Josh Fass

Yeah, noticed that in a few places, would need to refactor slightly so we have something like `parameter_names['atom'] = ['sigma', 'epsilon', 'charge']`, `parameter_names['torsion'] = ['periodicities', 'force_constants', 'phase_offsets']`, ..., or similar,...

Nice! Looks like something in this direction may be an improvement: would separate the computation of summary statistics from the generation of formatted reports, which are currently intertwined. A couple...

Thanks for kicking off this discussion! I think the main factors to consider are: (1) "correctness" -- does the batching / subsampling procedure produce an unbiased approximation of the objective...

>What are the performance characteristics as a function of graph size? Is there a performance penalty for small graphs, but then things scale linearly for larger batches of molecules wrangled...

>That this is impossible is the issue we're trying to address here. In order to batch the graphs have to be of the same size. Is there a reason the...

Some concrete numbers that would be helpful in this conversation are: * What is the maximum wall time per epoch we consider tolerable? (1 minute?) * What is the wall...

>Ideally we would just have a module where we list all the choices, and have arguments in a separate script allowing users to pick. Mostly agreed, hard part is defining...

Well-spotted -- thanks for catching this! Will fix this early next week, but I agree with assessment that this bug doesn't impact immediate results involving valence fitting. This also suggests...

I think we need at least a couple different dataset types, for the different tasks: * Interaction-type classification tasks (control) * Energy regression tasks * Fitting to QM * Fitting...

>Our target shouldn't care about the target-dependent offset, should it? I guess it depends what the goal is. For modeling the conformational distribution of a given molecule, any constant offset...