2023-08-21 Meeting notes

• Trello review

  • https://trello.com/c/okbqsXhz/643-look-into-hooking-up-evaluator-to-benchmarking-code

    • No progress. I have the starting materials, just haven’t taken action yet.

  • https://trello.com/c/wLNTkVRf/644-test-benchmarking-code-with-a-force-field-that-uses-plugins

    • https://openforcefieldgroup.slack.com/archives/C03T3LLVC1J/p1692634193346509

    • JW – Looks like some mols have super high ddEs

    • MT – Will look into that.

  • https://trello.com/c/Y8ernXOm/616-compare-results-to-existing-pavan-simon-scripts

    • LW – No progress on this. BW, any update?

    • BW – I ran the whole process through on a small dataset, and now I’m running on the whole industry set. It’s been converting molecules to records for 10 hours now.

    • LW – Might need another week on this one.

• MT – Not nominating for a new ticket, but dataset.to_records is some sort of bottleneck in QCSubmit.

  • BW – Actually profiled this last week, and it looks like the majority of time is spent on requests. I have a rust tool that seems to do well here.

  • MT – Does this hit the request limit?

  • BW – Apparently not. Just running a bunch of processes.

  • MT – When I load from JSON…

  • …

  • JW – I’ll focus on serialization pathways for retrieved data once I’ve got QCSubmit working again.

New tickets:

QCSubmit

LW – Is this a good place to discuss QCS planning?

JW – Yeah, we just released QCSubmit 0.5.0, which should work with legacy QCF. Next I’m finishing up some OFF toolkit stuff. Then on to QCSubmit.

LW – Should I expect API changes?

JW – No, I think we can get a minimal working version by just fixing interface to QCF, which I may simplify by stripping out as much QCSubmit caching as possible, while keeping external API the same. The caching may not even be necessary any more since the QCF refactor does a lot of stuff server-side. So the initial fixed release may run slow if there are some caches that we missed.

LW – How was Industry dataset issue resolved?

JW – RDK and OE backends will differ by about 50 mols. Ultimate fix will need to go in OFFTK.