Run torsion comparisons

[mattwthompson]

• Make a different main.py that runs torsions
  • This will use TorsionStore and some iteration of QCArchiveTorsionDataset
• Store some TorsionDrive dataset in datasets/
  • New version of download_and_filter_dataset.py (which is run once per dataset/iteration thereof, not in different versions of CI) OR
  • Just add a cached dataset into datasets/
  • The name of the dataset is reference in the "run config" file, which is used in workflow triggering
• Actually make models for storing torsion outputs
• Think about updating the existing workflow, which globs up whatever is there. Make sure it adds the torsion outputs, whatever they are
• Make sure the TorsionStore database are successfully uploaded to Zenodo
• Work with science team to establish a (rough) dataset for benchmarking torsions (100-200?)
  • As a stopgap, @ntBre can just filter out an arbitrary 100-200 torsions from the fitting dataset (some nuance with the filters that are applied)
• Build out basic metric(s) for qualitatively analyzing torsion profiles (i.e. without needing to look at dozens of plots)
  • Only want a ballpark estimate, leave room for more detailed analyzes in the future

[mattwthompson]

ref openforcefield/yammbs#76

[mattwthompson]

With openforcefield/yammbs@a7298a8, MM profiles and some (extremely approximate) metrics and be stored after optimization

[ntBre]

That's great news! Does that mean I should work on a new version of main.py and a dataset script?

[mattwthompson]

I started something in #13 - please suggest (or simply make) whatever changes stand out as useful to you.

In my local setup (with YAMMBS at that commit), these files script generates a plot, a metrics JSON, and an "outputs" JSON which stores the MM-minimized profiles: https://gist.github.com/mattwthompson/820328f2736193e3edb06eadc0bac742