What is the license for this software? about streaming-quantiles

[PavelVesely]

Hi Ted,
I'd actually suggest you to use the DataSketches implementation of KLL or ReqSketch for your testing and benchmarking. Their implementation is production-quality and also in Java, making the comparison meaningful, and they are under the Apache license. I believe the purpose of this repo is to demonstrate proof of concept.

[tdunning]

I looked at that implementation. But it only implements the uniform absolute error. What I want to compare against is the relative error streaming quantile estimator. Is there a relative error implementation in Java?

…

On Fri, Jan 29, 2021, 1:10 AM PavelVesely ***@***.***> wrote: Hi Ted, I'd actually suggest you to use the DataSketches implementation <https://github.com/apache/datasketches-java> of KLL or ReqSketch for your testing and benchmarking. Their implementation is production-quality and also in Java, making the comparison meaningful, and they are under the Apache license. I believe the purpose of this repo is to demonstrate proof of concept. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#5 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAB5E6VPY5HVC6CB36PVXEDS4J3RBANCNFSM4WYK7OOQ> .

[PavelVesely]

DataSketches very recently incorporated a Java implementation of the same relative error quantiles algorithm as here, and actually their code is based on my Python code in this repo (which was implementing the algorithm from this paper: https://arxiv.org/abs/2004.01668 ).
The Java implementation is not released yet (that should be done soon) but you can find it here: https://github.com/apache/datasketches-java/tree/master/src/main/java/org/apache/datasketches/req
I was helping them to develop it and I think it's pretty much ready to be used in experiments.

[tdunning]

Pavel, Thanks very much for that pointer. That is very helpful.

…

On Fri, Jan 29, 2021 at 10:01 AM PavelVesely ***@***.***> wrote: DataSketches very recently incorporated a Java implementation of the same relative error quantiles algorithm as here, and actually their code is based on my Python code in this repo (which was implementing the algorithm from this paper: https://arxiv.org/abs/2004.01668 ). The Java implementation is not released yet (that should be done soon) but you can find it here: https://github.com/apache/datasketches-java/tree/master/src/main/java/org/apache/datasketches/req I was helping them to develop it and I think it's pretty much ready to be used in experiments. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#5 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAB5E6RHIBYLCCKTBCFM7KDS4LZZJANCNFSM4WYK7OOQ> .

[tdunning]

I should point out that my particular needs are satisfied by Pavel's kind pointer, but the lack of clarity insofar as licensing this Python code still apply. I don't particularly care about that, but my pull request is still there for the taking if you guys want to clarify things (and assuming I got it right). Thanks again.

…

On Fri, Jan 29, 2021 at 12:45 PM Ted Dunning ***@***.***> wrote: Pavel, Thanks very much for that pointer. That is very helpful. On Fri, Jan 29, 2021 at 10:01 AM PavelVesely ***@***.***> wrote: > DataSketches very recently incorporated a Java implementation of the same > relative error quantiles algorithm as here, and actually their code is > based on my Python code in this repo (which was implementing the algorithm > from this paper: https://arxiv.org/abs/2004.01668 ). > The Java implementation is not released yet (that should be done soon) > but you can find it here: > https://github.com/apache/datasketches-java/tree/master/src/main/java/org/apache/datasketches/req > I was helping them to develop it and I think it's pretty much ready to be > used in experiments. > > — > You are receiving this because you authored the thread. > Reply to this email directly, view it on GitHub > <#5 (comment)>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/AAB5E6RHIBYLCCKTBCFM7KDS4LZZJANCNFSM4WYK7OOQ> > . >

[PavelVesely]

I'm fine with changing it, but this needs to be decided by Edo (the license header for relative error code here was already change to the Apache license).

[tdunning]

Speaking from my experience as VP of Incubator at Apache, the README statement carries a lot of weight and the contradiction between the headers and the README would, if this code were a dependency, completely prevent the use of any code in this repo. Taking a single file out of this repo when the README says what it says doesn't suffice at all. In my commercial life, I had a *major* customer completely nix a delivery because a dependency of a dependency of an Apache project had a license problem not all that different from this. I had to spend my Christmas break coding an alternative and convincing several Apache projects to accept it. It is good that there is a complete rewrite with a clean lineage in the Java form since that makes the currently unusable state of this code moot.

…

On Fri, Jan 29, 2021 at 3:13 PM PavelVesely ***@***.***> wrote: I'm fine with changing it, but this needs to be decided by Edo (the license header for relative error code here was already change to the Apache license). — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#5 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAB5E6RDFISPQLDSA6AQML3S4M6JNANCNFSM4WYK7OOQ> .

[edoliberty]

Hi Ted, sorry for just now noticing this thread.
This repo is a research project. The line "It should NOT be used for any commercial purposes" is a warm recommendation more than a legal usage guideline...

I'm happy to soften the language if you think it's worth doing. Regardless, apache datasketches should be the authoritative implementation for KLL, not my untested research python code...