wwaites
commented
1 month ago Reporting at every simulation step might well be too frequent. Often you want coarse reporting time (e.g. one simulation day) and there are many thousands of events between reporting times.
Computing the counts to report can be costly so even though you might be able recompute all the observables after every step and only save them periodically, in practice that can be wasteful.
AKA epiphenomena: we presently store almost all the data of the run, expecting it to be processed after the fact. This approach will eventually not be able to scale, so we should consider offering the user a syntax for saying what should get logged in real time as the model runs.
One approach for logging for specific rewrite rules is to embed the pattern in some larger context $L \hookrightarrow Ctx$, and we pick out a subset $Log \hookrightarrow Ctx$ which is the relevant values to log (this latter mono might preferably be a map in $C$-Par, since it may not be a proper subobject) whenever the rewrite fires (the $Log$ subset of data of all maps $Ctx\rightarrow World$ that make the triangle (with a match $L \rightarrow World$) commute.
One approach for general logging after any step is to give patterns and report the counts (or the data of all matches) of that pattern in each resulting state of the world.
The partial maps induced by rewrites give us the ability to have the full trajectory, but we should come up with means of getting longitudinal data that may be smaller than the entire trajectory.