def accuracy()

in domainbed/lib/misc.py [0:0]

• 23 lines of code
• 4 McCabe index (conditional complexity)

def accuracy(network, loader, weights, device):
    correct = 0
    total = 0
    weights_offset = 0

    network.eval()
    with torch.no_grad():
        for x, y in loader:
            x = x.to(device)
            y = y.to(device)
            p = network.predict(x)
            if weights is None:
                batch_weights = torch.ones(len(x))
            else:
                batch_weights = weights[weights_offset : weights_offset + len(x)]
                weights_offset += len(x)
            batch_weights = batch_weights.to(device)
            if p.size(1) == 1:
                correct += (p.gt(0).eq(y).float() * batch_weights.view(-1, 1)).sum().item()
            else:
                correct += (p.argmax(1).eq(y).float() * batch_weights).sum().item()
            total += batch_weights.sum().item()
    network.train()

    return correct / total