def create_train_test_valid_folders(dataset_train_test_valid_directory,
dataset_folders_to_create,
dataset_classes):
for dirname in dataset_folders_to_create:
Utils.create_folder_if_not_exists(
os.path.join(dataset_train_test_valid_directory, dirname))
for class_name in dataset_classes:
Utils.create_folder_if_not_exists(
os.path.join(dataset_train_test_valid_directory, dirname,
class_name))
def train_models(dataset_dataloader, dataset_test_dir_name,
dataset_train_dir_name, dataset_valid_dir_name,
training_cnn_models_to_train, training_criterion,
training_dropout, training_epochs_early_stopping,
training_epochs_count, training_feature_extract,
training_learning_rate, training_lr_gamma,
training_lr_step_size, training_model_output_directory,
training_momentum, training_optimizer,
training_save_best_model_enabled, training_scheduler,
training_use_gpu, training_use_early_stopping,
training_use_pretrained_models, training_use_softmax,
training_weight_decay, signals):
for i, model_id in enumerate(training_cnn_models_to_train):
signals['label_training_model_name_text_changed'].emit(model_id)
signals['console_append'].emit('----------\n' + model_id +
'\n----------')
model, accuracy_loss_history = ModelTraining.train_model(
model_id=model_id,
dataset_dataloader=dataset_dataloader,
dataset_test_dir_name=dataset_test_dir_name,
dataset_train_dir_name=dataset_train_dir_name,
dataset_valid_dir_name=dataset_valid_dir_name,
training_criterion=training_criterion,
training_dropout=training_dropout,
training_epochs_count=training_epochs_count,
training_epochs_early_stopping=training_epochs_early_stopping,
training_use_gpu=training_use_gpu,
training_use_early_stopping=training_use_early_stopping,
training_feature_extract=training_feature_extract,
training_learning_rate=training_learning_rate,
training_lr_gamma=training_lr_gamma,
training_lr_step_size=training_lr_step_size,
training_momentum=training_momentum,
training_optimizer=training_optimizer,
training_scheduler=training_scheduler,
training_use_pretrained_models=training_use_pretrained_models,
training_use_softmax=training_use_softmax,
signals=signals,
training_weight_decay=training_weight_decay)
if training_save_best_model_enabled:
path = Utils.create_folder_if_not_exists(
training_model_output_directory)
ModelIO.save(
model=model,
model_id=model_id,
output_dir=path,
classes=dataset_dataloader[dataset_train_dir_name].dataset.
classes,
feature_extract=training_feature_extract,
use_pretrained=training_use_pretrained_models,
training_dropout=training_dropout,
training_use_softmax=training_use_softmax)
signals['plot_train_valid_acc_loss_graph'].emit(
accuracy_loss_history, model_id)
signals['progressBar_training_set_value_changed'].emit(
(i + 1) / len(training_cnn_models_to_train) * 100)
def save(model, output_dir, model_id, classes, feature_extract,
use_pretrained, training_dropout, training_use_softmax):
output_dir = Utils.create_folder_if_not_exists(output_dir)
filename = model_id + DELIMITER + datetime.now().strftime(
"%Y%m%d_%H%M%S") + '.pth'
torch.save(model, os.path.join(output_dir, filename))
torch.save(
{
'classes': classes,
'feature_extract': feature_extract,
'model_id': model_id,
'model_state_dict': model.state_dict(),
'use_pretrained': use_pretrained,
'training_dropout': training_dropout,
'training_use_softmax': training_use_softmax
}, os.path.join(output_dir, filename))
def train_valid_acc_loss_graph(history, model_name, save,
training_evaluation_directory):
history = pd.DataFrame(
history,
columns=['train_acc', 'train_loss', 'valid_acc', 'valid_loss'])
fig = plt.figure(figsize=(15, 6))
locator = matplotlib.ticker.MultipleLocator(1)
fig.suptitle(model_name)
subfig = fig.add_subplot(122)
fig.gca().xaxis.set_major_locator(locator)
subfig.plot(history['train_acc'], label="Training")
subfig.plot(history['valid_acc'], label="Validation")
subfig.set_title('Model Accuracy')
subfig.set_xlabel('Epoch')
subfig.set_ylabel('Percentage')
subfig.legend(loc='upper left')
subfig = fig.add_subplot(121)
fig.gca().xaxis.set_major_locator(locator)
subfig.plot(history['train_loss'], label="Training")
subfig.plot(history['valid_loss'], label="Validation")
subfig.set_title('Model Loss')
subfig.set_xlabel('Epoch')
subfig.set_ylabel('Loss')
subfig.legend(loc='upper right')
plt.show()
if save:
path = os.path.join(training_evaluation_directory, model_name)
path = Utils.create_folder_if_not_exists(path)
plt.savefig(os.path.join(
path, 'valid_acc_loss_graph_' +
datetime.now().strftime("%Y%m%d_%H%M%S") + '.png'),
bbox_inches='tight')
