Benchmarks for the RMG v3.3 paper, comparing it against historical versions of RMG.
TODO: it appears that the CoW behavior of Python 2.7 and 3.7 has been reduced/eliminated in Python 3.11, meaning that rmg 3.3.x could, when adding support for 3.11, drop the check of .free memory and multiprocess exactly how the user requested without memory growing unbounded (check the corresponding docs about this, may also need to test Python 3.9 and 3.10)
TODO: choice of benchmarks is currently ch3no2 (diverse atomtypes), octane (solvated), c3h4 (classic combustion)
I have provided working installation files for Linux.
When I say working, I mean that you can run them today and they actually resolve an environment that allows RMG-Py to run.
The environment files included in the original code don't work anymore because new versions of many of the dependencies have been released; the environment files don't forbid conda from installing these, so they often install incompatible packages.
These are the three versions which I have set up, each of which has its own subdirectory:
2.4.1: last Python 2 release, uses Python 2.7 (commit is here on GitHub)3.0.0: first Python 3 release, uses Python 3.7 (commit is here on GitHub)3.3.x: version described in the RMG-Py 3.3 manuscript; supports Python 3.9 through 3.11, this setup uses 3.11 (todo: release this version of RMG-Py and include it here)
To actually run the install, just navigate to the corresponding subdirectory and run install.sh.
Within each version's subdirectory, there is a test.sh script that executes the tests, writing the results to both the terminal and a timestamped log file.
It will auto-magically run on all available CPUs.