examples/openff_strategy not working

[luedsch]

Hey @JMorado,

I find your work very interesting and wanted to give it a try. Unfortunately, I could not reproduce the examples that are provided in examples/openff_strategy. I ran all three python scripts available in that directory:

Both scripts, openff_system_creation.py and run_scripts/create_run_systems.py raised an NetworkXError: The node IntraMolecularBondedRemoval is not in the digraph. The logs are attached here openff_system_creation.txt and create_run_systems.txt.

The script benzene_mm_sol.py on the other hand raised an OpenMMException: All Forces must have identical exclusions. The log is attached here: benzene_mm_sol.txt.

I also faced an issue with building the environment, as I stated here #17 - maybe the raised errors / exceptions are related to this one.

It would be very nice if you could point out what I am missing here or if you could provide working examples.

Kind regards
luedsch

[JMorado]

Thanks for your interest!

Could you try running the examples in PR #21? They should run smoothly, but let me know if you encounter any problems or if there are other examples you'd find helpful.

The script benzene_mm_sol.py on the other hand raised an OpenMMException: All Forces must have identical exclusions. The log is attached here: benzene_mm_sol.txt.

This is related to openmm/openmm#5165. I thought I had fixed this previously in fes-ml, so hopefully you won't encounter this error when running the updated examples. Were you running on the CPU or GPU? I can't remember for sure, but I think this exception only occurred on one of the platform -- probably the CPU, based on openmm/openmm#4655.

[luedsch]

Were you running on the CPU or GPU?

Thanks for the fast reply! The .txt files I posted above where all from CPU runs. Running the run_scripts/create_run_systems.py as a job on our GPUs yields a similar result, see fes-ml_14624307.txt. The job script also tried to run the analysis.py, but it couldn't import pymbar. Looking at the different environment.yaml this is easily explained, as the fes-ml env doesn't include this dependency. I report back once I have run the updated examples.

[luedsch]

So, the given examples in PR #21 already work a bit better.

The 'ml_mm_mechanical_embedding' and 'ml_mm_vacuum' both output something like this,

2026-07-08 17:01:13 INFO     fes_ml.fes       AlchemicalState(modifications={'MLInterpolation': 1.0}) iteration 497 / 500
2026-07-08 17:01:26 INFO     fes_ml.fes       AlchemicalState(modifications={'MLInterpolation': 1.0}) iteration 498 / 500
2026-07-08 17:01:38 INFO     fes_ml.fes       AlchemicalState(modifications={'MLInterpolation': 1.0}) iteration 499 / 500
2026-07-08 17:01:50 INFO     fes_ml.fes       AlchemicalState(modifications={'MLInterpolation': 1.0}) iteration 500 / 500

which I assume to be correct. There are also a bunch of files that were created, which doesn't show any (obvious) malfunctions.

ml_mm_emle on the other hand fails with TypeError: EMLECalculator.__init__() got an unexpected keyword argument 'top_file', the full output of one window is attached to this OUTPUT_0.zip. I ran all the examples with the environment from the main branch with the adjustments I described in #17. Any idea on how to fix the emle part here?

[JMorado]

Glad the examples in #21 are working better for you. I've just merged #23, which should fix the EMLE-related issue you're seeing. Just run git pull followed by pip install -e ., and let me know if the EMLE example works now.

Previously, the OpenFF strategy was passing arguments to EMLECalculator that were intended for newer experimental versions of EMLE that I've been working on. That should now be fixed.

Are you still having any issues with the MM leg when you run it on a GPU?