[mlpack] Hello Everyone

Sat Feb 4 09:20:01 EST 2017

Thanks Marcus and Ryan for your replies.

Please accept my apologies for the late reply. This week at college has
been quite a busy one.

Following your advice, I went through previous year discussions on these
two ideas and was
able to find some interesting conversations on both of them. I came across
some papers [1, 2]
which were recommended as a good read to gain some theoretical knowledge of
the proposed
models.

I'm going to spend next few days reading those papers and also will try to
share with you some
new ideas/thoughts as they manifest for some constructive feedback and
discussion thereupon.

Also, I noticed that PR #831 was finally merged. I think it's also good
time for me to start making
a broader sense out of it to understand the current state of ANN modules.

About that -- it'd be great help if Marcus could summarise a bit about his
thoughts on the current
state and near-future changes that are expected in this module as it is
closely to Deep Learning
Modules project idea and has direct effect on its future shape.

Also, looking forward to the updated ideas wiki page. :)

[1] https://mailman.cc.gatech.edu/pipermail/mlpack/2016-March/000842.html
[2] http://knife.lugatgt.org/pipermail/mlpack/2016-March/002578.html

Thanks,
Vivek Pal

On Tue, Jan 31, 2017 at 1:15 AM, Marcus Edel <marcus.edel at fu-berlin.de>
wrote:

> Hello Vivek,
>
> >> Yes, I'll look through the list to gain some insights. I also noticed
> >> that there's one open PR (#831) with major changes for ANN module
> >> lined up. Since, it looks like it will be merged soon can you comment
> >> on this year's focus for the deep learning modules idea after this PR
> >> is finally merged or will it remain same as last year's of
> >> implementing RBM, DBN etc.?
> >
> > I'm not the mentor for the deep learning code so I can't say much there,
> > but I know that Marcus has merged it in the past few days.  I think when
> > the ideas page gets updated over the next week or two, there will be a
> > lot more information on the goals this year for the types of projects we
> > are interested in.
>
> I'm the one who came up with the 'Essential Deep Learning Modules' project
> idea.
> The focus of the Essential Deep Learning Modules' project will slightly
> shift
> and will be concentrated on more recent ideas. It doesn't mean an
> implementation
> of e.g Deep Belief Networks or Radial Basis Function Networks isn't
> welcome, we
> will definitely look over the code if someone opens a PR. As Ryan already
> said
> we will update the ideas page over the next week or two and add a bunch of
> interesting information.
>
> Thanks,
> Marcus
>
>
> > On 30 Jan 2017, at 19:47, Ryan Curtin <ryan at ratml.org> wrote:
> >
> > On Wed, Jan 25, 2017 at 11:23:20PM +0530, Vivek Pal wrote:
> >>> I'm still willing to mentor the CF idea.  If you search the mailing
> list
> >>> archives, I seem to remember there being a lot of discussion about the
> >>> idea last year, and maybe you can find some insights there.
> >>
> >> It's great that CF idea is still open. I will certainly look into the
> >> list archives to gather some useful information regarding the project.
> >> If I'm not wrong this idea was not one of the selected project last
> >> year, so I think it'd be worth finding out what was lacking. May be
> >> you want to add some input on that?
> >
> > We got very many good proposals last year (in total we received over
> > 100!) and we had only 6 slots.  If I remember right, the CF code did not
> > receive too many proposals.  It's possible that last year, there was
> > simply more interest in other projects from both the mentor and the
> > student side.
> >
> > I'll be going through the Ideas List soon and try and think of good
> > projects for improving the CF code.
> >
> >>> In the end, anything that helps the CF code get easier to use or
> >>> more functional so that users can easily build a recommendation
> >>> system would be a great project.
> >>
> >> Will the focus be on making existing CF code more functional this year
> >> as compared to last year's focus on implementing alternatives to k-NN
> >> based CF?
> >
> > I'm not picky---both could be good projects.  Or even a project that
> > factored in both ideas could be good also.  In the end, if I mentor a
> > project on the CF code, I am looking for a clear proposal that will have
> > a tangible benefit.  This could take the form of improvements to the
> > existing code or adding new approaches for collaborative filtering.
> >
> >>> I also remember a lot of discussion about the deep learning modules
> >>> project; I'm sure you can find a lot about that one in the archives
> too.
> >>
> >> Yes, I'll look through the list to gain some insights. I also noticed
> >> that there's one open PR (#831) with major changes for ANN module
> >> lined up. Since, it looks like it will be merged soon can you comment
> >> on this year's focus for the deep learning modules idea after this PR
> >> is finally merged or will it remain same as last year's of
> >> implementing RBM, DBN etc.?
> >
> > I'm not the mentor for the deep learning code so I can't say much there,
> > but I know that Marcus has merged it in the past few days.  I think when
> > the ideas page gets updated over the next week or two, there will be a
> > lot more information on the goals this year for the types of projects we
> > are interested in.
> >
> > But, also, keep in mind that you are always welcome to propose your own
> > project idea!  I have seen some great proposals along those lines, where
> > the student has identified a problem (or something lacking inside
> > mlpack) and then developed a proposal that addresses that problem.
> > Yannis' project last year on LSH tuning is a great example of this type
> > of approach.
> >
> > I hope this is helpful---let me know if I can clarify anything.
> >
> > Thanks,
> >
> > Ryan
> >
> > --
> > Ryan Curtin    | "Happy premise #2: There is no giant foot trying
> > ryan at ratml.org | to squash me." - Kit Ramsey
> > _______________________________________________
> > mlpack mailing list
> > mlpack at lists.mlpack.org
> > http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://knife.lugatgt.org/pipermail/mlpack/attachments/20170204/74864dd0/attachment.html>