# set grad_enabled setting
torch.set_grad_enabled(True)
for param in model.features.parameters():
param.requires_grad = False
# set loss function and optimizer
criterion = nn.NLLLoss()
optimizer = optim.SGD(model.classifier.parameters(), lr=args["learning_rate"])
# tramsfer the model and train
model = model.to(device)
model = model_functions.train_model(model, args["epochs"], device, optimizer,
criterion, train_loader, valid_loader)
# test the model with training dataset
model_functions.test_model(model, device, optimizer, criterion, test_loader)
# make the checpoint dictionary
checkpoint = {
'epochs': args["epochs"],
'arch': args["arch"],
'classifier_dict': classifier_dict,
'cumulative_epochs': args["epochs"],
'state_dict': model.state_dict(),
'class_to_idx': train_data.class_to_idx,
'optimizer_state_dict': optimizer.state_dict()
}
# save the checkpoint in the file
torch.save(checkpoint, args["save_path"])
model = train_model(model=model,
target_size=target_size,
batch_size=args.batch_size,
list_of_categories=categories,
train_epoch=args.epochs,
steps_per_epoch=args.steps_per_epoch,
validation_split=validation_split,
train_folder=train_folder,
checkpoint_path=checkpoint_path,
tensorboard_path=tensorboard_path,
as_gray=as_gray)
# else just eval
test_model(model=model,
target_size=target_size,
batch_size=args.batch_size,
list_of_categories=categories,
as_gray=as_gray,
test_folder=test_folder,
result_folder=result_folder,
image_type=image_type)
else:
if not args.eval: # Training
model = train_model(model=model,
target_size=target_size,
batch_size=args.batch_size,
list_of_categories=categories,
train_epoch=args.epochs,
steps_per_epoch=args.steps_per_epoch,
validation_split=validation_split,
train_folder=train_folder,
checkpoint_path=checkpoint_path,
tensorboard_path=tensorboard_path,
val_folder=val_folder,
checkpoint_path=checkpoint_path,
tensorboard_path=tensorboard_path,
as_gray=as_gray,
data_gen_args=data_gen_args,
early_stop_number=args.early_stop,
image_type=image_type)
print("Done training")
model.load_weights(str(checkpoint_path))
# else just eval
test_model(model=model,
target_size=target_size,
batch_size=args.batch_size,
list_of_categories=categories,
as_gray=as_gray,
test_folder=test_folder,
result_folder=result_folder,
image_type=image_type,
save_images=True,
max_images_saved=0)
else:
print(tf.device)
if not args.eval: # Training
model = train_model(model=model,
target_size=target_size,
batch_size=args.batch_size,
list_of_categories=categories,
train_epoch=args.epochs,
steps_per_epoch=args.steps_per_epoch,
validation_split=validation_split,
train_folder=train_folder,