def eval_classif()

in utils/evaluator.py [0:0]

• 16 lines of code
• 6 McCabe index (conditional complexity)

    def eval_classif(self, scores, data_loader):
        """
        Evaluate classification.
        """
        params = self.params
        self.model.eval()

        # stats
        losses=[]
        accuracies = []
        topk = [1, 5, 10, 20, 50, 100, 200, 500]
        topk = [k for k in topk if k <= params.num_classes]

        for _, images, targets in data_loader:
            images = images.cuda()
            output = self.model(images)
            loss = F.cross_entropy(output, targets.cuda(non_blocking=True), reduction='mean')
            accuracies.append(accuracy(output.cpu(), targets, topk=tuple(topk)))
            losses.append(loss.item())

        # loss
        scores['valid_loss']=np.mean(losses)
        # accuracy
        for i_k, k in enumerate(topk):
            scores['valid_top%d_acc' % k] = np.mean([x[i_k] for x in accuracies])