def test_classifications(self):
false_positives = 0
false_negatives = 0
correct = 0
wrong = 0
engine = create_engine('sqlite:///data.db')
Session = sessionmaker(bind=engine)
session = Session()
training_data = session.query(model.Corpus).all()
training_values = [rec.title + ' ' + rec.text for rec in training_data]
training_targets = [rec.category for rec in training_data]
training_values, testing_values, training_targets, testing_targets = cross_validation.train_test_split(training_values, training_targets, test_size=0.3, random_state=0)
classifier = main.Classifier(training_values, training_targets)
for (i, message_text) in enumerate(testing_values):
classification = classifier.classify(message_text)[0]
if testing_targets[i] == 'good' and classification != 'good':
false_positives += 1
elif testing_targets[i] != 'good' and classification == 'good':
false_negatives += 1
elif testing_targets[i] == classification:
correct += 1
else:
wrong += 1
print('{} false positives ({})'.format(false_positives, float(false_positives)/len(testing_values)))
print('{} false negatives ({})'.format(false_negatives, float(false_negatives)/len(testing_values)))
print('{} correct ({})'.format(correct, float(correct)/len(testing_values)))
print('{} wrong ({})'.format(wrong, float(wrong)/len(testing_values)))
def test_classifications(self):
false_positives = 0
false_negatives = 0
correct = 0
wrong = 0
engine = create_engine('sqlite:///data.db')
Session = sessionmaker(bind=engine)
session = Session()
training_data = session.query(model.Corpus).all()
training_values = [rec.title + ' ' + rec.text for rec in training_data]
training_targets = [rec.category for rec in training_data]
training_values, testing_values, training_targets, testing_targets = cross_validation.train_test_split(
training_values, training_targets, test_size=0.2, random_state=0)
classifier = main.Classifier(training_values, training_targets)
for (i, message_text) in enumerate(testing_values):
classification = classifier.classify(message_text)[0]
if testing_targets[i] == 'good' and classification != 'good':
false_positives += 1
print(message_text)
print('[Suspected {}; actually good]'.format(classification))
print('---')
elif testing_targets[i] != 'good' and classification == 'good':
false_negatives += 1
elif testing_targets[i] == classification:
correct += 1
else:
wrong += 1
print(message_text)
print('[Suspected {}; actually {}]'.format(
classification, testing_targets[i]))
print('---')
print('{} false positives ({})'.format(
false_positives,
float(false_positives) / len(testing_values)))
print('{} false negatives ({})'.format(
false_negatives,
float(false_negatives) / len(testing_values)))
print('{} correct ({})'.format(correct,
float(correct) / len(testing_values)))
print('{} wrong ({})'.format(wrong,
float(wrong) / len(testing_values)))
if float(false_positives) / len(testing_values) > 0.05:
raise Exception('False positive rate too high!')
elif float(correct) / len(testing_values) < 0.6:
raise Exception('Correct identification rate too low!')
def makeDocx(self):
self.progressBar.setProperty("value", 0)
cls = classifier.Classifier()
nature = 0
for n, line in enumerate(self.lines):
found = cls.predict(line)
found_class = cls.getClass(found)
if (found_class == "N"):
nature += 1
self.progressBar.setProperty("value", (n / len(self.lines)) * 100)
self.progressBar.setProperty("value", 100)
cls.createDocx(self.inputFile.split("/")[-1])
self.labelNumOfNature.setText("Znaleziono " + str(nature) +
" zdań o przyrodzie\nz " +
str(len(self.lines)) + " wszytkich zdań")
import main
import model
import settings
param_grid = [
{
'C': [1, 5, 10, 50],
'loss': ['hinge', 'squared_hinge'],
'tol': [1e-6, 1e-4, 1e-2, 1e-1],
'multi_class': ['ovr', 'crammer_singer'],
'class_weight': ['balanced']
}
]
if __name__ == '__main__':
engine = create_engine(settings.DATABASE_URI)
Session = sessionmaker(bind=engine)
session = Session()
data = session.query(model.Corpus).all()
data_values = [col.title + ' ' + col.text for col in data]
data_targets = [col.category for col in data]
classifier = main.Classifier()
classifier.vectorizer.fit_transform(data_values)
grid_search = GridSearchCV(classifier.classifier, param_grid, n_jobs=-1)
grid_search.fit(classifier.vectorizer.transform(data_values), data_targets)
print('Best score: {}'.format(grid_search.best_score_))
parameters = grid_search.best_estimator_.get_params()
for parameter in parameters.keys():
print("{} - {}".format(parameter, parameters[parameter]))