5.1. Cross-Validation: evaluating estimator performance¶
Learning the parameters of a prediction function and testing it on the same data is a methodological mistake: a model that would just repeat the labels of the samples that it has just seen would have a perfect score but would fail to predict anything useful on yet-unseen data.
To avoid over-fitting, we have to define two different sets : a training set X_train, y_train which is used for learning the parameters of a predictive model, and a testing set X_test, y_test which is used for evaluating the fitted predictive model.
In scikit-learn such a random split can be quickly computed with the train_test_split helper function. Let load the iris data set to fit a linear Support Vector Machine model on it:
>>> import numpy as np >>> from sklearn import cross_validation >>> from sklearn import datasets >>> from sklearn import svm >>> iris = datasets.load_iris() >>> iris.data.shape, iris.target.shape ((150, 4), (150,))
We can now quickly sample a training set while holding out 40% of the data for testing (evaluating) our classifier:
>>> X_train, X_test, y_train, y_test = cross_validation.train_test_split( ... iris.data, iris.target, test_size=0.4, random_state=0) >>> X_train.shape, y_train.shape ((90, 4), (90,)) >>> X_test.shape, y_test.shape ((60, 4), (60,)) >>> clf = svm.SVC(kernel='linear', C=1).fit(X_train, y_train) >>> clf.score(X_test, y_test) 0.96...
However, by defining these two sets, we drastically reduce the number of samples which can be used for learning the model, and the results can depend on a particular random choice for the pair of (train, test) sets.
A solution is to split the whole data several consecutive times in different train set and test set, and to return the averaged value of the prediction scores obtained with the different sets. Such a procedure is called cross-validation. This approach can be computationally expensive, but does not waste too much data (as it is the case when fixing an arbitrary test set), which is a major advantage in problem such as inverse inference where the number of samples is very small.
5.1.1. Computing cross-validated metrics¶
The simplest way to use perform cross-validation in to call the cross_val_score helper function on the estimator and the dataset.
The following example demonstrates how to estimate the accuracy of a linear kernel Support Vector Machine on the iris dataset by splitting the data and fitting a model and computing the score 5 consecutive times (with different splits each time):
>>> clf = svm.SVC(kernel='linear', C=1) >>> scores = cross_validation.cross_val_score( ... clf, iris.data, iris.target, cv=5) ... >>> scores array([ 1. ..., 0.96..., 0.9 ..., 0.96..., 1. ])
The mean score and the standard deviation of the score estimate are hence given by:
>>> print "Accuracy: %0.2f (+/- %0.2f)" % (scores.mean(), scores.std() / 2) Accuracy: 0.97 (+/- 0.02)
By default, the score computed at each CV iteration is the score method of the estimator. It is possible to change this by passing a custom scoring function, e.g. from the metrics module:
>>> from sklearn import metrics >>> cross_validation.cross_val_score(clf, iris.data, iris.target, cv=5, ... score_func=metrics.f1_score) ... array([ 1. ..., 0.96..., 0.89..., 0.96..., 1. ])
In the case of the Iris dataset, the samples are balanced across target classes hence the accuracy and the F1-score are almost equal.
It is also possible to use othe cross validation strategies by passing a cross validation iterator instead, for instance:
>>> n_samples = iris.data.shape >>> cv = cross_validation.ShuffleSplit(n_samples, n_iter=3, ... test_size=0.3, random_state=0) >>> cross_validation.cross_val_score(clf, iris.data, iris.target, cv=cv) ... array([ 0.97..., 0.97..., 1. ])
The available cross validation iterators are introduced in the following.
5.1.2. Cross validation iterators¶
The following sections list utilities to generate boolean masks or indices that can be used to generate dataset splits according to different cross validation strategies.
Boolean mask vs integer indices
Most cross validators support generating both boolean masks or integer indices to select the samples from a given fold.
When the data matrix is sparse, only the integer indices will work as expected. Integer indexing is hence the default behavior (since version 0.10).
You can explicitly pass indices=False to the constructor of the CV object (when supported) to use the boolean mask method instead.
KFold divides all the samples in math:K groups of samples, called folds (if , this is equivalent to the Leave One Out strategy), of equal sizes (if possible). The prediction function is learned using folds, and the fold left out is used for test.
Example of 2-fold:
>>> import numpy as np >>> from sklearn.cross_validation import KFold >>> X = np.array([[0., 0.], [1., 1.], [-1., -1.], [2., 2.]]) >>> Y = np.array([0, 1, 0, 1]) >>> kf = KFold(len(Y), n_folds=2, indices=False) >>> print kf sklearn.cross_validation.KFold(n=4, n_folds=2) >>> for train, test in kf: ... print train, test [False False True True] [ True True False False] [ True True False False] [False False True True]
Each fold is constituted by two arrays: the first one is related to the training set, and the second one to the test set. Thus, one can create the training/test sets using:
>>> X_train, X_test, y_train, y_test = X[train], X[test], Y[train], Y[test]
If X or Y are scipy.sparse matrices, train and test need to be integer indices. It can be obtained by setting the parameter indices to True when creating the cross-validation procedure:
>>> X = np.array([[0., 0.], [1., 1.], [-1., -1.], [2., 2.]]) >>> Y = np.array([0, 1, 0, 1]) >>> kf = KFold(len(Y), n_folds=2, indices=True) >>> for train, test in kf: ... print train, test [2 3] [0 1] [0 1] [2 3]
220.127.116.11. Stratified K-Fold¶
StratifiedKFold is a variation of K-fold, which returns stratified folds, i.e which creates folds by preserving the same percentage for each target class as in the complete set.
