<div dir="ltr">Hi Ryan<div><br></div><div>I wanted to clear that the problem we are solving is essentially k-center one, right ? and not k-means. We already have an implementation of dual tree k-means.</div><div>From my understanding, the difference is that in k-centers, the center points or centroids are the points that necessarily belong to the dataset, and for k-means, it can be any point in the space.</div><div>However, in the project description, it's written that we need to find k-points which may not be in the set. But that is essentially k-means problem.</div><div>This got me a little confused, and I wanted to clarify what we are solving here.</div><div><br></div><div>Modifying the k-means problem to perform k-centers isn't much of a problem. What do you think ?</div><div><br></div><div>Regards</div><div>Abhinav</div><div><div><img src="https://t.yesware.com/t/395300192eaa89f8ac1d56afc3b0015c9edd19e7/1a81fb6268601deddbcaae4c4450460a/spacer.gif" style="border:0; width:0; height:0; overflow:hidden;" width="0" height="0"><img src="http://t.yesware.com/t/395300192eaa89f8ac1d56afc3b0015c9edd19e7/1a81fb6268601deddbcaae4c4450460a/spacer.gif" style="border:0; width:0; height:0; overflow:hidden;" width="0" height="0"><font face="yw-395300192eaa89f8ac1d56afc3b0015c9edd19e7-1a81fb6268601deddbcaae4c4450460a--to" style="display: none;"></font></div></div></div><div class="gmail_extra"><br><div class="gmail_quote">On Tue, Mar 1, 2016 at 8:11 PM, Ryan Curtin <span dir="ltr"><<a href="mailto:ryan@ratml.org" target="_blank">ryan@ratml.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><span class="">On Tue, Mar 01, 2016 at 03:08:28PM +0530, Abhinav Agarwalla wrote:<br>
> Hello<br>
><br>
> I worked really hard on understanding dual tree algorithms, and mlpack's<br>
> implementation of the same last year, but unfortunately we were not<br>
> selected. I still wanted to do the project, but got busy. I am really glad<br>
> that this time we are indeed selected.<br>
> I have gone through the literature, particularly N-body problems by A.<br>
> Gray, and cover trees one. I am also going through the list, that was<br>
> shared in an earlier post.<br>
> I was hoping if either Bill or you could share the idea, and the work that<br>
> needs to be done.<br>
><br>
> Considering that organisation slots are fixed and numerous projects, what<br>
> is your priority for this particular project ?<br>
> I also couldn't find any issues surrounding this. Is there any<br>
> implementation that I can start with ?<br>
<br>
</span>Hi Abhinav,<br>
<br>
I remember our discussions from last year; probably the best<br>
documentation on this project at the moment are these emails, which you<br>
may have already seen:<br>
<br>
<a href="https://mailman.cc.gatech.edu/pipermail/mlpack/2015-February/000610.html" rel="noreferrer" target="_blank">https://mailman.cc.gatech.edu/pipermail/mlpack/2015-February/000610.html</a><br>
<a href="https://mailman.cc.gatech.edu/pipermail/mlpack/2014-March/000330.html" rel="noreferrer" target="_blank">https://mailman.cc.gatech.edu/pipermail/mlpack/2014-March/000330.html</a><br>
<br>
A good thing to do, if you are familiar with dual-tree algorithms now,<br>
would be to study the k-centers problem and think about possibilities<br>
for an appropriate BaseCase() and Score() function. This is a research<br>
project, so there is not yet a known dual-tree algorithm to solve the<br>
k-centers problem; there will be no reference implementation to consult.<br>
<br>
Another idea might be to go read about the Gonzalez algorithm for<br>
k-centers and then implement it to see how it works.<br>
<br>
About the priority of the projects: all of the projects on the ideas<br>
list are important---otherwise they wouldn't be there. :) It's not<br>
possible to say at this point which projects will have slots allocated<br>
to them and which will not; that depends on the applicants, the mentors,<br>
and how many slots Google gives us this year.<br>
<br>
I hope this is helpful. Please let me know if I can clarify anything.<br>
<br>
Thanks!<br>
<span class="HOEnZb"><font color="#888888"><br>
Ryan<br>
<br>
--<br>
Ryan Curtin | "Open the pig!"<br>
<a href="mailto:ryan@ratml.org">ryan@ratml.org</a> | - Frank Moses<br>
</font></span></blockquote></div><br></div>