modelop / augustus

Augustus is an open source system for building and scoring statistical models designed to work with data sets that are too large to fit into memory
43 stars 21 forks source link

v4.2 #1

Open robert-howley-zocdoc opened 8 years ago

robert-howley-zocdoc commented 8 years ago

Is there a timeline for or any intent to support PMML Spec 4.2 in Augustus? I noticed that there hasn't been a commit in a few years and wasn't sure if there were any plans on reviving this.

Thanks

yomiBazuaye commented 7 years ago

Hi, I just want to echo Robert's question: anyone working on support for 4.2?

Also: does anyone know of any other open source Python-PMML model consumers?

rtbs-dev commented 7 years ago

Wondering similar, though more-so for v4.3 ... With the implementation of Bayes-Nets in 4.3, I'd like to use augustus to parse to and from PMML with something like PyMC3 (as opposed to the more standard sklearn parsers I see everywhere, since there's no generic BN implementation in sklearn).

In the mean-time, any advice on how to proceed?

AkisAya commented 5 years ago

hey I've encountered the similar problem as robert did. wonder is there a plan to support pmml 4.3?

or any other open-source pmml parser for python, please let me know, thx

rtbs-dev commented 5 years ago

@AkisAya I wrote a very rough parser to-and-from PyMC3 here, as part of an upcoming paper to illustrate 4.3.

Just note it uses sympy (read, eval) to parse symbolic dependencies. There are some issues with highly complex networks from a sampling perspective, but this was built with non-linear-gaussian models in mind: example.

I went ahead and just pushed a commit to update to networkx 2.0. Still, it would be nice to see a bit more robust support for pmml-->python. Not sure if this project is still active.