Example of stratified 2-fold:
>>> from sklearn.cross_validation import StratifiedKFold >>> X = [[0., 0.], ... [1., 1.], ... [-1., -1.], ... [2., 2.], ... [3., 3.], ... [4., 4.], ... [0., 1.]] >>> Y = [0, 0, 0, 1, 1, 1, 0] >>> skf = StratifiedKFold(Y, 2) >>> print skf sklearn.cross_validation.StratifiedKFold(labels=[0 0 0 1 1 1 0], n_folds=2) >>> for train, test in skf: ... print train, test [1 4 6] [0 2 3 5] [0 2 3 5] [1 4 6]
18.104.22.168. Leave-One-Out - LOO¶
LeaveOneOut (or LOO) is a simple cross-validation. Each learning set is created by taking all the samples except one, the test set being the sample left out. Thus, for n samples, we have n different learning sets and n different tests set. This cross-validation procedure does not waste much data as only one sample is removed from the learning set:
>>> from sklearn.cross_validation import LeaveOneOut >>> X = np.array([[0., 0.], [1., 1.], [-1., -1.], [2., 2.]]) >>> Y = np.array([0, 1, 0, 1]) >>> loo = LeaveOneOut(len(Y)) >>> print loo sklearn.cross_validation.LeaveOneOut(n=4) >>> for train, test in loo: ... print train, test [1 2 3]  [0 2 3]  [0 1 3]  [0 1 2] 
22.214.171.124. Leave-P-Out - LPO¶
LeavePOut is very similar to Leave-One-Out, as it creates all the possible training/test sets by removing samples from the complete set.
Example of Leave-2-Out:
>>> from sklearn.cross_validation import LeavePOut >>> X = [[0., 0.], [1., 1.], [-1., -1.], [2., 2.]] >>> Y = [0, 1, 0, 1] >>> lpo = LeavePOut(len(Y), 2) >>> print lpo sklearn.cross_validation.LeavePOut(n=4, p=2) >>> for train, test in lpo: ... 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]
126.96.36.199. Leave-One-Label-Out - LOLO¶
LeaveOneLabelOut (LOLO) is a cross-validation scheme which holds out the samples according to a third-party provided label. This label information can be used to encode arbitrary domain specific stratifications of the samples as integers.
Each training set is thus constituted by all the samples except the ones related to a specific label.
For example, in the cases of multiple experiments, LOLO can be used to create a cross-validation based on the different experiments: we create a training set using the samples of all the experiments except one:
>>> from sklearn.cross_validation import LeaveOneLabelOut >>> X = [[0., 0.], [1., 1.], [-1., -1.], [2., 2.]] >>> Y = [0, 1, 0, 1] >>> labels = [1, 1, 2, 2] >>> lolo = LeaveOneLabelOut(labels) >>> print lolo sklearn.cross_validation.LeaveOneLabelOut(labels=[1, 1, 2, 2]) >>> for train, test in lolo: ... print train, test [2 3] [0 1] [0 1] [2 3]
Another common application is to use time information: for instance the labels could be the year of collection of the samples and thus allow for cross-validation against time-based splits.
LeavePLabelOut is similar as Leave-One-Label-Out, but removes samples related to labels for each training/test set.
Example of Leave-2-Label Out:
>>> from sklearn.cross_validation import LeavePLabelOut >>> X = [[0., 0.], [1., 1.], [-1., -1.], [2., 2.], [3., 3.], [4., 4.]] >>> Y = [0, 1, 0, 1, 0, 1] >>> labels = [1, 1, 2, 2, 3, 3] >>> lplo = LeavePLabelOut(labels, 2) >>> print lplo sklearn.cross_validation.LeavePLabelOut(labels=[1, 1, 2, 2, 3, 3], p=2) >>> for train, test in lplo: ... print train, test [4 5] [0 1 2 3] [2 3] [0 1 4 5] [0 1] [2 3 4 5]
188.8.131.52. Random permutations cross-validation a.k.a. Shuffle & Split¶
The ShuffleSplit iterator will generate a user defined number of independent train / test dataset splits. Samples are first shuffled and then splitted into a pair of train and test sets.
It is possible to control the randomness for reproducibility of the results by explicitly seeding the random_state pseudo random number generator.
Here is a usage example:
>>> ss = cross_validation.ShuffleSplit(5, n_iter=3, test_size=0.25, ... random_state=0) >>> len(ss) 3 >>> print ss ShuffleSplit(5, n_iter=3, test_size=0.25, indices=True, ...) >>> for train_index, test_index in ss: ... print train_index, test_index ... [1 3 4] [2 0] [1 4 3] [0 2] [4 0 2] [1 3]
184.108.40.206. See also¶
StratifiedShuffleSplit is a variation of ShuffleSplit, which returns stratified splits, i.e which creates splits by preserving the same percentage for each target class as in the complete set.
220.127.116.11. Bootstrapping cross-validation¶
Bootstrapping is a general statistics technique that iterates the computation of an estimator on a resampled dataset.
The Bootstrap iterator will generate a user defined number of independent train / test dataset splits. Samples are then drawn (with replacement) on each side of the split. It furthermore possible to control the size of the train and test subset to make their union smaller than the total dataset if it is very large.
Contrary to other cross-validation strategies, bootstrapping will allow some samples to occur several times in each splits.
>>> bs = cross_validation.Bootstrap(9, random_state=0) >>> len(bs) 3 >>> print bs Bootstrap(9, n_iter=3, train_size=5, test_size=4, random_state=0)
>>> for train_index, test_index in bs: ... print train_index, test_index ... [1 8 7 7 8] [0 3 0 5] [5 4 2 4 2] [6 7 1 0] [4 7 0 1 1] [5 3 6 5]