CODEWITHSUNDEEP
pythontensorflowinstallationpython-wheelincompatibility
roman
roman

Reputation: 1370

TensorFlow 1.9.0 and Python 3.6.5

I'm trying to upgrade to Tensorflow 1.9 within a conda environment (Ubuntu 16.04). I am using python 3.6.5. When I try this:

source activate myenv

sudo -H pip install --ignore-installed --upgrade https://storage.googleapis.com/tensorflow/linux/cpu/tensorflow-1.9.0rc0-cp36-cp36m-linux_x86_64.whl

I get the error:

tensorflow-1.9.0rc0-cp36-cp36m-linux_x86_64.whl is not a supported wheel on this platform.

Seems strange because the same thing worked fine for TF 1.8

TensorFlow seems to install fine without sudo -H but then when I try:

 python -c "import tensorflow as tf; print(tf.__version__)"

I get the following error:

from tensorflow.python.keras._impl.keras.backend import abs
ImportError: cannot import name 'abs'

I can't install from conda because it still has 1.8 when I check with:

conda install -c conda-forge tensorflow

Upvotes: 1

Views: 3824

Answers (2)

sandes
sandes

Reputation: 2267

With sudo you're installing locally. So, remove sudo -H to install over your Environment.

also, you need the python developer library locally:

  • Ubuntu: apt-get install python3-dev

  • RHEL/Fedora: dnf install python3-devel

  • Mac OS: check your environment variables or try re-installing?

  • Windows: check your environment variables or try re-installing?

Upvotes: 2

coder3101
coder3101

Reputation: 4165

Sometimes because of outdated pip also this can happen . Try this inside that environment and let me know

python -m pip install --upgrade pip

Try to check if conda has 1.9 

conda install -c conda-forge tensorflow

I figured out that this Tensorflow is a pre-release not the complete release, as a result you can upgrade it using pip directly.

You can remove your installed release and try to install this else wait for couple of weeks after that you can directly update via conda forge or pip.

Upvotes: 0

Related Questions