如何在 pytorch 中评估训练有素的 model？

Question

I have trained a model and save model using torch.save. Then after training I have loaded the model using train.load but I am getting this error

Traceback (most recent call last):
  File "/home/fsdfs.py", line 219, in <module>
    test(model, 'cuda', testloader)
  File "/home/fsdfs.py", line 201, in test
    model.eval()
AttributeError: 'collections.OrderedDict' object has no attribute 'eval'

Here is my code for test part

model = torch.load("train_5.pth")

def test(model, device, test_loader):
    model.eval()
    test_loss = 0
    correct = 0
    with torch.no_grad():
        for data, target in test_loader:
            data, target = data.to('cuda'), target.to('cuda')
            output = model(data)
            #test_loss += f.cross_entropy(output, target, reduction='sum').item() # sum up batch loss
            pred = output.argmax(1, keepdim=True) # get the index of the max log-probability 
            print(pred, target)
            correct += pred.eq(target.view_as(pred)).sum().item()

    test_loss /= len(test_loader.dataset)

    print('\nTest set: Accuracy: {}/{} ({:.0f}%)\n'.format(
         correct, len(test_loader.dataset),
        100. * correct / len(test_loader.dataset)))


test(model, 'cuda', testloader)

I have commented training part of the code in the file, so in a way this and loading the data part is all that is there in the file now.

What am I doing wrong?

Answer 1

Like @jodag has said. you probably have saved a state_dict instead of a model, which is recommended by the community as well.

This link explains the difference between two. To keep my answer self contained, I copy the snippet from the documentation. Here is the recommended way:

Save:

torch.save(model.state_dict(), PATH)

Load:

model = TheModelClass(*args, **kwargs)
model.load_state_dict(torch.load(PATH))
model.eval()

You could also save the entire model instead of saving the state_dict, if you really need to use the model the way you do.

Save:

torch.save(model, PATH)

Load:

# Model class must be defined somewhere
model = torch.load(PATH)
model.eval()