Closed ressy closed 1 year ago
Thanks a lot for reporting! I have opened https://github.com/bioconda/bioconda-recipes/pull/40289 for the time being until I have time to investigate this further.
BTW, you can now set Conda to use the Mamba solver with these commands:
conda install -n base conda-libmamba-solver
conda config --set solver libmamba
(I hope they make it the default soon.)
Oh wow, I had no idea conda had yielded to mamba to that extent. Thanks!
Also pretty sure it's seaborn's problem here. I'm not familiar with seaborn but for what it's worth this crashes with pandas 2.0.0 but works with 1.5.3:
import seaborn
import pandas
table = pandas.DataFrame(data = {
"x": ["A", "A", "A", "B", "B", "B"],
"y": [5, 7, 6, 2, 3, 2.5]})
seaborn.boxenplot(data=table, x="x", y="y")
Oh wow, I had no idea conda had yielded to mamba to that extent. Thanks!
I think the intention by the Mamba folks was from the beginning to get the fast solver back into Conda. Good to see it happening.
Also pretty sure it's seaborn's problem here. I'm not familiar with seaborn but for what it's worth this crashes with pandas 2.0.0 but works with 1.5.3: [...]
Ah, thanks for the reproducer. I’ll see whether I can get IgDiscover to use the most recent seaborn version. I saw some commits regarding pandas 2 compatibility.
The Bioconda package is now updated and I also changed the Conda lock file to use Pandas 2.0.0, thanks again for reporting.
pandas 2.0.0 just came out a few days ago, and it looks like there must be some compatibility-breaking changes compared with the previous major version; IgDiscover crashes with pandas 2.0.0 but works with 1.5.3. I don't think you'll catch this with the current tests because it's pinned at 1.5.3 in conda-linux-64.lock.
A minimal reproducible example with the test dataset starting from scratch (just with mamba instead of conda, since conda uses an obscene amount of RAM for dependency resolution):
I get this IgDiscover and pandas:
And it fails on the rule
plot_errorhistograms
with:If I instead force it to pandas 1 like this:
I get these packages:
...and it all works as expected. Maybe for the time being pandas should be pinned at version 1 in environment.yml? (Unless it's an easy update for new pandas; I just have no idea. Looks to me like it might be seaborn's fault based on that
mode.use_inf_as_null
thing in the traceback.)