sys.exit()
print(net)
optimizer = optim.Adam(net.params, lr=learning_rate)
criterion = nn.CrossEntropyLoss()
'''Convert to cuda if available'''
if torch.cuda.is_available() and cuda:
print("CUDA is available, training on GPU")
print("Number of available devices: {}".format(torch.cuda.device_count()))
print("Using device: {}".format(cuda_device))
torch.cuda.device(args.device)
net.cuda()
criterion = criterion.cuda()
else:
print("CUDA is NOT available, training on CPU")
'''Train and evaluate model'''
for i in range(1, num_epochs + 1):
train(i, net, trainloader, criterion, optimizer, cuda, batch_size)
print("Results on training set")
evaluate(i, net, trainloader, criterion, cuda, batch_size)
print("Results on validation set")
evaluate(i, net, testloader, criterion, cuda, batch_size)
net.cpu()
torch.save(net.state_dict(), model_path + name + "_" + str(i) + ".pth")
if torch.cuda.is_available() and cuda:
net.cuda()
if i % 2 == 0:
'''Decay learning rate'''
learning_rate = learning_rate * 0.95
optimizer = optim.Adam(net.params, lr=learning_rate)
print(f"Learning rate: {learning_rate}")
