Scalars tutorial for TensorBoard and Keras

[manivaradarajan]

This is a very quick introduction to viewing default scalar summaries when developing a Keras model in TensorFlow. It's intended to be very simple to show that virtually no extra code is required to get started using TensorBoard to see how loss is changing during training.

To some extent this repeats some of the material in the "Quick Start", but since it's very focussed I don't think the repetition is a bad thing.

[GalOshri]

Thank you for adding this!

A few general thoughts/questions:

• Would it be better to have one tutorial for the scalar plugin as opposed to multiple (e.g. include gradient tape here as well)? Or maybe it's worth breaking things down a bit, just need to make sure the titles match (gradient tape can also be used with Keras).
• Is there more information that might be valuable to include in this tutorial?
  • For example, explaining the left pane of scalars? Probably don't need to go into too much detail.
  • Additionally, explaining that you can add custom Keras metrics and they will appear in the callback? We don't need to explain everything but should give a sense of all the different capabilities here. WDYT?

[manivaradarajan]

Thanks for the feedback. I've added an example of custom scalars. I decided that using GradientTape would make it very complicated. Hopefully this complements the original material.

[GalOshri]

The learning rate example is great. Adding some comments.

[manivaradarajan]

Addressed all comments, PTAL

[manivaradarajan]

Addressed all comments, PTAL

[GalOshri]

LGTM!