3

I have two Python 3.7.11 environments managed by Anaconda, and I'm trying to sort out reproducibility issues between the two. I've pinpointed those issues to numpy versions, and thus I've cloned my base environment to gradually upgrade numpy to find where they stem from. While comparing versions between both environments, I've realized that the first one has both numpy and numpy-base listed, which apparently happens. However, numpy and numpy-base have different versions. If I run conda list numpy, I get the following :

# Name                    Version                   Build  Channel
numpy                     1.21.2                   pypi_0    pypi
numpy-base                1.16.5           py37hde5b4d6_0

While the same command within my second environment, in which I've upgraded and downgraded numpy several times, returns :

# Name                    Version                   Build  Channel
numpy                     1.16.5                   pypi_0    pypi
numpy-base                1.16.5           py37hde5b4d6_0

Printing numpy.__version__ in both environments returns 1.16.5.

However, pip show numpy returns 1.21.2 for the first environment and 1.16.5 for the second, as pip show numpy-base apparently doesn't work.

From what I've read in the answer I've linked above, it seems that numpy-base is the version actually doing calculations, but as my results aren't reproducible between the two environments, could it cause issues ?

BptGrm
  • 39
  • 1
  • 3
    I feel like this is an artifact from mixing pypi and conda packages. You can probably save yourself a lot of sanity when installing numpy from conda repositories exclusively :-) – cel Dec 16 '21 at 13:39
  • @cel This does seem to be an issue, I was at least able to reproduce the layout of the first environment by installing 1.21.2 with pypi and then 1.16.5 with conda. Now I just have to find a way to get reproducible results with a clean install ! – BptGrm Dec 16 '21 at 14:11

0 Answers0