Closed mortonjt closed 2 years ago
Can you upgrade the numpy package in the first environment to 1.13 or newer? There were some issues with the Tensorflow 1.5.1 wheels requiring a newer NumPy version than the requirements expressed, cf tensorflow/tensorflow#16488, which were copied into the conda package. Upgrading NumPy usually fixes the issue.
The later issue I believe is caused by the minimum version of macOS being 10.12 for Tensorflow 1.10.0.
Makes sense. Thanks!
At the moment, qiime2 is fixed at numpy=1.12 -- but we are currently in discussion about bumping up the numpy version there.
That being said, I was still having issues with the standalone TF installation with conda-forge (even without specifying the numpy version or other qiime2 dependencies). The anaconda version seems to be working though via
conda create -n tf-test2 tensorflow -c anaconda
source activate tf-test2
python -c "import tensorflow as tf; print(tf.__version__)"
'1.12.0'
The anaconda package (but please use defaults
not anaconda
) is built from source and will work with macOS 10.11 and above. The conda-forge
package is a repackage of the wheels provided by Google, which appear to support macOS 10.12 and above. Building Tensorflow from source, see #6, could solve this problem but is quite challenging.
I believe this has been addressed in recent versions.
i would suggest adding a tensorflow pinning >=2.6
. should help with these builds.
Yeah, this is an ancient issue, let's close it (but please feel free to open a new one if there are any remaining problems!)
Issue: Tensorflow seems to be breaking across multiple versions.
I'm trying to install tensorflow with another package, qiime2. That install procedure is as follows
Environment (
conda list
):Details about
conda
and system (conda info
):I'm trying to conda install tensorflow=1.5 and get the following error when importing in python
Has anyone else noticed these sorts of errors? Not entirely sure why this is the case, especially since numpy=1.12 should be compatible.
I then tried to see if I could install tensorflow by itself, and still had issues.
Below are the environment details for the 2nd attempt
Environment (
conda list
):Details about
conda
and system (conda info
):Any insights will be greatly appreciated! And this resource is extremely useful, particularly for tools that heavily rely on the conda build!