2020-11-13 QCA Submission Meeting notes

Wavefunctions

Trevor

• Ordering is canonical on QCArchive for wavefunction basis components; doesn’t appear to be same order as psi4

• BP: Psi4 should probably support the approach taken by QCSchema, not the other way around

• JH: will raise an issue on Psi4 for supporting QCSchema’s way of storing wavefunction components; see what the devs say about if this is the desired approach

BCC Refit Study

Josh

• Ready to go; just needs a re-roll of the dataset with QCSubmit
  
  

ANI1,cxx

Josh

• We’ll hold off on submission; want to let ANI2X make more progress before tossing this in; torsiondrives are proceeding, but slowly

PEPCONF

David

• Trevor reviewed, ready to go. One concern is large molecules with many conformers. This will be a big dataset, with some taking perhaps a long time to compute.

  • asked John for thoughts. Should be ready to proceed when we get them.

Phenyl Set 1 v2.0

David

• Josh reviewed; comments need to be addressed

  • problem molecules converge with latest psi4!

  • David will re-roll with testing given feedback

Protomers/tautomers OptimizationDataset

Pavan

• Dataset prepared; Josh provided feedback, Pavan is addressing

• Update of an old dataset with new information

MM compute specs for CHO sandbox dataset

David

• Ready for review! Trevor is interested in reviewing.

• Do we want validations for MM specs that try and do create_openmm_system (see if they can be parameterized)?

  • JH: actually step after this may fail more often, assigning charges

STANDARDSv3

Trevor

• STANDARDSv3 is for our consumption, guides our approach to datasets, automation

• DD: are there upstream changes we need to support the behaviour we want?

  • metadata overwriting is needed

  • should work, will need to test

• Ready for review! TG would like reviews from everyone if possible

• DD: Aiming for adoption by Dec. 1; once adopted, we can then spin out issues everywhere to implement the vision

• TG: CMILES for fitting is a hard problem, but worth trying to plot a course here; at the very least, filter out molecules whose final structures don’t produce the same CMILES on entry; would like solution to work in RDKit at some point

• TG: Reliable CMILES generation is dependent on good provenance for software versions, so making this

Dataset index

Josh

• JH: need to add the function to PR

Local Optimization Executor

Trevor

• TG: Working on it

Do we need more compute?

David

• TG: I think we’re okay; we have a lot of datasets right now

  • Can we measure e.g. optimizations per day?

• TG: Could probably get this from the database itself; recall Daniel provided these numbers regularly.

• BP: Can we assemble a document with a list of metrics we would like to see / would be useful for decision-making; for the dashboard we want to know this.

• DD: I’ll create a thread on Slack for assembling desired metrics

• JH: Is there any plan on QCArchive for a tool that can give predictions on the length of time a calculation will take?

  • BP: There are issues with this, but we are working on making this work

  • JH: Does it only work with single-point data, or things like Optimizations?

  • TG: Seconds per electron would be useful

  • DD: Having a validation step that uses the model to give a resource use estimate would be incredibly useful for dataset submission decision-making, capacity planning

  • JH: If there are datasets needed to help with this work, we’d be able to assist

• TG: interested in exploring this; will give this a shot from manager data

PCM support

Josh

• PCM support should all be in place; will be testing with Simon’s dataset

Public QCA

Ben

• Still getting issues with services failing (e.g. torsiondrives). Need to solve this today before managers spun up.

• DD: we’ll hold off on manager upgrade and deployment until Ben gives the go-ahead.