Write tutorial on how to save and restore both model and the TensorFlow graph

[dustinvtran]

The process is straightforward. See, e.g., test_saver.py. It's worth having an example that users can build from as a template—either as a tutorial or written as a section in some other location of the docs.

The tutorial should provide links to the following references:

• https://www.tensorflow.org/programmers_guide/variables#saving_and_restoring
• https://www.tensorflow.org/programmers_guide/meta_graph
• https://www.tensorflow.org/programmers_guide/supervisor

[gnublet]

Can I try this one?

[dustinvtran]

Sure, feel free. Happy to provide feedback if you want to submit a PR while still unfinished.

[pswpswpsw]

actually if one just simply want the coefficient parameter. I think it is faster and much, much much more clean to use the tensorflow save model API.