outputs = net(batch_x)
loss = criterion(outputs, batch_y)
optimizer.zero_grad()
loss.backward()
optimizer.step()
running_loss = loss.item()
total_loss += running_loss
else:
break
print('Epoch %d RMSE loss: %.4f' %
(epoch, total_loss / steps_per_epoch))
torch.save(net.state_dict(), model_name + '_.pt')
else:
net.load_state_dict(torch.load(model_name + '.pt'))
net.eval()
with torch.no_grad():
yhat = []
# test_y = []
test_y = pd.read_csv('ch2_final_eval.csv')['steering_angle'].values
# composed = transforms.Compose([Rescale(image_size), Preprocess(model_name), ToTensor()])
# dataset = UdacityDataset(dataset_path, ['HMB1', 'HMB2', 'HMB4', 'HMB5','HMB6'], composed)
# steps_per_epoch = int(len(dataset) / batch_size)
# train_generator = DataLoader(dataset, batch_size=batch_size, shuffle=True, num_workers=8)
test_composed = transforms.Compose(
batch_y = batch_y.to(device)
outputs = net(batch_x)
loss = criterion(outputs, batch_y)
optimizer.zero_grad()
loss.backward()
optimizer.step()
running_loss = loss.item()
total_loss += running_loss
else:
break
print('Epoch %d RMSE loss: %.4f' %
(epoch, total_loss / steps_per_epoch))
torch.save(net.state_dict(),
'adv_training_models/' + advt_model + '.pt')
net.eval()
with torch.no_grad():
yhat = []
# test_y = []
test_y = pd.read_csv(
'ch2_final_eval.csv')['steering_angle'].values
# composed = transforms.Compose([Rescale(image_size), Preprocess(model_name), ToTensor()])
# dataset = UdacityDataset(dataset_path, ['HMB1', 'HMB2', 'HMB4', 'HMB5','HMB6'], composed)
# steps_per_epoch = int(len(dataset) / batch_size)
# train_generator = DataLoader(dataset, batch_size=batch_size, shuffle=True, num_workers=8)