Probability: module 'tensorflow._api.v1.compat' has no attribute 'v2'

[Mekkhaoula]

Problem definition:
In

probability/tensorflow_probability/python/optimizer/sgld.py

I receive the following error when running class StochasticGradientLangevinDynamics(tf.compat.v2.optimizers.Optimizer): : module 'tensorflow._api.v1.compat' has no attribute 'v2'

Install process:

I installed tensorflow and tensorflow-probability, both through Conda cmds, in one of my existing Conda environments.
I also should specify that I have for each environment a specific IPython kernel, so that shouldn't be a problem with elsewhere existing TF installations.

Conda commands were:
conda install -n myenvname tensorflow
conda install -c conda-forge tensorflow-probability

Versions:

tensorflow 1.13.1
tensorflow-probability 0.5.0

Any help or assistance is much appreciated, thanks!

[csuter]

Hi there, we (TFP) don't maintain the conda build and I'm not sure how it works; I seem to recall it does not adhere to our prescribed installation procedure.

The most recent stable version of TFP is tensorflow-probability==0.6.0, which is tested and stable against tensorflow==1.13.1

Can you somehow force 0.6.0 to be installed instead of 0.5.0?

[truittll]

I got this error as well. I updated tensor flow-probability and tensor flow using pip. Now I am getting a new error "No module named tensorflow.compat.v2". I did use the prescribed installation procedure

[brianwa84]

Do you have a stack trace for the import error? Did you install with pip? Can you share your pip list?

[haolsun]

Maybe you can use the code in Branch r0.5. The master is different from the Branch. I think there are some compatibility issues for the TF 1.13. TF 2.0 is suitable for TF-probability 0.6. @mkcliona

[csuter]

Thanks, @haolsun, but this is not quite right (maybe there was a typo?).

TFP 0.6 is compatible with TF 1.13 -- it will not work with TF 2.0 (alpha or nightly)

tfp-nightly will work with tf-nIghtly and tf-nightly-2.0-preview, and might work with TF 2.0 alpha, although we don't have continuous integration testing for this combination (it did work several weeks ago, but tfp-nightly has continued progressing while TF 2.0a hasn't been updated AFAIK).

[GauravRaturi]

Still getting the module 'TensorFlow._api.v1.compat' has no attribute 'v2' error. I have TensorFlow version 1.13.1 and TensorFlow probability 0.9.0. Kindly give the solution.

[brianwa84]

Can you upgrade your TF build? TFP 0.9 is compatible with TF 2.1

Brian Patton | Software Engineer | [email protected]

On Tue, Jan 28, 2020 at 12:31 AM GauravRaturi [e-mail]
wrote:

Still getting the module 'TensorFlow._api.v1.compat' has no attribute 'v2'
error. I have TensorFlow version 1.13.1 and TensorFlow probability 0.9.0.
Kindly give the solution.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/tensorflow/probability/issues/354?email_source=notifications&email_token=AFJFSIZUOI4MADWSLA7SDQ3Q767MHA5CNFSM4HED2R52YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEKCCYZQ#issuecomment-579087462,
or unsubscribe
https://github.com/notifications/unsubscribe-auth/AFJFSI5M3AKS4JGMSBMQBS3Q767MHANCNFSM4HED2R5Q
.

[brianwa84]

More info on TFP releases & compatibility is here
https://github.com/tensorflow/probability/releases

Brian Patton | Software Engineer | [email protected]

On Tue, Jan 28, 2020 at 10:50 AM Brian Patton 🚀 [e-mail] wrote:

Can you upgrade your TF build? TFP 0.9 is compatible with TF 2.1

Brian Patton | Software Engineer | [email protected]

On Tue, Jan 28, 2020 at 12:31 AM GauravRaturi [e-mail]
wrote:

Still getting the module 'TensorFlow._api.v1.compat' has no attribute
'v2' error. I have TensorFlow version 1.13.1 and TensorFlow probability
0.9.0. Kindly give the solution.

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
https://github.com/tensorflow/probability/issues/354?email_source=notifications&email_token=AFJFSIZUOI4MADWSLA7SDQ3Q767MHA5CNFSM4HED2R52YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEKCCYZQ#issuecomment-579087462,
or unsubscribe
https://github.com/notifications/unsubscribe-auth/AFJFSI5M3AKS4JGMSBMQBS3Q767MHANCNFSM4HED2R5Q
.

[wuyexingkong]

hi, i had the same problem,Has the problem been solved?

[GauravRaturi]

hi, i had the same problem,Has the problem been solved?
Yes, the problem has solved I updated by TensorFlow to 1.15.0. Let me know if the problem persists.

[GauravRaturi]

Update the CUDA and CuDNN to the suitable version if you are using GPU version. I have used TensorFlow-cpu==1.15.0.

[wuyexingkong]

i sovled the problem by the way you told me.

[fkhjoy]

Someone solved by downloading this version of models since we are using TF 1.13

[HajerAa]

@fkhjoy after downloading it, where should i put the models-1.13.0 folder to solve this please???

[fkhjoy]

@fkhjoy after downloading it, where should i put the models-1.13.0 folder to solve this please???

Where your tensorflow is installed
In my case it's ..\Lib\site-packages\tensorflow\
In this directory my models is located. Just put it there.

[wangshuo2020]

@fkhjoy after downloading it, where should i put the models-1.13.0 folder to solve this please???

Where your tensorflow is installed
In my case it's ..\Lib\site-packages\tensorflow
In this directory my models is located. Just put it there.

why some files are corrupted when extracting files? And I just need put this folder in the directionary ..\Lib\site-packages\tensorflow? It didn't seem to work.

[fkhjoy]

@fkhjoy after downloading it, where should i put the models-1.13.0 folder to solve this please???

Where your tensorflow is installed
In my case it's ..\Lib\site-packages\tensorflow
In this directory my models is located. Just put it there.

why some files are corrupted when extracting files? And I just need put this folder in the directionary ..\Lib\site-packages\tensorflow? It didn't seem to work.

Extract in different directory and copy-paste it to the desired directory. There is no reason to be corrupted. Try from this link

[wangshuo2020]

@fkhjoy after downloading it, where should i put the models-1.13.0 folder to solve this please???

Where your tensorflow is installed
In my case it's ..\Lib\site-packages\tensorflow
In this directory my models is located. Just put it there.

why some files are corrupted when extracting files? And I just need put this folder in the directionary ..\Lib\site-packages\tensorflow? It didn't seem to work.

Extract in different directory and copy-paste it to the desired directory. There is no reason to be corrupted. Try from this link

I downloaded the 1.13.0 model again. And it works well. May be some error occurred during last download.