Jon Malkin
Jon Malkin
The binary format is defined here: https://github.com/apache/datasketches-java/blob/master/src/main/java/org/apache/datasketches/kll/KllPreambleUtil.java Obviously the format for items cannot be defined for arbitrary type T, and the exact size of the levels array is not fixed....
Interesting. I believe we install everything into a single datasketches directory, so we'd assumed the library would typically be installed. You're probable not doing that? Certainly not trying to tell...
If I try to add any new matrix sketches I think I'd try using Eigen (we have Frequent Directions in java for approximate SVD, for instance, but there may be...
@AlexanderSaydakov Several months later, I don't know if we want to revisit this yet?
@tjstum I know it's been a while but we haven't forgotten this. We discovered a change we merged recently does break the API slightly in some edge cases so we'll...
I believe it offers accuracy guarantees that t-digest cannot, while providing dual-tailed relative error. It’s numeric so it can do things REQ can’t (two-sided) and is likely smaller, too. I...
We've debated this before since it seems like a reasonable thing to do. The challenge is that in most cases it will break the sketch error distribution -- and still...
Just an update on where we are. While still skeptical of whether we really want to add a "shoot yourself in the foot" button in general, it's not an unreasonable...
Well, it's not serialized as part of the sketch and I guess that limits the utility, but at some point we moved the random_bit into common_defs.hpp where it's directly user-facing....
@RAMitchell Going to close this for now since we have _a_ solution even if perhaps not ideal. We'd be reluctant to change the serialization format even if we have a...