def cross_validate(self, data=None, num_folds=10):
"""
Split the training data into N folds. Use N-1 folds as training data
and 1 fold as the testing data during each iteration.
Outputs the confusion matrix and other accuracy measurement stats the end of N-CV.
"""
if data is None:
data = self.data
if len(data) < num_folds:
raise ValueError(
'Not enough data to make {} folds'.format(num_folds))
folds = folds_indexes(data, num_folds)
folds_accuracy = []
cm = ConfusionMatrix(np.unique(data.index.values))
for i, (train_idx, test_idx) in enumerate(folds):
# find the indexes of other folds data in the dataframe
print('Fold {}'.format(i))
training_data = data.iloc[train_idx]
test_data = data.iloc[test_idx]
self.train(training_data)
# predict and evaluate
predictions = self.predict(test_data.values)
real_vals = test_data.index.values
accuracy = accuracy_score(predictions, real_vals)
cm.update(predictions, real_vals)
# clean up
folds_accuracy.append(accuracy)
self.svm_units = []
print(accuracy)
print('Mean Accuracy : {}'.format(np.mean(folds_accuracy)))
print('***** Detailed Summary *****')
print(cm.statistics())
print('***** Confusion Matrix *****')
print(cm)
cm.save(folder='../experiments/')
return self
