[mlpack] [GSOC 2016] Contribute for MLPack

[Parijat Dewangan]

Hello,


   - I have uploaded my draft on GSOC 2016 site. Please review it.



I have few doubts regarding the implementation of trees. Currently, I have
mentioned 5 trees which would be implemented in the project time - Vantage
point trees,k-means trees, random projection trees, Bregman ball trees,
segment trees.. Among these, I have coded vantage point trees based on a
paper. You can find the code on my github.

Please review the other trees, whether it would be fine to implement them.
Please tell if you have some tree type in mind which I have missed. Also, I
was

On Fri, Mar 11, 2016 at 8:26 PM, Ryan Curtin <ryan at ratml.org> wrote:

> On Fri, Mar 11, 2016 at 06:00:28PM +0530, Parijat Dewangan wrote:
> > Hi,
> >
> > I went through various research papers to have a better understanding of
> > the dual tree algorithms and various trees. Currently, I am focusing on
> > vantage point trees, by referring the following papers.
> >
> >    -  "IMPROVING DUAL-TREE ALGORITHMS"
> >    <http://www.ratml.org/pub/pdf/2015improving.pdf> Thesis by Ryan
> Curtin .
> >    -  "Data Structures and Algorithms for Nearest Neighbor Search in
> >    General Metric Spaces by Peter N. Yianilos*"
> >
> > I am completely comfortable with the MLPack API after going through the
> > above thesis of Ryan Curtin. So, I was thinking of coding vantage point
> > trees. What do you suggest ?
>
> Sure, vantage point trees would be interesting.  Note that vantage point
> trees are actually the same as metric trees, so we'll have to provide
> some documentation somewhere indicating that they are the same thing.
>
> > Should I provide you with the pseudo code of Vantage Point Trees? Or
> should
> > I try fixing some issues? I was thinking of taking up  issue #275.
> > https://github.com/mlpack/mlpack/issues/275.
>
> If you can do #275 without breaking any of the tests, please feel free
> and I'd be happy to merge in the improvement.  It will be a significant
> refactoring.
>
> Another possibility from there would be to implement a leaf size
> parameter for cover trees, that would cause the tree building process to
> terminate when the number of points in a node was small enough.  But I
> think that would be a lot more difficult and maybe we can save that for
> another day... :)
>
> Thanks,
>
> Ryan
>
> --
> Ryan Curtin    | "Like, with jetpacks?"
> ryan at ratml.org |   - Scott
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cc.gatech.edu/pipermail/mlpack/attachments/20160319/8ceb9de1/attachment-0002.html>