def __call__()

in lib/metrics.py [0:0]

• 20 lines of code
• 3 McCabe index (conditional complexity)

    def __call__(self, net, epoch, args, all_logs):
        """
        Evaluates the current state of the network without
        and with quantization and stores a checkpoint.
        """
        print("Valiation at epoch %d" % epoch)
        # also store current state of network + arguments
        res, score = evaluate(net, self.xq, self.xb, self.gt,
                              self.quantizers, self.best_key)
        all_logs[-1]['val'] = res
        if self.checkpoint_dir:
            fname = join(self.checkpoint_dir, "checkpoint.pth")
            print("storing", fname)
            torch.save({
                'state_dict': net.state_dict(),
                'epoch': epoch,
                'args': args,
                'logs': all_logs
            }, fname)
            if score > self.best_score:
                print("%s score improves (%g > %g), keeping as best"  % (
                    self.best_key, score, self.best_score))
                self.best_score = score
                shutil.copyfile(fname, fname + '.best')

        return res