`ImportError: cannot import name 'abs'` when importing TFP in Python 3 (and in Python 2)

[matthew-mcateer]

`ImportError: cannot import name 'abs'` when importing TFP in Python 3 (and in Python 2)

[Red-Eyed]

I've dealt with this problem (in conda env on Ubuntu 18.04):

pip uninstall tensorflow protobuf --yes
find $CONDA_PREFIX -name "tensorflow" | xargs -Ipkg rm -rfv pkg
pip install --ignore-installed --upgrade https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-1.9.0-cp36-cp36m-linux_x86_64.whl --no-cache-dir

[matthew-mcateer]

This appeared to be fixable with the following steps:

1. uninstalling tensorflow
2. uninstalling protobuf
3. reinstalling tensorlfow (which should come along with the correct protobuf version.

[strubell]

FWIW if you're using a local install (pip install --user) rather than conda you can replace the second command with this:
find ~/.local/ -name "tensorflow" | xargs -Ipkg rm -rfv pkg

[AramNasser]

The problem is that you updated the tensorflow but not tensorflow_gpu.

pip install --upgrade tensorflow
pip install --upgrade tensorflow-gpu

Best wishes.

[pascalwhoop]

I suspect this is not fixed by the protobuf version but rather by the reinstall of tensorflow which installs 1.8.0 I assume. I built 1.9.0 from source and see the same error.

What I can see is that the _impl folder is missing under the site-packages folder of the keras package included in tensorflow. hence, the resolution fails.

But it seems to have something to do with this commit

manually deleting the tensorflow folder inside the virtual environment and reinstalling the locally installed variant helped. I assume it has something to do with __pycache__ files.

[addisaden]

I got the problem too. Tensorflow 1.8.0 was installed as sudo and Tensorflow 1.10.0 installed as user triggered the problem.

sudo pip3 uninstall tensorflow
pip3 install tensorflow

solved it.

[Dok11]

@rodrigob did you found solution?
I have same error (windows 10, anaconda) at from keras import backend as K

  File "\Anaconda3\envs\art\lib\site-packages\tensorflow\python\keras\engine\__init__.py", line 21, in <module>
    from tensorflow.python.keras.engine.base_layer import InputSpec
  File "\Anaconda3\envs\art\lib\site-packages\tensorflow\python\keras\engine\base_layer.py", line 33, in <module>
    from tensorflow.python.keras import backend
  File "\Anaconda3\envs\art\lib\site-packages\tensorflow\python\keras\backend\__init__.py", line 22, in <module>
    from tensorflow.python.keras._impl.keras.backend import abs
ImportError: cannot import name 'abs'

[Dok11]

@Tarunyellogi I installed 1.8 instead 1.9 and its worked for me

[sraddhanjali]

Deleting the virtualenv which stored caches of tensorflow installation did the trick. Making a new env, and then installing tensorflow again likely installed the packages afresh.

[PallawiSinghal]

Hello everyone,
Please make a new environment. I was struggling the whole day with this problem. I was about to quit but for the last time I decided to create a new environment and it worked.

[assaad]

which protobuf version is supported? thanks!

[SCIUGU]

hi, I got the same problem with my program, but neither reinstalling tf nor deleting the folder worked.
I have problems with installing via web, so I download the wheel files of the modules and use pip install locally.
I'm trying to run tf 1.9.0 and I'm trying with all the latest releases of the submodules.
any hints on what to try?

[SCIUGU]

thank you very much

[guotong1988]

Thank you！

[jasperzhong]

I also encounter this problem. I have installed tensorflow 1.8.0 before and I just pip install tensorflow-gpu==1.9.0.
Then import tensorflow as tf, I saw this error.
So I uninstall tensorflow 1.8.0 and reinstall tensorflow-gpu 1.9.0.
And the error is gone! The problem is solved.

[rodrigob]

Same problem. Uninstalling & reinstalling did not help :(

[coolsgupta]

Hi,
I am using tensorflow 1.10.0
I agree with matthew-mcateer, It worked on my machine by uninstalling tensorflow and protobuf and reinstalling tensorflow.

[strubell]

@rdinse good point -- the original solution also has this problem (except for conda). I guess if you have more important things in .local then you should rm with more care :)

[Tarunyellogi]

Hey Dok11 , im getting the same error, please ping me if you got any solution for that.
https://www.linkedin.com/in/tarun-yellogi-51b41466/

[christojet]

I have the same issue that Dok11 and Tarunyellogi

[pratham16]

Deleting directories of older tensorflow installations (after uninstalling tf) and then installing tf-1.10.0 worked for me.

[yahyadogan72]

Hi,
1-) conda uninstall tensorflow-gpu and conda uninstall protobuf,
2-) Deleting directories of older tensorflow installations,
3-) conda install tensorflow-gpu==1.8
worked for me.

but, conda install tensorflow-gpu==1.10 not worked for me.

[alexcpn]

The answer from AramNaseer solved this . But when I upgraded the tensofflow and tensorflow-gpu got this error - tensorflow cannot load libcublas.so.9 (I had cuda 8 installed, now need to upgrade it and hope everything else works)

"As our installation pages list, prebuilt TF pip packages on pypi are built with CUDA 9.0 support only:
https://www.tensorflow.org/install/gpu"

tensorflow/tensorflow#20569 (comment)

[sidharthskumar]

how to create a new environment

[yustiks]

I got the problem too. Tensorflow 1.8.0 was installed as sudo and Tensorflow 1.10.0 installed as user triggered the problem.

sudo pip3 uninstall tensorflow
pip3 install tensorflow

solved it.

worked for me

[sohel5100]

This appeared to be fixable with the following steps:

1. uninstalling tensorflow
2. uninstalling protobuf
3. reinstalling tensorlfow (which should come along with the correct protobuf version.

worked for me

1. pip uninstall tensorflow
2. pip uninstall protobuf
3. pip install tensorflow==1.4.0