@posco Is it known CMS behavior that ++ w.r.t. heavy hitters is kinda "broken" because for heavy hitters ++ is not associative?
-
-
@avibryant@posco True. AFAIU the intersection is the problem as it's not associative. It introduces "left-biased" monoculture when reducing -
@avibryant@posco s/intersection/subset/
End of conversation
New conversation -
-
-
@avibryant@posco CMS-"local" HH on "right" CMS's are less likely to become global HH when reducing.Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@avibryant@posco The problem is similar to micro vs macro average.Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@avibryant@posco For special case of 2 CMS you can trick by unioning both HHs, re-querying estimates from /combined/ CMS, then dropping HH -
-
@avibryant@posco Not in general, given the way the monad magic adds 1 CMS at a time. For the case of N CMS you must union all N HHs at once -
@avibryant@posco And even then the problem is not yet solved. I'll try to summarize in a GH issue. Still hope I'm mistaken. ;-) -
-
-
@avibryant@posco Sure, np. Given this limitation of topN CMS, I begin to feel it should not be included in AB or only with a big warning :)
End of conversation
New conversation -
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.