But the training network has a different number of hidden nodes (7 instead of 5) and a different random seed for weight initialization, because using the same number of hidden nodes and seeds would ...