[mlpack] GSoC 2014 : Introduction and Interests

[Ryan Curtin]

On Wed, Mar 05, 2014 at 10:29:40AM +0530, Anand Soni wrote:
> Marcus,
> Thanks for the guidance! I will look up on these metrics.
> 
> As far as visual performance analysis is concerned, I came across
> d3.js and hexbins.js. Once we have the required data we can generate
> good visualizations using these tools. I plan on using either one or
> both of them.
> 
> Also, I was reading up on dual trees. Why exactly does mlpack need
> dual tree implementations? I am interested to work on k-d trees.

I'm confused about your question.  mlpack implements dual-tree
algorithms both because they are fast and also because they are the line
of research conducted by the lab that originally designed the library.
Often, dual-tree algorithms perform other algorithms for the same task;
for exact nearest neighbor search, there is really not a better strategy
than using a dual-tree algorithm.

The kd-tree is already implemented in mlpack and is the most stable tree
type, so there is not too much work left to be done for it.  If you are
interested in the implementation of tree types, take a look on the Ideas
page and there is a list of other possible tree types there.

> Is it alright to submit proposals for two projects?

Yes, this is fine.  But know that every mlpack Summer of Code project is
very in-depth, and the quality of the application is far more important
than the quantity of the application.  That is to say, your chances are
better if you submit one good application than two mediocre
applications.

I hope that helps.

Thanks,

Ryan

-- 
Ryan Curtin    | "I was misinformed."
ryan at ratml.org |   - Rick Blaine