General topics | JW – What was your read on OBS/BSS meeting? MT – This meeting in principle seemed like a waste of time. I think it could have been an email, and I’m not particularly interested in sire or its implementation details (unless it becomes my job). So this is irrelevant until they submit issues/PRs. In general we have a bad track record with external promises - They often don’t actually come through, or we receive something that we don’t actually need. So I’m interested to see what materializes from this. DM did a good job of being “bad cop” and disowning responsibility for AMBER stuff in the short term. JM also seemed to have kinda mishandled the meeting - The part that he was there for centered on stuff that’s basically in the documentation, and then he posted his real question as he was leaving. JW – Mostly agree - I do like that we ended up not being on the hook for anything and that the conclusion was “they may open issues or PRs” MT – I’m also wary of their statements “we tried some stuff and it didn’t work” - I’m not a psychic and unless they open issues or actually say what the problem was then I don’t care. It just seems like a way to signal that they’ve invested time and gain social capital in the discussion without actually needing to provide any value. JW – Agree, next time we meet we should say “oh for that stuff that didn’t work please open issues”.
MT (rollover) – As manager, I’m asking who is responsible for determining what I develop in interchange JW – Literal answer is “governing board > PM > Tech lead > Matt”. In practice, each level up will provide less frequent and detailed direction. I think I’ve not provided enough direction, and probably need to reach up to PM to figure out how different directions of interchange directions should prioritized based on our strategy. I’m also open to supporting other structures if you have an alternative proposal, though I’d need Mobley to agree to them. MT – It’s accurate to say that I’ve not been given enough direction as this model would include if it worked well. I’m not pushing to have Mobley be the person I have to please. I think I’d misread what Mobley wrote up about reproducing the SAMPL6 files/energies. I’d like to know more clearly who I have to make happy and when I can make decisions myself. So it’s not clear how to safeguard myself against doing the wrong thing and getting a “no” from a person who could tell me that, but only after the fact. I think I could improve on making decisions and just going for it, but there’s this constant crosswind of complaints/ambiguously-authoried redirection requests. JW – I could improve a lot and get better at knowing what direction we’re taking, what we’re not taking, and why. I’ve been drifting away from the details recently. DM has a good idea for roadmap planning which is “hard decisions that should be made and hold for about a year”. Can we raise some number of these that will clarify interchange developments and user acquisition? MT – I think translating to monthly/quarterly objectives will be a hard challenge. A lot of operational stuff has improved with the ability to now say “no”. Right now everything seems to have about the same level of priority, and since I’m tied into everything I just kinda know how to prioritize in the short term. Like, I know that JH could use the DEXP stuff right now, and that people don’t need vsites immediately. Specific questions could be: Where do we sit with respect to AMBER and GAFF stuff? Are we expected to make Interchange.from_gaff("gaff-2.11", mol)? Do we really want from_openmm, from_gromacs, from_prmtop working? It would be a ton of work to get it working at a publication-level of validation. It seems like from_smirnoff should cover the majority of our mandate. How do we want to acquire interchange users and get bug reports/feedback/direction? We need to know how workflows will actually use things to focus development/UX effort. In the next year, what progress (if any) do we want to make toward doing direct parameter optimization using just Interchange and something like a jax optimizer?
MT – I could use multicomponent PDB loading. For the SAMPL stuff, I’m doing Topology.from_openmm, with waters, ions, and molecules loaded from SDF.
|