A straightforward line search approach on the expected empirical loss for stochastic deep learning problems