To evaluate no matter whether the resulting network at step is certainly a winning ticket, train the pruned, untrained network and examine its convergence behavior and accuracy.