recommenders fix keras compatibility bug in factorized_top

The bug is summarized by the following issue comment. https://github.com/tensorflow/recommenders/issues/712#issuecomment-2041163592

May 17 '24 01:05 getumen

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

May 17 '24 01:05 google-cla[bot]

@maciejkula

May 21 '24 12:05 mat10599

+1 for this PR getting merged in. It was a blocker for me until I manually applied this fix.

Jun 21 '24 04:06 ghost

Can we please get this merged?

Jul 09 '24 11:07 takoyaro

I have the same issue. most active contributor seems to be @maciejkula, can you please tell us how to get this merged and released?

Edit: This change is not sufficient to fix the bug. We also need to change how the counter is accessed, this part in the Streamingclass throws an exception

      """Enumerates rows in each batch using a total element counter."""

      starting_counter = self._counter.read_value()

Who is running CI/CD for the tutorial code? https://www.tensorflow.org/recommenders/examples/quickstart It throws an exception due to the bug in this PR, so it appears the tensorflow recommender library might not be maintained if even the quickstart does not work

Jul 09 '24 14:07 rmminusrslash

@ZhaoyueCheng, it looks like you are the only active commiter to main in the last 6 mo. Can you give us a hand?

Jul 12 '24 20:07 czroth

I don't have permission to merge PRs actually, but since this repository is mostly running with legacy keras, would using TF_USE_LEGACY_KERAS=1 work as a workaround with older keras version? (also suggested in https://github.com/tensorflow/recommenders/issues/712#issuecomment-2112567754)

Jul 22 '24 15:07 ZhaoyueCheng

I don't have permission to merge PRs actually, but since this repository is mostly running with legacy keras, would using TF_USE_LEGACY_KERAS=1 work as a workaround with older keras version? (also suggested in #712 (comment))

Passing TF_USE_LEGACY_KERAS=1 is not doing it for me, using the legacy version of keras is making my notebook kernel crash without any logs :

[error] Disposing session as kernel process died ExitCode: undefined, Reason:

It should be related to the fact that I’m using some of keras latest features, however at this stage of my project it’s frustrating to go back and adjust the code in question.

Jul 31 '24 17:07 Az-r-ow

@rmminusrslash that issue is independant, it has been reported at https://github.com/tensorflow/recommenders/issues/722 and there is this patch https://github.com/tensorflow/recommenders/pull/723 linked to it