def main():
# Data loading and preprocessing
nltk.download('stopwords')
driver = NILCDriver(emb_type='glove', subtype='cbow', dimension=300)
driver.download_extract()
g1 = NewsG1Driver(file_name='g1_final.txt')
g1.preprocess()
g1.save_sentences()
# MCI data
mci = MCIDriver(file_name='data_set_cohmetrix_cn_trh_1.0.json')
mci_data, mci_target = mci.preprocess()
with open('data/processed_g1_final.txt') as g1_file:
data = [line for line in g1_file]
# Model training
rnn_ae = Autoencoder(encoding_dim=500, embedding_dim=300, embedding_file=driver.file_name + '.txt')
x_train, y_train, word_index, output_shape = rnn_ae.data_formatting(data)
x_test, _, _, _ = rnn_ae.data_formatting(mci_data)
rnn_ae.build(word_index, output_shape)
rnn_ae.fit(x_train, y_train, 10, 128)
rnn_ae.save_model_single_file()
x_train_clean = rnn_ae.encoder.predict(x_train)
x_train_noise = rnn_ae.noise_encoder.predict(x_train)
x_train_encoded = np.concatenate([x_train_clean, x_train_noise])
y_train_encoded = [0] * len(x_train_clean) + [1] * len(x_train_noise)
x_test_encoded = rnn_ae.encoder.predict(x_test)
y_test_encoded = mci_target
clf = LogisticRegression()
clf.fit(x_train_encoded, y_train_encoded)
print("Accuracy: %0.2f" % accuracy_score(y_test_encoded, clf.predict(x_test_encoded)))
