Tensorflow speed of individual train/test

Instead of having individuals with their own models, look into possible speed up with individuals instead only containing parameters of the model.
Most of the time associated with the training is in the set-up of the network, not the training or testing. It is possible that there is some extra load time associated with TF that we can avoid.

No files, tests, or entry points are named. Start by locating the TensorFlow network setup and the individual training/testing flow, then measure setup time against training and testing time; done means a supported design for avoiding repeated setup and evidence of its speed impact.

Written by the indexing model from the issue text.