When running Synoptic on large logs, there's no way of gauging how long model
inference might take (30 seconds? 5 hours?). It would be very useful for large
experiments to have some (even very approximate) progress notification.
For refinement, adding an approximate progress indicator could simply involve
recording the number of invariant violations when refinement first starts
(initialViolations) and occasionally reporting:
"Refinement approximately (currentNumViolations / initialViolations * 100)%
complete"
For coarsening, we could start by computing or approximating the number of
merges that will be attempted (mergesToAttempt). We could compute it exactly by
counting the number of nodes of each event type in the partition graph and
calculating the number of permutations for each, or we could approximate it by
calculating the permutations of something like (numTotalNodes / numEventTypes),
which simplifies the problem by assuming nodes are equally distributed among
event types. Either way, we could then occasionally report (attemptedMerges /
mergesToAttempt).
Original issue reported on code.google.com by tonyohm...@gmail.com on 12 Mar 2014 at 4:26
Original issue reported on code.google.com by
tonyohm...@gmail.com
on 12 Mar 2014 at 4:26