Cannot install perses > 0.8.1 with python > 3.7 because of pymbar

[zhang-ivy]

If my conda environment has python > 3.8, running mamba install -c conda-forge perses will install perses 0.8.1.

If I try to install a more recent version of perses, running mamba install -c conda-forge --only-deps perses=0.9.1:

Problem: package perses-0.9.1-pyh8a188c0_0 requires pymbar 3.0.3, but none of the providers can be installed

which is because pymbar 3.0.3 requires python >=3.7,<3.8.0a0.

We are requiring the older version of pymbar (3.0.3) because of choderalab/pymbar#419.

Possible solutions:

• get a bugfix release of pymbar out asap (so that we don't have to rely on the older version of pymbar)
• update pymbar 3.0.3 to support newer versions of python.

[mikemhenry]

related:
conda-forge/perses-feedstock#11

[mikemhenry]

So we can fix this in the pinning in the next release, since our unit tests pass with pymbar 3.0.5 but there could be issues. @zhang-ivy Can you make an env with perses 0.9.1 or 0.9.2 and pymbar 3.0.5 and run your workflow to make sure there will be no major issues?
You can create the env by using the yaml here https://github.com/choderalab/perses/blob/master/devtools/conda-envs/test_env.yaml then doing a pip install .

[zhang-ivy]

Can you make an env with perses 0.9.1 or 0.9.2 and pymbar 3.0.5

I already tried this. pymbar 3.0.5 is problematic for analyzing repex simulations (I'm seeing the same issue as reported by the pymbar issue that I linked above). I think the only solutions here are the ones I mentioned above. We don't want to make pymbar 3.0.5 a dependency.

[jchodera]

Can you remind me what the issue is with analyzing repex trajectories with the latest pymbar? I can dive in and help fix that on our end.

[zhang-ivy]

See details at choderalab/pymbar#419.
To quote Michael Shirts: "for the majority of 3.0.3 -> 3.0.5 issues, the default solver switched (which is good - the "adaptive" solver is much more consistent), but the default maximum number of iterations is too low (250), and thus MBAR doesn't go to completion."

Josh fass's fix here fixed the issue for me: choderalab/pymbar#425 -- I installed from master to get this fix. I think we might just need a new release now.

[ijpulidos]

@mikemhenry will be releasing a new bugfix for pymbar that should solve this issue.

[mikemhenry]

fixed with conda-forge/pymbar-feedstock#19