delimiter=',')
# Generate abnormal data
abnormal_data = adg.generate_abnormal_data_from_raw_data(overwrite_data=False,
generate_graph=True,
show_graph=False)
abnormal_data = abnormal_data[:, 1:]
abnormal_ratios = []
for i in range(results_array.shape[0]):
print('======================== Iteration {} ========================'.
format(i))
ae_tests_score, ae_tests_scores = aeu.test_trained_ae_model(
test_data=abnormal_data,
model_dir_path=trained_model_path,
iteration_number=i)
# Refer to the deep_ae_summary_results.csv
threshold_value = results_array['threshold_value'][i]
abnormal_ratio = sum(
[score > threshold_value
for score in ae_tests_scores]) / float(len(ae_tests_scores))
abnormal_ratios.append(abnormal_ratio)
output_string = '\n{0:.2f}% of abnormal samples are detected as abnormal.'.format(
abnormal_ratio * 100.0)
print(output_string)
print('==============================================================')
hidden_units=best_layer_type)
autoencoder.train(train_data=train_data,
save_model=True,
test_saved_model=True,
model_dir_path=saved_ae_network_path,
iteration_number=i,
print_and_plot_history=True,
show_plots=False)
# Get the scores
ae_train_score = autoencoder.global_mse
ae_train_scores = autoencoder.mse_per_sample
ae_validate_score, ae_validate_scores = aeu.test_trained_ae_model(
test_data=validation_data,
model_dir_path=saved_ae_network_path,
iteration_number=i)
ae_tests_score, ae_tests_scores = aeu.test_trained_ae_model(
test_data=abnormal_data,
model_dir_path=saved_ae_network_path,
iteration_number=i)
# Format the test scores: [Iteration, Test_sample_No, Score]
if i == 0:
with open(os.path.join(data.dir_name, test_score_filename),
'wb') as score_file:
score_file.write(b'iteration,test_sample_no,score\n')
rows_number = len(ae_tests_scores)
test_scores_array = np.hstack((np.full(