loss_hist = []
# Train
start = time.time()
for t in range(5000):
X_train_mini, y_train_mini = get_mini_batching(X_train, y_train,
size_batch)
# Berechne die Vorhersage (foward step)
outputs = model(X_train_mini)
# Berechne den Fehler (Ausgabe des Fehlers alle 50 Iterationen)
loss = criterion(outputs, y_train_mini.long())
# Berechne die Gradienten und Aktualisiere die Gewichte (backward step)
optimizer.zero_grad()
loss.backward()
optimizer.step()
# Berechne den Fehler (Ausgabe des Fehlers alle 100 Iterationen)
if t % 50 == 0:
loss_hist.append(loss.item())
#print(t, loss.item())
print(time.time() - start)
if args.s_model:
name = f'logreg_{n_features}.sav'
pre.save_model(model, name=name)
#shuffle datafram in-place
train_features = train_features.sample(frac=1).reset_index(drop=True)
pre.set('training_data' + '_' + str(n_words), train_features)
pre.save('training_data' + '_' + str(n_words))
#transform the test data
TEST_PATH_POS = 'test/pos/'
TEST_PATH_NEG = 'test/neg/'
pre.load_all_texts_from_directory(path=TEST_PATH_POS, name='raw_pos_test')
pre.load_all_texts_from_directory(path=TEST_PATH_NEG, name='raw_neg_test')
print('transform test data')
texts_list_test = pd.concat(
[pre.data['raw_pos_test'], pre.data['raw_neg_test']])
test_features = text_transformer.transform(texts_list_test)
test_features['target'] = np.append(
np.ones(len(pre.data['raw_pos_test'])),
np.zeros(len(pre.data['raw_neg_test'])))
test_features = test_features.sample(frac=1).reset_index(drop=True)
pre.set('test_data' + '_' + str(n_words), test_features)
pre.save('test_data' + '_' + str(n_words))
if args.s_model:
name = f'word_transformer_{n_words}.sav'
pre.save_model(text_transformer, name=name)