[FEATURE] TensorFlow 1.13 update

[loomlike]

Description

Update TensorFlow package version

Other Comments

Require to refactor deprecated functions and changed APIs

[miguelgfierro]

We should also discuss about TF 2.0, it is still in alpha https://github.com/tensorflow/tensorflow/releases but it will have some interesting new features (and probably breaking changes)

[loomlike]

@miguelgfierro Do we want to skip 1.13 and directly move to 2.0?

We don't suffer any bugs from 1.12 and thus not sure moving from 1.12 to 1.13 would give any benefit to us. TF2.0, on the other hand, seems has attractive features as you said, e.g. NCF reference model, distributed training API and cleaner APIs.

[miguelgfierro]

yeah that makes sense. I wouldn't rush unless we see a big benefit, TF 2.0 could be buggy

[gramhagen]

yeah, it would be nice to have a branch where tf 2.0 example was built out so we could compare, ncf might be a good starting point.

[loomlike]

Btw, tested 1.13 with unit/test_tf_utils.py and smoke/test_notebooks_gpu.py and all tests got passed.

[miguelgfierro]

Cool, we can upgrade them

[loomlike]

@miguelgfierro TF already released 1.14 lol... and I discussed with @gramhagen about the version. We think it makes sense to just stick with the current version we have for now (1.12) since nothing has been broken nor are buggy. Also not very exciting features in .13 and .14 for us.

So we will wait until 2.0 officially released. Is that make sense?

[miguelgfierro]

So we will wait until 2.0 officially released. Is that make sense?

yeah that makes sense

[miguelgfierro]

Closing this, feel free to reopen if needed