[ASK] Stopped the support of new Tensorflow versions >= 2.16

[miguelgfierro]

Description

We had so many problems with Tensorflow, in terms of security, breaking changes, etc, that the team decided to stop the support of Tensorflow over 2.16.

Here are some of the issues we had:

• [BUG] Upgrade to Tensorflow>2.10.1 breaks several of the DeepRec algos #2018
• [BUG] tensorflow-estimator is removed from tensorflow 2.16.1 #2072
• [BUG] NameError in ImplicitCF #2064
• TensorFlow 2.6 upgrade #1565
• Security alert with TF 2.7. Moved to 2.8.4 #2017
• [BUG] Error in the AzureML tests with CUBLAS #1883
• [BUG] Wide and Deep model raise error --- no attribute 'NanLossDuringTrainingError' #1969
• [ASK] Multinomial VAE not working #1915
• [BUG] Error in the benchmark notebook #1825
• [BUG] Errors in benchmark notebook #1560
• [BUG] AttributeError: module 'tensorflow.keras.initializers' has no attribute 'truncated_normal'\n in multinomial_vae.py #1513
• [FEATURE] TensorFlow 2.0 #953
• [FEATURE] TensorFlow 1.13 update #683

Other Comments

If a developer wants to add new code with TF dependencies, please consider this situation.

[jmarrietar]

Hi @miguelgfierro I would like to ask you about this situation, is planned to migrate from TF to Pytorch in all the Algorithms recommenders use?. I suspect that keeping an old version of TF might have future problems with other dependencies. What do you think?

[miguelgfierro]

Yes, we would like to migrate step by step the algorithms.