.])

在cross_val_score 中同样可使用pipeline 进行流水线操作

In [24]: from sklearn import preprocessingIn [25]: from sklearn.pipeline import make_pipelineIn [26]: clf = make_pipeline(preprocessing.StandardScaler(), svm.SVC(C=1))In [27]: cross_val_score(clf, iris.data, iris.target, cv=cv)Out[27]: array([ 0.97777778, 0.93333333, 0.95555556])

3.cross_val_predict cross_val_predict 与cross_val_score 很相像，不过不同于返回的是评测效果，cross_val_predict 返回的是estimator 的分类结果（或回归值），这个对于后期模型的改善很重要，可以通过该预测输出对比实际目标值，准确定位到预测出错的地方，为我们参数优化及问题排查十分的重要。

In [28]: from sklearn.model_selection import cross_val_predictIn [29]: from sklearn import metricsIn [30]: predicted = cross_val_predict(clf, iris.data, iris.target, cv=10)In [31]: predictedOut[31]: array([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, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1,1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 1, 2, 2, 2, 2, 2, 2, 2, 2,2, 2, 2, 2, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 2, 2, 2, 2,2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2])In [32]: metrics.accuracy_score(iris.target, predicted)Out[32]: 0.96666666666666667

4.KFold K折交叉验证，这是将数据集分成K份的官方给定方案，所谓K折就是将数据集通过K次分割，使得所有数据既在训练集出现过，又在测试集出现过，当然，每次分割中不会有重叠。相当于无放回抽样。

In [33]: from sklearn.model_selection import KFoldIn [34]: X = ['a','b','c','d']In [35]: kf = KFold(n_splits=2)In [36]: for train, test in kf.split(X):...:print train, test...:print np.array(X)[train], np.array(X)[test]...:print '\n'...:[2 3] [0 1]['c' 'd'] ['a' 'b'][0 1] [2 3]['a' 'b'] ['c' 'd']

5.LeaveOneOut LeaveOneOut 其实就是KFold 的一个特例，因为使用次数比较多，因此独立的定义出来，完全可以通过KFold 实现。

In [37]: from sklearn.model_selection import LeaveOneOutIn [38]: X = [1,2,3,4]In [39]: loo = LeaveOneOut()In [41]: for train, test in loo.split(X):...:print train, test...:[1 2 3] [0][0 2 3] [1][0 1 3] [2][0 1 2] [3]#使用KFold实现LeaveOneOtutIn [42]: kf = KFold(n_splits=len(X))In [43]: for train, test in kf.split(X):...:print train, test...:[1 2 3] [0][0 2 3] [1][0 1 3] [2][0 1 2] [3]

6.LeavePOut 这个也是KFold 的一个特例，用KFold 实现起来稍麻烦些，跟LeaveOneOut 也很像。

In [44]: from sklearn.model_selection import LeavePOutIn [45]: X = np.ones(4)In [46]: lpo = LeavePOut(p=2)In [47]: for train, test in lpo.split(X):...:print train, test...:[2 3] [0 1][1 3] [0 2][1 2] [0 3][0 3] [1 2][0 2] [1 3][0 1] [2 3]

7.ShuffleSplit ShuffleSplit 咋一看用法跟LeavePOut 很像，其实两者完全不一样，LeavePOut 是使得数据集经过数次分割后，所有的测试集出现的元素的集合即是完整的数据集，即无放回的抽样，而ShuffleSplit 则是有放回的抽样，只能说经过一个足够大的抽样次数后，保证测试集出现了完成的数据集的倍数。

In [48]: from sklearn.model_selection import ShuffleSplitIn [49]: X = np.arange(5)In [50]: ss = ShuffleSplit(n_splits=3, test_size=.25, random_state=0)In [51]: for train_index, test_index in ss.split(X):...:print train_index, test_index...:[1 3 4] [2 0][1 4 3] [0 2][4 0 2] [1 3]

