Modifying kmeans algo

[Ankit Goel <an...@gmail.com>]

Hi all,
If one wanted to modify the kmeans algorithm given with the mahout package,
how would/should one go about doing it?

Also what is the function that can be used to find the median point between
2 or more vectors? As in I want the median point in vector format so that I
can use it as a new center maybe.

Thanks.

-- 
Regards,
Ankit Goel
http://about.me/ankitgoel

Re: Modifying kmeans algo

[Ted Dunning <te...@gmail.com>]

On Tue, Sep 22, 2015 at 5:51 PM, Ankit Goel <an...@gmail.com> wrote:

> What I wanted to do was modify the clustering algorithm, in  hopes of
> experimenting with different versions of it. I'm not much hung over the MR
> part of things, rather the clustering algo itself.
>

Have at it.

All yours.


> Secondly or alternatively, I wanted to which part or library calculates the
> centroid. (In hopes of reconfiguring when the algo splits the clusters,
> increasing the number of clusters and thus calculating a new centroid).
>

I think it was just simple vector addition. Nothing fancy at all.

Re: Modifying kmeans algo

[Ankit Goel <an...@gmail.com>]

Hi,
What I wanted to do was modify the clustering algorithm, in  hopes of
experimenting with different versions of it. I'm not much hung over the MR
part of things, rather the clustering algo itself.

Secondly or alternatively, I wanted to which part or library calculates the
centroid. (In hopes of reconfiguring when the algo splits the clusters,
increasing the number of clusters and thus calculating a new centroid).

On Tue, Sep 22, 2015 at 6:09 AM, Ted Dunning <te...@gmail.com> wrote:

> On Mon, Sep 21, 2015 at 4:44 PM, Ankit Goel <an...@gmail.com>
> wrote:
>
> > If one wanted to modify the kmeans algorithm given with the mahout
> package,
> > how would/should one go about doing it?
> >
>
> If you want to modify the old map reduce code, please go right ahead.  The
> project members will not be maintaining that code going forward, however,
> so that modification will be all yours.
>
>
> >
> > Also what is the function that can be used to find the median point
> between
> > 2 or more vectors? As in I want the median point in vector format so
> that I
> > can use it as a new center maybe.
> >
>
> It sounds like you want to compute the medoid of several vectors [1] or
> possibly the geometric median [2]. Neither is particularly easy to compute
> and Mahout supports neither.
>
> You may also have wanted the vector mean.  That is trivial to compute ...
> just add up the vectors and divide by the number of vectors.
>
>
> [1] https://en.wikipedia.org/wiki/Medoid
>
> [2] https://en.wikipedia.org/wiki/Geometric_median
>



-- 
Regards,
Ankit Goel
http://about.me/ankitgoel

Re: Modifying kmeans algo

[Ted Dunning <te...@gmail.com>]

On Mon, Sep 21, 2015 at 4:44 PM, Ankit Goel <an...@gmail.com> wrote:

> If one wanted to modify the kmeans algorithm given with the mahout package,
> how would/should one go about doing it?
>

If you want to modify the old map reduce code, please go right ahead.  The
project members will not be maintaining that code going forward, however,
so that modification will be all yours.


>
> Also what is the function that can be used to find the median point between
> 2 or more vectors? As in I want the median point in vector format so that I
> can use it as a new center maybe.
>

It sounds like you want to compute the medoid of several vectors [1] or
possibly the geometric median [2]. Neither is particularly easy to compute
and Mahout supports neither.

You may also have wanted the vector mean.  That is trivial to compute ...
just add up the vectors and divide by the number of vectors.


[1] https://en.wikipedia.org/wiki/Medoid

[2] https://en.wikipedia.org/wiki/Geometric_median