Does this package work for tensorflow 1.15?

[Jingnan-Jia]

I found that the last date of commit is 2 years ago, so maybe this package is not applied in tensorflow1.15? Does anyone could make sure it?? It can not work in my codes with tf1.15. I need to know if it is because of the version of tensorflow or my own codes.

[wen8411]

1.15 could be the last version on which this works. The package uses tf.contrib.graph_editor while starting from 2.0 tf.contrib is removed. Is there plan to have this work in TF2?

[fernandocamargoai]

@wen8411 but do you know if it works in eager mode of tf 1.15? And the contrib module was basically moved to tf-addons, so it would just need to be refactored here.

[wen8411]

Hi, It worked for me in 1.15 but I did not try eager mode. I searched tf-addons in 2.0 but couldn’t even import tfa. Checked tfa repo but didn’t find graph_editor. There are posts discussing this breaks in 2.0.

…

On Fri, Feb 7, 2020 at 8:18 AM Fernando Camargo ***@***.***> wrote: @wen8411 <https://github.com/wen8411> but do you know if it works in eager mode of tf 1.15? And the contrib module was basically moved to tf-addons, so it would just need to be refactored here. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#49?email_source=notifications&email_token=AGYNNTVJ34YSAUUPT5LZKMDRBVUSNA5CNFSM4KCAWFL2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOELDBOUA#issuecomment-583407440>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGYNNTQXAJFBMPFBHDZ36HDRBVUSNANCNFSM4KCAWFLQ> .

[golden0080]

@wen8411 is it possible to describe what kind of models worked with this repo?

I'm adopting this in my model, but got some errors like:

ValueError: Operation 'training/Adam/gradients/header/conv_4/batch_normalization_63/cond/ReadVariableOp_2/Switch' has no attr named '_XlaCompile'.

[golden0080]

@wen8411 And to your point of graph_editor in TF 2.x, it's removed from TF source code basically. They were sunsetting contrib modules and that included graph_editor.

[nouiz]

The errors give me the impression that this repo doesn't support XLA. Try disabling XLA.

[golden0080]

Thank you @nouiz for the suggestion. After disabling XLA, I was able to get different errors:

ValueError: Tensor conversion requested dtype float32 for Tensor with dtype resource: <tf.Tensor 'training/Adam/gradients/gradients_1/training/Adam/gradients/header/conv_4/batch_normalization_63/cond/ReadVariableOp_2/Switch_grad/Switch_1:1' shape=() dtype=resource>
TypeError: Tensors in list passed to 'inputs' of 'Merge' Op have types [float32, resource] that don't all match.

I know this may be related to my model architecture, any ideas why this is happening?