Why is bioconda recipe more constrained than setup.py?

[corneliusroemer]

I noticed that the bioconda pins are much more restrictive than the pins in setup.py

What's the reason for this? Can I relax the bioconda pins so that pathogen-embed doesn't constrain all the environments it's installed in, most notably nextstrain conda-base?

See setup.py constraints:

pathogen-embed/setup.py

Lines 26 to 37 in 617a561

	install_requires=['numpy',
	'pandas',
	"biopython",
	'scikit-learn >=1.3,<1.5',
	'umap-learn ==0.5.*',
	# Pin Numba at maximum supported version for the pinned umap-learn version.
	# For more details see:
	# https://numba.readthedocs.io/en/stable/reference/deprecation.html#deprecation-of-object-mode-fall-back-behaviour-when-using-jit
	'numba <0.59.0',
	'matplotlib',
	'hdbscan'
	],

vs. bioconda recipe constraints:

    - python >=3.8
    - numpy >=1.24.4,<2
    - pandas >=1.2.0,<2
    - biopython >=1.83,<2
    - scikit-learn >=1.3,<1.5
    - umap-learn >=0.5.0,<0.6.0
    - numba <0.59.0
    - matplotlib-base >=3,<4
    - hdbscan >=0.8.36,<0.9.0

https://github.com/bioconda/bioconda-recipes/blob/127cf1a35701ee0a808af57380d515d4865a5ef5/recipes/pathogen-embed/meta.yaml#L27-L35

[huddlej]

@corneliusroemer It's an oversight on my part that setup.py isn't as constrained as the Bioconda recipe. I'd plan to eventually follow Augur's lead in that respect.

Are there specific packages in the Bioconda recipe that are causing issues in the Nextstrain conda recipe?

[victorlin]

I noticed this discrepancy in nextstrain/public#12, specifically for pandas. PyPI (used by docker-base) allows pandas v3 to be installed, while conda pins at v1.

@huddlej would it be possible to test against pandas v2 and support it in the conda recipe?

[huddlej]

@victorlin Totally! The seasonal-flu CI workflow runs pathogen-distance, pathogen-embed, and pathogen-cluster (triggered by this config flag). If you wanted to bump pandas in test environment and run the CI workflow in that environment, you could at least quickly catch runtime errors.

I won't have time this week to check the scientific validity of the output from the later pandas, but I could check next week.