[mlpack] mlpack Contribution

Ankit Aggarwal ankit.ankit.aggarwal at gmail.com
Sun May 7 13:44:51 EDT 2017


Hi Ryan

So I have been going through your Ph.D. thesis explaining the dual-tree
algorithms and how the BaseCase and Score are defined especially the
chapter-7(Algorithms).

So here is approach what I think for Fast-K center algorithm
1) Choose any arbitrary center and define a radius 'r' for that node. Now
for that node prune the neighbor nodes which are far by (alpha)*r (can be
same as borkuva algorithm pruning method)
2) The value of alpha can be iterated depending on the value of K.

Please let me know whether I'm going in right direction or not.

Ankit Aggarwal

On Tue, May 2, 2017 at 1:06 AM, Ryan Curtin <ryan at ratml.org> wrote:

> On Tue, May 02, 2017 at 12:22:37AM +0530, Ankit Aggarwal wrote:
> > Hi Ryan
> >
> > I was going through the "Tree independent Dual tree Algorithm"
> paper(wrote
> > by you) which talks about the some of the algorithms.
> >
> > Can you explain how does the tight upper bound of k-nearest neighbor work
> > intuitively?
> >
> > "The basic problem is: given a set of points and integer k, choose k
> points
> > that minimize the maximum distance between a point in the set and its
> > nearest neighbor in the set"
> >
> > How are you thinking to extend it to Dual tree algorithms?
>
> Hi Ankit,
>
> The upper bounds that are used in dual-tree nearest neighbor search are
> fairly straightforward extensions of the triangle inequality.  The
> bounds are simply bounds on the furthest possible distance between a
> query point and its k-nearest neighbor.  If you'd like further
> clarification, you can consult the references of the paper or you can
> also read my Ph.D. thesis for a longer version.
>
> For developing the k-centers algorithm, the idea will center around
> determining when you can prune a point from a particular set.
>
> Realistically I would say that if you are not 100% familiar with
> dual-tree algorithms, it would be very difficult to quickly come up with
> a solution to this problem.
>
> Thanks,
>
> Ryan
>
> --
> Ryan Curtin    | "I'm just going to shoot you once!"
> ryan at ratml.org |   - Joseph Dunn
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://knife.lugatgt.org/pipermail/mlpack/attachments/20170507/7725ad46/attachment.html>


More information about the mlpack mailing list