def test_IsSystemRunningThroughWithAllFeatures(self):
system = System(self.testdata)
# Turn on features
system.use_dependency_is_root()
system.use_dependency_order()
system.use_aspect()
system.use_tense()
system.use_same_tense()
system.use_same_aspect()
system.use_dependency_type()
system.use_dct()
system.use_type()
system.use_same_polarity()
system.use_polarity()
system.use_class()
system.use_entity_distance()
system.use_textual_order()
system.use_duration()
system.use_duration_difference()
system.use_same_pos()
system.use_pos()
system.use_temporal_signal()
system.create_features()
system.train()
system.eval()
self.assertTrue(True)
def _run(self):
data = Data()
data.training = self.training
system = System(data)
system.use_best_feature_set()
system.create_features()
system.train()
system.eval(quiet=True)
self.result_event_event = system.evaluation_accuracy_event_event
self.result_event_timex = system.evaluation_accuracy_event_timex
def measure_time(data):
start = time.time()
system = System(data)
system.use_all_features()
system.use_feature_selection()
system.create_features()
system.train()
system.save_classifiers()
system.eval()
system.create_confidence_scores()
system.apply_global_model()
end = time.time()
return end-start
def _run_systems(self):
for k in range(1, self.max_len+1):
features = list(set(self._feature_series(k, self.features_event_event) + self._feature_series(k, self.features_event_timex)))
print features
data = Data()
data.training = TrainingSet(False, False, "data/training/TBAQ-cleaned/TimeBank/")
system = System(data, features)
system.create_features()
system.cross_validation()
now = list(set(self._feature_series(k, self.features_event_event)))
if k > 1:
prev = list(set(self._feature_series(k-1, self.features_event_event)))
if k > 1:
if now != prev:
self.accuracies_event_event.append(system.crossval_accuracy_event_event)
print system.crossval_accuracy_event_event
else:
self.accuracies_event_event.append(system.crossval_accuracy_event_event)
print system.crossval_accuracy_event_event
now = list(set(self._feature_series(k, self.features_event_timex)))
if k > 1:
prev = list(set(self._feature_series(k-1, self.features_event_timex)))
if k > 1:
if now != prev:
self.accuracies_event_timex.append(system.crossval_accuracy_event_timex)
print system.crossval_accuracy_event_timex
else:
self.accuracies_event_timex.append(system.crossval_accuracy_event_timex)
print system.crossval_accuracy_event_timex
print
# Leave out i for testing
train_X, test_X = leave_out(pieces_X, i)
train_y, test_y = leave_out(pieces_y, i)
# Transform back to sparse
train_X = transform_to_sparse_matrix(train_X)
test_X = transform_to_sparse_matrix(test_X)
clf.fit(train_X, train_y)
predicted = clf.predict(test_X)
accs.append(accuracy_score(test_y, predicted))
return np.mean(accs)
data = Data()
data.training = TrainingSet(False, False, "data/training/TBAQ-cleaned/TimeBank/")
system = System(data, ["lemma", "token"])
system.create_features()
X_event_event, y_event_event = system.training_event_event
X_event_timex, y_event_timex = system.training_event_timex
X_event_event = transform_to_list(X_event_event)
X_event_timex = transform_to_list(X_event_timex)
print kfold(X_event_event, y_event_event, 5)
print kfold(X_event_timex, y_event_timex, 5)
class Distance:
def __init__(self):
self.data = Data()
self.system = System(self.data)
self.system.use_all_features()
self.system.use_feature_selection()
self.system.create_features()
self.system.train()
self.system.save_predictions_to_relations()
self.relations = self._get_relations()
same_sentence, not_same_sentence = self._sort_relations_after_in_same_sentence()
print "Same sentence event-event:"
truth, predicted = self._get_truth_and_prediction(same_sentence)
print Result(truth, predicted)
print "Same sentence event-timex:"
truth, predicted = self._get_truth_and_prediction(same_sentence, event_event=False)
print Result(truth, predicted)
print "Not same sentence event-event:"
truth, predicted = self._get_truth_and_prediction(not_same_sentence)
print Result(truth, predicted)
print "Not same sentence event-timex:"
truth, predicted = self._get_truth_and_prediction(not_same_sentence, event_event=False)
print Result(truth, predicted)
def _get_truth_and_prediction(self, rels, event_event=True):
prediction = []
truth = []
for rel in rels:
if event_event and rel.is_event_event():
prediction.append(rel.predicted_class)
truth.append(rel.relation_type)
elif not event_event and rel.is_event_timex():
prediction.append(rel.predicted_class)
truth.append(rel.relation_type)
return truth, prediction
def _get_relations(self):
rels = []
for text_obj in self.data.test.text_objects:
for relation in text_obj.relations:
if relation.is_event_event():
rels.append(relation)
elif relation.is_event_timex():
rels.append(relation)
return rels
def _sort_relations_after_in_same_sentence(self):
same_sentence = []
not_same_sentence = []
for rel in self.relations:
if rel.source.sentence == rel.target.sentence:
same_sentence.append(rel)
else:
not_same_sentence.append(rel)
return (same_sentence, not_same_sentence)
