net.load() after change in network layers is bad

[dsblank]

1. create network
2. net.save()
3. change network (say, hidden layer size)
4. net.load()

No error message, until you train, etc. conx net != keras net