You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Use the round-based approach: at the start of each round, we know the number of scans that will occur. Therefore, we can select an adequate batch size beforehand.
To keep the estimator synced with the round-based scheme, at each empty! call we can double an internal value for the number of scans.
Difficulties
Since the Shared object is not seen by recorder constructors, there is no way of knowing if we are starting from round 1 or round > 1. So we can only keep the estimator in sync if it started in sync.
The text was updated successfully, but these errors were encountered:
RE the issue with not knowing the initial number of scans: we can learn after the first mistake. So perhaps the first round the estimate is NaN, but the following round should work.
Idea
empty!
call we can double an internal value for the number of scans.Difficulties
Shared
object is not seen by recorder constructors, there is no way of knowing if we are starting from round 1 or round > 1. So we can only keep the estimator in sync if it started in sync.The text was updated successfully, but these errors were encountered: