# training time
history = attr.model.fit_generator(
attr.train_generator,
steps_per_epoch=attr.steps_train,
epochs=attr.epochs,
validation_data=attr.validation_generator,
validation_steps=attr.steps_valid,
use_multiprocessing=True,
workers=multiprocessing.cpu_count() - 1,
callbacks=callbacks)
# plot loss and accuracy
plot_train_stats(history, attr.curr_basename + '-training_loss.png', attr.curr_basename + '-training_accuracy.png')
# make sure that the best weights are loaded (even if restore_best_weights is already true)
attr.model.load_weights(filepath=attr.curr_basename + "-ckweights.h5")
# save model with weights for later reuse
save_model(attr)
# delete ckweights to save space - model file already has the best weights
os.remove(attr.curr_basename + "-ckweights.h5")
# create confusion matrix and report with accuracy, precision, recall, f-score
write_summary_txt(attr, NETWORK_FORMAT, IMAGE_FORMAT, ['negative', 'positive'], time_callback, callbacks[1].stopped_epoch)
K.clear_session()
copy_to_s3(attr)
# os.system("sudo poweroff")
attr.model.save(attr.summ_basename + '-weights.h5')
# Plot train stats
plot_train_stats(history, attr.summ_basename + '-training_loss.png', attr.summ_basename + '-training_accuracy.png')
# Get the filenames from the generator
fnames = attr.test_generator.filenames
# Get the ground truth from generator
ground_truth = attr.test_generator.classes
# Get the label to class mapping from the generator
label2index = attr.test_generator.class_indices
# Getting the mapping from class index to class label
idx2label = dict((v,k) for k,v in label2index.items())
# Get the predictions from the model using the generator
predictions = attr.model.predict_generator(attr.test_generator, steps=attr.steps_test, verbose=1)
predicted_classes = np.argmax(predictions, axis=1)
errors = np.where(predicted_classes != ground_truth)[0]
res="No of errors = {}/{}".format(len(errors), attr.test_generator.samples)
with open(attr.summ_basename + "-predicts.txt", "a") as f:
f.write(res)
print(res)
f.close()
write_summary_txt(attr, "Unimodal", "2D", ['negative', 'positive'])
copy_to_s3(attr)