batch_loss = np.mean(losses)
epoch_losses.append(batch_loss)