Pytorch model accuracy test

Question

I'm using Pytorch to classify a series of images. The NN is defined as follows:

model = models.vgg16(pretrained=True)
model.cuda()
for param in model.parameters(): param.requires_grad = False

classifier = nn.Sequential(OrderedDict([
                           ('fc1', nn.Linear(25088, 4096)),
                           ('relu', nn.ReLU()),
                           ('fc2', nn.Linear(4096, 102)),
                           ('output', nn.LogSoftmax(dim=1))
                           ]))

model.classifier = classifier

The criterions and optimizers are as follows:

criterion = nn.NLLLoss()
optimizer = optim.Adam(model.classifier.parameters(), lr=0.001)

My validation function is as follows:

def validation(model, testloader, criterion):
    test_loss = 0
    accuracy = 0
    for images, labels in testloader:

        images.resize_(images.shape[0], 784)

        output = model.forward(images)
        test_loss += criterion(output, labels).item()

        ps = torch.exp(output)
        equality = (labels.data == ps.max(dim=1)[1])
        accuracy += equality.type(torch.FloatTensor).mean()

    return test_loss, accuracy

This is the piece of code that is throwing the following error:

RuntimeError: input has less dimensions than expected

epochs = 3
print_every = 40
steps = 0
running_loss = 0
testloader = dataloaders['test']

# change to cuda
model.to('cuda')

for e in range(epochs):
    running_loss = 0
    for ii, (inputs, labels) in enumerate(dataloaders['train']):
        steps += 1

        inputs, labels = inputs.to('cuda'), labels.to('cuda')

        optimizer.zero_grad()

        # Forward and backward passes
        outputs = model.forward(inputs)
        loss = criterion(outputs, labels)
        loss.backward()
        optimizer.step()

        running_loss += loss.item()

        if steps % print_every == 0:
            model.eval()
            with torch.no_grad():
                test_loss, accuracy = validation(model, testloader, criterion)

            print("Epoch: {}/{}.. ".format(e+1, epochs),
                  "Training Loss: {:.3f}.. ".format(running_loss/print_every),
                  "Test Loss: {:.3f}.. ".format(test_loss/len(testloader)),
                  "Test Accuracy: {:.3f}".format(accuracy/len(testloader)))

            running_loss = 0

Any help?

Answer 1

Just in case it helps someone.

If you don't have a GPU system (say you are developing on a laptop and will eventually test on a server with GPU) you can do the same using:

if torch.cuda.is_available():
        inputs =inputs.to('cuda')
    else:
        inputs = inputs.to('cuda')

Also, if you are wondering why there is a LogSoftmax, instead of Softmax that is because he is using NLLLoss as his loss function. You can read more about softmax here

Answer 2

I needed to change the validation function as follows:

def validation(model, testloader, criterion):
    test_loss = 0
    accuracy = 0

    for inputs, classes in testloader:
        inputs = inputs.to('cuda')
        output = model.forward(inputs)
        test_loss += criterion(output, labels).item()

        ps = torch.exp(output)
        equality = (labels.data == ps.max(dim=1)[1])
        accuracy += equality.type(torch.FloatTensor).mean()

    return test_loss, accuracy

inputs need to be converted to 'cuda': inputs.to('cuda')