Touch-ups to the environment.yml file

[jonwzheng]

Hi folks, thanks for all your work on this great software package.

I was checking out the environment.yml file after one of my labmates was experiencing issues with solving the conda environment.

A few things that caught my attention:

• defaults and anaconda channels: Large organizations need to pay for the defaults channel and so it may be good to move away from defaults to conda-forge. I was able to successfully install the environment by commenting out the defaults channel. I think you can also switch many of the anaconda packages to conda-forge. (See https://www.anaconda.com/blog/is-conda-free)
• We moved away from some of the rmg channel packages in RMG-Py, to conda-forge (see https://github.com/ReactionMechanismGenerator/RMG-Py/blob/main/environment.yml) to help with upkeep. Are there any plans to do this? The ones I could immediately find:
  • gprof2dot
  • muq (https://anaconda.org/conda-forge/muq)
  • numdifftools
  • quantities
  • rdkit
• Any plans to update to ensure Python 3.9 or higher? 3.7 is end-of-life, and upgrading may help with continuity with the other conda packages

[calvinp0]

Hey mate,

Yes! We are intending to move from 3.7 to 3.9 soon. On top of that, also updating the packages and their channels. We are just working on decoupling parts of our code from RMG-Py and hopefully will have that done soon. It's our current focus

[jonwzheng]

Thanks Calvin, looking forward to it!
in the meantime what is the recommended install procedure?
The instructions on https://reactionmechanismgenerator.github.io/ARC/advanced.html say to install the ARC env and then install RMG, but RMG doesn't support 3.7 anymore.
Is there a docker config?

[calvinp0]

Hey @jonwzheng
We do have a Docker currently, hosted on my account on Dockerhub although I don't think the ARC version on there has the latest commits so may have to git pull if utilizing the docker.

You can find it here

I have tried to do an updated version of our Docker of ARC but pulling back to earlier commits of RMG-Py. The issue though is RMS - I have been attempting to find a version that doesn't use PythonCall, but still uses PyCall, but have not had any success thus far

[jonwzheng]

Thanks!

Let us know if there's something we could do to help. We moved away from PyCall to PythonCall (see RMG-Py discussion ReactionMechanismGenerator/RMG-Py#2640, ReactionMechanismGenerator/RMG-Py#2741; RMS PR ReactionMechanismGenerator/ReactionMechanismSimulator.jl#268 )

[calvinp0]

Yeah, and that is kinda the issue with the versioning/tags on RMS - trying to find one that is still the old version with PyCall. I think I did in the end - version 0.4.0, but now getting some difficulties with other julia packages. It's becoming sort of like whack-a-mole unfortunately.

But maybe you know this off the top of your head - if our old Docker version with RMG-Py 3.7, will it work with an RMS that requires PythonCall or is RMG utilising in its code PyCall when calling on RMS?

[jonwzheng]

New versions of RMG rely on the PythonCall interface to use the RMS features in the RMG-Py code (in addition to the RMS code). It won't work correctly with PyCall. This change was made simultaneously with our update to Python 3.9.

Is there something blocking the update to Python 3.9 that wouldn't allow you to use those new versions of RMG-Py and RMS simultaneously?

[calvinp0]

Okay, so I found a commit of RMS that worked and I think it was the latest version before moving over to Python 3.9 RMG.

Here it is

Regarding the usage of RMG 3.9, we had decided that we wanted to reduce our reliance on RMG-Py as a whole since it became overly complex to main ARC and also the installation process was a bit more full-on than we had liked. So, we have decided to separate entirely from RMG-Py in terms of needing it in the same Python environment. We extracted parts of molecule repo that Matt made and then call on RMG-Py through a subprocess. So, right now, we have all but completed that. We are just trying to get our tests to pass.