Use Zero-One loss function as a model-metric for keras neural networks

Question

I know that zero-one loss is rather a simple metric for calculating the accuracy of a model estimator. However, even simple, for multi-label classi