def test_smaller_dictionary(self):
taxonomy_dictionary = self.taxonomy_dictionary
taxonomy_dictionary['Stochastic'] = taxonomy_dictionary[
'Stochastic'][:-1]
classifier = HierarchicalRandomForest(self.taxonomy_dictionary)
with self.assertRaises(Exception):
classifier.fit(self.features, self.labels)
def test_larger_dictionary(self):
taxonomy_dictionary = self.taxonomy_dictionary
taxonomy_dictionary[
'Stochastic'] = taxonomy_dictionary['Stochastic'] + ['new class']
classifier = HierarchicalRandomForest(self.taxonomy_dictionary)
classifier.fit(self.features, self.labels)
predicted_probs = classifier.predict_proba(self.features)
self.assertEqual(predicted_probs.shape, (len(self.features), 15))
def test_fit(self):
classifier = HierarchicalRandomForest(self.taxonomy_dictionary)
classifier.fit(self.features, self.labels)
predicted_classes = classifier.predict(self.features)
recall_values = balanced_recall(predicted_classes, self.labels)
self.assertGreater(
recall_values, 0.95,
'Training balanced recall has to be greater than 95 %')
def test_feature_order_shuffle(self):
classifier = HierarchicalRandomForest(self.taxonomy_dictionary)
classifier.fit(self.features, self.labels)
new_columns = self.features.columns.values
np.random.shuffle(new_columns)
predicted_classes = classifier.predict(self.features[new_columns])
recall_values = balanced_recall(predicted_classes, self.labels)
self.assertGreater(
recall_values, 0.95,
'Training balanced recall has to be greater than 95 %')
def test_fit(self):
taxonomy_dictionary = {
'Stochastic': ['LPV', 'QSO', 'YSO', 'CV/Nova', 'Blazar', 'AGN'],
'Periodic': ['RRL', 'EB', 'DSCT', 'Ceph', 'Periodic-Other'],
'Transient': ['SNIa', 'SNII', 'SNIbc']
}
model = HierarchicalRandomForest(taxonomy_dictionary)
model.fit(self.train_features, self.train_labels)
probs = model.predict_proba(self.train_features)
print(probs.head())
predicted_classes = model.predict(self.train_features)
print(predicted_classes.head())
def test_predict_in_pipeline(self):
classifier = HierarchicalRandomForest(self.taxonomy_dictionary)
classifier.fit(self.features, self.labels)
features = self.features.iloc[0]
predicted_classes = classifier.predict_in_pipeline(features)
n_classes = len(classifier.get_list_of_classes())
self.dict_sum_one(predicted_classes['probabilities'])
self.assertEqual(n_classes, len(predicted_classes['probabilities']))
self.dict_sum_one(predicted_classes['hierarchical']['top'])
self.assertEqual(3, len(predicted_classes['hierarchical']['top']))
for children_probs in predicted_classes['hierarchical'][
'children'].values():
self.dict_sum_one(children_probs)
def test_predict_proba(self):
classifier = HierarchicalRandomForest(self.taxonomy_dictionary)
classifier.fit(self.features, self.labels)
predicted_probs = classifier.predict_proba(self.features)
self.is_sum_one(predicted_probs)
def test_save_and_load_hierarchical_rf(self):
classifier = HierarchicalRandomForest(self.taxonomy_dictionary)
classifier.fit(self.features, self.labels)
predicted_probs = classifier.predict_proba(self.features)
classifier.save_model(self.tmp_dir)
classifier2 = HierarchicalRandomForest(self.taxonomy_dictionary)
classifier2.load_model(self.tmp_dir)
predicted_probs2 = classifier2.predict_proba(self.features)
self.assertTrue((predicted_probs == predicted_probs2).all(axis=None))
stratify=labels.loc[train_val_oids].classALeRCE.values,
test_size=0.15 / 0.7,
random_state=0)
# needed for color augmentation
training_features = features.loc[train_oids]
training_labels = labels.loc[train_oids]
training_features.to_pickle('training_features.pkl')
training_labels.to_pickle('training_labels.pkl')
taxonomy_dictionary = {
'Stochastic': ['LPV', 'QSO', 'YSO', 'CV/Nova', 'Blazar', 'AGN'],
'Periodic': ['RRL', 'EB', 'DSCT', 'Ceph', 'Periodic-Other'],
'Transient': ['SNIa', 'SNII', 'SNIbc', 'SLSN']
}
classifier = HierarchicalRandomForest(taxonomy_dictionary)
classifier.fit(features.loc[train_oids], labels.loc[train_oids])
def evaluate_classifier(classifier):
test_predictions = classifier.predict(features.loc[test_oids])
test_confusion_matrix = confusion_matrix(test_predictions,
labels.loc[test_oids],
classifier.get_list_of_classes())
print(test_confusion_matrix)
test_predictions_proba = classifier.predict_proba(features.loc[test_oids])
kaggle_score_value = kaggle_score(test_predictions_proba,