Closing the Gap Between Theory and Practice During Alternating Optimization for GANs