def train_model(self,
clf_type,
features,
labels,
params=None,
logging=False):
skf = StratifiedKFold(n_splits=10, shuffle=True, random_state=23)
# params used for setting model after training
if params is None:
clf = self._get_classifier(clf_type)
else:
if clf_type == 'svm':
clf = SVC(C=params['C'],
gamma=params['Gamma'],
kernel='rbf',
decision_function_shape='ovr',
probability=True)
elif clf_type == 'lgr':
clf = LogisticRegression(C=params['C'], solver='newton-cg')
_C, _Gamma, _Scores, _Accuracies, _Precisions, _Recalls = np.array(
[]), np.array([]), np.array([]), np.array([]), np.array(
[]), np.array([])
selected_params = dict()
_Probabilities, _Indicies, _True_Labels = list(), list(), list()
for train_idx, test_idx in skf.split(features, labels):
# if clf_type == 'xgb':
# if logging:
# watchlist = [(dfeatures, 'train')]
# dfeatures = xgb.DMatrix(
# features[train_idx], labels=labels[train_idx])
# bst = xgb.train(clf, dfeatures, watchlist, 5)
# else:
clf.fit(features[train_idx], labels[train_idx])
if params is None:
if logging:
Log.classifier_details(clf_type, clf)
if clf_type == 'svm':
_C = np.append(_C, clf.best_params_['C'])
_Gamma = np.append(_Gamma, clf.best_params_['gamma'])
_Scores = np.append(_Scores, clf.best_score_)
else:
predictions = clf.predict(features[test_idx])
_Probabilities.append(clf.predict_proba(features[test_idx]))
_Indicies.append(test_idx)
_True_Labels.append(labels[test_idx])
_Accuracies = np.append(
_Accuracies, accuracy_score(predictions, labels[test_idx]))
from sklearn.metrics import precision_score, recall_score
_Precisions = np.append(
_Precisions,
precision_score(predictions,
labels[test_idx],
average='macro'))
_Recalls = np.append(
_Precisions,
recall_score(predictions,
labels[test_idx],
average='macro'))
print('Accuracy score: ',
accuracy_score(predictions, labels[test_idx]))
if logging and params is None:
Log.cross_val_details(_C, _Gamma, _Scores)
if params is None:
selected_params = {
'C': np.dot(np.mean(_C),
np.mean(_Scores).T),
'Gamma': np.dot(np.mean(_Gamma),
np.mean(_Scores).T)
}
return _Accuracies, selected_params, _Precisions, _Recalls
