def accuracy()

in models.py [0:0]

• 21 lines of code
• 5 McCabe index (conditional complexity)

    def accuracy(self, loader):
        nb_groups = loader.dataset.nb_groups
        nb_labels = loader.dataset.nb_labels
        corrects = torch.zeros(nb_groups * nb_labels)
        totals = torch.zeros(nb_groups * nb_labels)
        self.eval()
        with torch.no_grad():
            for i, x, y, g in loader:
                predictions = self.predict(x.cuda())
                if predictions.squeeze().ndim == 1:
                    predictions = (predictions > 0).cpu().eq(y).float()
                else:
                    predictions = predictions.argmax(1).cpu().eq(y).float()
                groups = (nb_groups * y + g)
                for gi in groups.unique():
                    corrects[gi] += predictions[groups == gi].sum()
                    totals[gi] += (groups == gi).sum()
        corrects, totals = corrects.tolist(), totals.tolist()
        self.train()
        return sum(corrects) / sum(totals),\
            [c/t for c, t in zip(corrects, totals)]