jaimergp

Hm, is this vendoring `modest` and `lextor`? Looks like the repo has git submodules. If these are dynamically linked, they should be provided as separate conda-forge packages and added as...

Thanks @carterbox!

Some ideas for memory profiling [here](https://www.pluralsight.com/blog/tutorials/how-to-profile-memory-usage-in-python).

Thanks @carterbox!

Hm, line endings are getting in the way :(

I'd strongly suggest running `python -m cprofile -o profile.dat whatever_pymbar_script.py` to profile all the calls and then visualize the output with `snakeviz`. It will produce a visualization [similar to this...

`cudatoolkit` does not ship `cudnn`, but `cudnn` is [available ](https://anaconda.org/anaconda/cudnn/files )as a separate package in `defaults`. More info: https://github.com/conda-forge/conda-forge.github.io/issues/687#issuecomment-450268618

Is this going to be optional or required? `cudatoolkit` + `cudnn` amount to ~700-900MB depending on the OS and version.

Cool. We can play around with the conda packaging to provide easy options for both approaches.

This is the diff between two versions: https://github.com/choderalab/pymbar/compare/3.0.3...3.0.5 This [particular change](https://github.com/choderalab/pymbar/compare/3.0.3...3.0.5#diff-f5541a3c868c7773917580b717127325539785ff68b367eebe1427e822c878e6R383) in the default tolerance might be interesting to analyze. Otherwise, the standard approach here is to run `git bisect...