madlib-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Frank McQuillan <fmcquil...@pivotal.io>
Subject Re: Apache MADlib next release discussion
Date Wed, 27 Sep 2017 20:04:18 GMT
I made you the owner of that JIRA Ed.

On Wed, Sep 27, 2017 at 12:11 PM, Ed Espino <espino@apache.org> wrote:

> As it relates to MADlib dev community growth, I'm definitely interested in
> taking a stab at https://issues.apache.org/jira/browse/MADLIB-1025
> (Support
> modern versions of gcc). Working on a few v1.12 issues, I am a bit familiar
> with the gcc issue from the end user perspective.  I will grab this issue
> and start the hunt for a cure!
>
> Thanks,
> -=e
>
> On Wed, Sep 27, 2017 at 11:43 AM, Frank McQuillan <fmcquillan@pivotal.io>
> wrote:
>
> > Thanks Ed for your offer to contribute.  Are you interested in this JIRA:
> >
> > Support modern versions of gcc
> > https://issues.apache.org/jira/browse/MADLIB-1025
> > ?
> > Seems like a gap we should fill.
> >
> >
> > Looking thru the 2.0 JIRAs, the emerging themes seem to be: performance
> > improvements, more graph algorithms, misc feature improvements:.  For
> e.g.,
> >
> > [performance]
> > Add mini-batching to IGD framework
> > https://issues.apache.org/jira/browse/MADLIB-1048
> >
> >
> > [graph]
> > HITS
> > https://issues.apache.org/jira/browse/MADLIB-1124
> >
> > graph cut
> > https://issues.apache.org/jira/browse/MADLIB-1074
> >
> >
> > [feature impr]
> > k-NN distance functions
> > https://issues.apache.org/jira/browse/MADLIB-1059
> >
> > Neural nets grouping
> > https://issues.apache.org/jira/browse/MADLIB-1149
> >
> >
> > This seems reasonable to me, to have various different types of things
> > coming into 1.13 rather than a single theme . The mini-batching is
> > something I am excited about since it could help performance for a number
> > of different modules that use gradient descent.
> >
> > But I'll wait till next week to get more comments from the community
> before
> > making any version changes.
> >
> > Frank
> >
> > On Tue, Sep 26, 2017 at 2:40 PM, Ed Espino <espino@apache.org> wrote:
> >
> > > I like the idea of not rushing the Apache MADlib v2.0 and giving the
> new
> > > interface time to sufficiently bake. Beyond the three items marked
> > resolved
> > > for v2.0, do we have a sense of what the theme will be for MADlib
> v1.13?
> > >
> > > Current items targeted with v2.0:
> > > https://issues.apache.org/jira/projects/MADLIB/versions/12333619
> > >
> > > FYI: I'm looking for ways to contribute to v1.13!
> > >
> > > -=e
> > >
> > > On Tue, Sep 26, 2017 at 2:17 PM, Frank McQuillan <
> fmcquillan@pivotal.io>
> > > wrote:
> > >
> > > > MADlib community,
> > > >
> > > > tl;dr
> > > >
> > > > Proposing we do a 1.13 release in Nov before Thanksgiving, and a 2.0
> > > after
> > > > that around Jan.
> > > >
> > > >
> > > > More
> > > >
> > > > The next release is currently numbered 2.0 as per current JIRAs,
> which
> > > > includes interface changes
> > > > https://issues.apache.org/jira/browse/MADLIB-1157
> > > >
> > > > Because of the interface changes, the 2.0 release is not backward
> > > > compatibility as per semantic versioning that Apache MADlib follows
> > > > http://semver.org/
> > > >
> > > > We have started a couple of wiki pages to discuss ideas for the new
> 2.0
> > > > interface:
> > > >
> > > > Ideas
> > > > https://cwiki.apache.org/confluence/display/MADLIB/
> > Interface+2.0+-+Ideas
> > > > IO standardization
> > > > https://cwiki.apache.org/confluence/display/MADLIB/
> Interface+2.0+-+IO+
> > > > Standardization
> > > >
> > > > but it is not baked yet.  So I am wondering if we ought to do a 1.13
> in
> > > Nov
> > > > before Thanksgiving to give us more time to collectively make a plan
> on
> > > the
> > > > interface, which would go in 2.0 around Jan or so.
> > > >
> > > > Please let me know what you think, and also comment/add/subtract on
> > > what’s
> > > > proposed for the interface changes.
> > > >
> > > > Frank
> > > >
> > >
> > >
> > >
> > > --
> > > *Ed Espino*
> > >
> >
>
>
>
> --
> *Ed Espino*
>

Mime
View raw message