8.StratifiedKFold 这个就比较好玩了，通过指定分组，对测试集进行无放回抽样。

In [52]: from sklearn.model_selection import StratifiedKFoldIn [53]: X = np.ones(10)In [54]: y = [0,0,0,0,1,1,1,1,1,1]In [55]: skf = StratifiedKFold(n_splits=3)In [56]: for train, test in skf.split(X,y):...:print train, test...:[2 3 6 7 8 9] [0 1 4 5][0 1 3 4 5 8 9] [2 6 7][0 1 2 4 5 6 7] [3 8 9]

9.GroupKFold 这个跟StratifiedKFold 比较像，不过测试集是按照一定分组进行打乱的，即先分堆，然后把这些堆打乱，每个堆里的顺序还是固定不变的。

In [57]: from sklearn.model_selection import GroupKFoldIn [58]: X = [.1, .2, 2.2, 2.4, 2.3, 4.55, 5.8, 8.8, 9, 10]In [59]: y = ['a','b','b','b','c','c','c','d','d','d']In [60]: groups = [1,1,1,2,2,2,3,3,3,3]In [61]: gkf = GroupKFold(n_splits=3)In [62]: for train, test in gkf.split(X,y,groups=groups):...:print train, test...:[0 1 2 3 4 5] [6 7 8 9][0 1 2 6 7 8 9] [3 4 5][3 4 5 6 7 8 9] [0 1 2]

10.LeaveOneGroupOut 这个是在GroupKFold 上的基础上混乱度又减小了，按照给定的分组方式将测试集分割下来。

In [63]: from sklearn.model_selection import LeaveOneGroupOutIn [64]: X = [1, 5, 10, 50, 60, 70, 80]In [65]: y = [0, 1, 1, 2, 2, 2, 2]In [66]: groups = [1, 1, 2, 2, 3, 3, 3]In [67]: logo = LeaveOneGroupOut()In [68]: for train, test in logo.split(X, y, groups=groups):...:print train, test...:[2 3 4 5 6] [0 1][0 1 4 5 6] [2 3][0 1 2 3] [4 5 6]

11.LeavePGroupsOut 这个没啥可说的，跟上面那个一样，只是一个是单组，一个是多组

from sklearn.model_selection import LeavePGroupsOutX = np.arange(6)y = [1, 1, 1, 2, 2, 2]groups = [1, 1, 2, 2, 3, 3]lpgo = LeavePGroupsOut(n_groups=2)for train, test in lpgo.split(X, y, groups=groups):print train, test[4 5] [0 1 2 3][2 3] [0 1 4 5][0 1] [2 3 4 5]

12.GroupShuffleSplit 这个是有放回抽样

In [75]: from sklearn.model_selection import GroupShuffleSplitIn [76]: X = [.1, .2, 2.2, 2.4, 2.3, 4.55, 5.8, .001]In [77]: y = ['a', 'b','b', 'b', 'c','c', 'c', 'a']In [78]: groups = [1,1,2,2,3,3,4,4]In [79]: gss = GroupShuffleSplit(n_splits=4, test_size=.5, random_state=0)In [80]: for train, test in gss.split(X, y, groups=groups):...:print train, test...:[0 1 2 3] [4 5 6 7][2 3 6 7] [0 1 4 5][2 3 4 5] [0 1 6 7][4 5 6 7] [0 1 2 3]

13.TimeSeriesSplit 针对时间序列的处理，防止未来数据的使用，分割时是将数据进行从前到后切割（这个说法其实不太恰当，因为切割是延续性的。。）

In [81]: from sklearn.model_selection import TimeSeriesSplitIn [82]: X = np.array([[1,2],[3,4],[1,2],[3,4],[1,2],[3,4]])In [83]: tscv = TimeSeriesSplit(n_splits=3)In [84]: for train, test in tscv.split(X):...:print train, test...:[0 1 2] [3][0 1 2 3] [4][0 1 2 3 4] [5]

这个repo 用来记录一些python技巧、书籍、学习链接等，欢迎star github地址