

In [ ]:

    
%matplotlib notebook
import matplotlib.pyplot as plt
import numpy as np

Cross-Validation



In [ ]:

    
from sklearn.datasets import load_iris



In [ ]:

    
iris = load_iris()
X = iris.data
y = iris.target



In [ ]:

    
from sklearn.cross_validation import cross_val_score
from sklearn.svm import LinearSVC



In [ ]:

    
cross_val_score(LinearSVC(), X, y, cv=5)



In [ ]:

    
cross_val_score(LinearSVC(), X, y, cv=5, scoring="f1_macro")

Let's go to a binary task for a moment



In [ ]:

    
y % 2



In [ ]:

    
cross_val_score(LinearSVC(), X, y % 2)



In [ ]:

    
cross_val_score(LinearSVC(), X, y % 2, scoring="average_precision")



In [ ]:

    
cross_val_score(LinearSVC(), X, y % 2, scoring="roc_auc")



In [ ]:

    
from sklearn.metrics.scorer import SCORERS
print(SCORERS.keys())

There are other ways to do cross-valiation



In [ ]:

    
from sklearn.cross_validation import ShuffleSplit

shuffle_split = ShuffleSplit(len(X), 10, test_size=.4)
cross_val_score(LinearSVC(), X, y, cv=shuffle_split)



In [ ]:

    
from sklearn.cross_validation import StratifiedKFold, KFold, ShuffleSplit

def plot_cv(cv, n_samples):
    masks = []
    for train, test in cv:
        mask = np.zeros(n_samples, dtype=bool)
        mask[test] = 1
        masks.append(mask)
    plt.figure(figsize=(10, 4))
    plt.subplots_adjust(left=0, bottom=0, right=1, top=1)
    plt.imshow(masks, interpolation='none')



In [ ]:

    
plot_cv(StratifiedKFold(y, n_folds=5), len(y))



In [ ]:

    
plot_cv(KFold(len(iris.target), n_folds=5), len(iris.target))



In [ ]:

    
plot_cv(ShuffleSplit(len(iris.target), n_iter=20, test_size=.2), 
        len(iris.target))

Exercises

Use KFold cross validation and StratifiedKFold cross validation (3 or 5 folds) for LinearSVC on the iris dataset. Why are the results so different? How could you get more similar results?



In [ ]:

    
# %load solutions/cross_validation_iris.py