Ted Dunning

My feeling for this is that addition is more robust than subtraction. As such, we can build a history of digests for relatively small periods and occasionally combine them into...

Restarting. Yes. This approach to decay sounds pretty reasonable, but you will have very little idea about accuracy as things begin to decay. I suppose that you can still say...

On the topic of integration with Prometheus, I think that sounds like a great idea. How can I help?

OK. That's interesting. Sad as well. Things like Hdr histograms have constant bin boundaries so this idea of having a different time series for each bin (or for each cumulative...

This proposed encoding looks totally usable. On Mon, Jan 18, 2021 at 4:23 PM ajwerner wrote: > One possibility is to not use an additive measure if Prometheus allows >...

I think that this test is flawed by having assumptions out of view. The number of centroids was *always* defined as

Are you suggesting that we have different serialization models for, say, protibuf and any other format? On Jun 28, 2017 11:29 AM, "Ankur Jain" wrote: > Isn't it possible to...

I think your design is fine. It bottlenecks the essential qualities of the t-digest. How about we find a better name for the SerializationModel (it might have other uses, even)....

DigestModel sounds fine! On Thu, Jun 29, 2017 at 1:59 AM, Ankur Jain wrote: > Or DigestModel? I see this as a dto representing digests. > Sure, I have some...

Yes, currently different fields are serialized, but we should synchronize that. The minimum set includes: - compression factor - min and max values observed - centroid positions - centroid weights...