Divergence Checking

If a training divergence occurs during training, the diverged output node produces network predictions approaching all 0's or 1's (normalized). This check scans the network output node predictions for this behavior and indicates any suspect nodes. Any indicated output node should be plotted to verify the diverged condition.