Replies: 2 comments 1 reply
-
At the time of writing this comment, numpy, scipy and matplotlib have also withdrawn support for python 3.6. I think we should also do the same and add support for python 3.9. |
Beta Was this translation helpful? Give feedback.
1 reply
-
Any updates on the subject please? |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
As of today, running
python setup.py install
on the dev branch fails. The current packages insetup.py
are as follows:For python 3.6, the failure happens because the latest scipy version 1.6.0 gets installed (tf->numpy->scipy) which requires a minimum of python 3.7. To make the installation successful, specific versions of numpy as well as scipy should be mentioned as below :
To support python 3.7+, the numpy version still needs to be mentioned. This is because the latest tf (2.4.1) needs
numpy~=1.19.2
, but tensorflow by itself doesn't always get the required numpy version (for me locally it gets 1.20rc2, which is incompatible with tf 2.4.1). So, even to support py3.7+, the following setup.py requirements need to put in:In summary, to support py3.6, an older version of scipy needs to be mentioned and we can no longer rely on tensorflow to get the compatible numpy version as well.
We would want to support py3.6, but that comes at the cost of using old versions of scipy/numpy.
Any thoughts @datitran @clennan ?
Beta Was this translation helpful? Give feedback.
All reactions