scikit-learn Cookbook , Second Edition - Second Edition

Cross-validation with the noise parameter

You might wonder if that is the best noise parameter, alpha=5? To figure this out, try some cross-validation.

1. First, produce a cross-validation score with alpha=5. Note the scorer within the cross_val_score object is neg_mean_absolute_error, as the default R-squared score is hard to read for this dataset:

from sklearn.model_selection import cross_val_score gpr5 = GaussianProcessRegressor(alpha=5,                                n_restarts_optimizer=20,                                kernel = mixed_kernel) scores_5 = (cross_val_score(gpr5,                             boston_X[train_set],                             boston_y[train_set],                             cv = 4,                             scoring = 'neg_mean_absolute_error'))

2. Look at the scores in scores_5:

def score_mini_report(scores_list):     print "List of scores: ", scores_list print "Mean of scores: ", ...