Have TQDMProgressBar work in evaluate mode as well

[zaccharieramzi]

Describe the feature and the current behavior/state.

Right now, the TQDMProgressBar callback can only be used when training a model. When evaluating nothing happens. And even if I try to "cheat" by manually setting the on_test_batch_(begin|end) methods, it fails with an error about mode.

Relevant information

• Are you willing to contribute it (yes/no): don't know
• Are you willing to maintain it going forward? (yes/no): no
• Is there a relevant academic paper? (if so, where): no
• Is there already an implementation in another framework? (if so, where): no
• Was it part of tf.contrib? (if so, where): no

Which API type would this fall under (layer, metric, optimizer, etc.) callbacks

Who will benefit with this feature? everyone using tqdm to evaluate in a notebook: I want something fancier than the default evaluate progressbar

Any other info. It failed in both tf2.2 and tf2.1

You can find the colab gist here.

[gabrieldemarmiesse]

I don't think it would be hard to do because keras exposes methods for that: e.g. https://www.tensorflow.org/api_docs/python/tf/keras/callbacks/Callback#on_test_batch_begin . Contributions welcome!

[ashutosh1919]

@gabrieldemarmiesse , please review PR #38447

[gabrieldemarmiesse]

As a small note, I believe the TQDM callback should have a cleanup as we removed the "samples" mode and a refactoring would be welcome before a PR to add this feature. There is definitly some dead code/renaming to do.

[ashutosh1919]

Yep. Got it. Let me do that too.

[ashutosh1919]

@gabrieldemarmiesse , I have created PR for adding methods on_test_batch_begin() and on_test_batch_end(). I have also looked at removing unused code related to samples mode. I have found one clue here. But since, I am not aware about past implementation and changes, I couldn't do on my own. Please give me hints or point me the direction where I can find one.

[shun-lin]

ops, sorry I have forgot about evaluate() when I wrote it and thanks @ashutosh1919 so much for the contribution!

[ashutosh1919]

@zaccharieramzi and @gabrieldemarmiesse , please review PR. Creates as below.