Model Examples Need Saving and Evaluation Added

[nfmcclure]

Make sure that all significant models save the model space (graph, variables, and definitions) properly and have a script that allows for (1) loading of most recent model, and (2) evaluation of new input.

[jimthompson5802]

@nfmcclure just checking to see if you want help with this. I have some examples of model save/restore in the context of transfer learning using the MNIST data set.

In my examples, I partition the MNIST digits into even and odd data sets. I train a CNN model to recognize even digits. The even digit model is saved to disk. The even digit model is restored from disk and trained on the odd digit data set with updating only the layers used for classification.

MNIST/transfer_learning_example_tf_keras performs all the steps described above.

MNIST/transfer_learning_example_tf_keras_restore_only performs only the model restore and retraining of odd digits only.

If you think the above examples are a reasonable approach, then let me know which one of your modules you'd like me to try first as test case.