Currently minibatch is already implemented for the training set in the atflow.dataset.Dataset. However, when running and evaluating a complex network, you would want to evaluate validation/test set in batches too. Add minibatch_validation and minibatch_test.