General updates | MT Taking an intro to deep learning course, a little bit of time (<2h) each day Free, online, pretty accessible: https://lightning.ai/courses/deep-learning-fundamentals/ JW – Very cool. I did the Stanford ML course with Andrew Ng and really liked it - was very hands-on and I feel I got a good working understanding. MT – I’d heard that one is good but somewhat outdated (eg doesn’t use pytorch)
Cancel OpenMM meeting Friday? JW – Worth mentioning OpenMM 8.2 noises from interchange regression tests and OpenFE? MT – Was unable to isolate our regression test failures to OpenMM. Unfamiliar with OpenFE issues. Should I follow up? JW – I think OpenFE may be avoiding it due to an abundance of caution until it’s validated. Not worth following up. I fixed some numerical errors recently where ARM mac CI was failing by forcing use of Reference platform in OpenMM. Possibly something up with OS GPU presentation/OpenMM platform seleciton logic. MT – Good to know.
Aiming to work Mon & Tues of next week, but unstructured (i.e. no agile) Any odds and ends you want me to look at? Triage/support, chipping away at little bits of debt, or whatever catches my attention
Rough pass of all tutorials done, a little confused and pessimistic about the direction of the project MT – Basically all gromacs stuff fails since we can’t read/write itp files. JW – Writing itps wouldn’t be a requirement - if we can write a top that gets the same energy then that’s a win. MT – I estimate that reading itps will be a large amount of work. Can’t recall if we can read tops. JW – Added to task tracker JW – I wonder how many of these tutorials we could roundtrip using OpenMM’s parser. MT – That’s possible, could be worth testing. Kinda kicks the can down the road for a “good” solution and would require OpenMM as major dep. Also adds complexity/information loss by going through different steps+tools. JW – Added to task tracker Added new section about next steps to bottom of agenda
Working with Brent on getting torsions running on YDS - Currently torsiondrives are at MVP stage - Can regenerate results and do other handy tasks easily. So I’m working with BW on getting this running. ETA to getting torsions running on YDS is pretty soon (days). JW – Cool. Thanks for reaching out to work with BW. I think it’s good that the torsiondrive data structure is pretty analysis-agnostic, since the science team is just starting to figure out what kinds of analysis they’ll want to do, and this will offer tehm lots of flexibility.
Re: AmberTools issues, a little frustrating to document issues, ruffle a lot of feathers about it, and then get feedback about how people were not sufficiently informed about it hand off tasks to Amber folks and not really get anything in return (mailing list kinda dead? http://dev-archive.ambermd.org/202410/ ) JW – I don’t think we should spend significant energy supporting AmberTools. We can spend an hour helping MH or something as part of our “neighborly OMSF” duties. MT – I’m frustrated that people are asking for summary of current status and I’d made that available before. Unclear that Amber devs are making progress on this. I do have ideas for what to try but if I’m the last one to have opened a PR then I’ll be seen as the responsible party. Ideally MH would take the lead on this and I could probably provide valuable support.
JW Did I get monthly goals right for ad board? I was preparing after hours and didn’t want to bother you but realized that I’m less aware of what you’re working on with agile/big picture style planning is less evident (but perhaps less necessary). I think solution is for me to start joining planning meetings. Re: PE can’t make OpenFF meeting Fri - I saw that us and OpenFE were making noises about OpenMM 8.2 - Worth bringing up with him? Zhiyi followup. MT – Yes, please join sprint planning. Next one will be Dec 2 MT – You kinda got monthly goals right. There’s kinda a vacuum of establishing larger goals. I would need more guidance from org level about big objectives to plan my monthly objectives. So I think this is where JW and JE should come in. MT – Would love for there to be suggestions prepared for sprint planning so it’s not just status quo (me thinking of what I should do) JW – Agree, I’ll start recording to-do items for whole team on my prioritization spreadsheet (which may change form over time) https://docs.google.com/spreadsheets/d/1SNLSRNC73Ybv-s7o7GTFJD0YL5RpmRvpLBakbHPF5GE/edit?gid=0#gid=0
Further Zhiyi followup? MT – If CHARMM-GUI’s system prep infrastructure is a 10, I suspect WZW’s is a 2 (where 0 is good). I suspect WZW is still doing a bit of sketchy unvalidated parmed stuff but less than charmm-gui. So I’d take their use cases with a grain of salt, and not rush to thinking that these pathways are production-ready and ready for us to make into a product. MT – Can we drop testing of intel macs?
|
Review of tutorial results | Goal: Digest current results and think of major options for JE to consider. |