help-octave
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: K means.


From: Søren Hauberg
Subject: Re: K means.
Date: Fri, 09 Dec 2011 08:00:17 +0100

fre, 09 12 2011 kl. 01:52 +0000, skrev Carnë Draug:
> On 8 December 2011 20:01, Søren Hauberg <address@hidden> wrote:
> > tor, 08 12 2011 kl. 14:35 -0500, skrev Ben Abbott:
> >> On Dec 8, 2011, at 2:04 AM, Søren Hauberg wrote:
> >>
> >> > ons, 07 12 2011 kl. 22:37 -0500, skrev Jordi Gutiérrez Hermoso:
> >> >> On 7 December 2011 19:52, Prachi Jain <address@hidden> wrote:
> >> >>> I was wondering is there any command by which we can find K means.
> >> >>
> >> >> We don't have such a function in Octave yet, but there are several
> >> >> implementations on the web. Would you like to help us get this
> >> >> function into Octave?
> >> >
> >> > Attached is an implementation I did some years ago; it's missing
> >> > documentation, but otherwise is in a reasonable state.
> >> >
> >> > Søren
> >>
> >> Soren,
> >>
> >> Is your implementation compatible with Matlab's?
> >>
> >>       http://www.mathworks.com/help/toolbox/stats/kmeans.html
> >
> > Not quite, it seems.
> >
> > My implementation requires an explicit initialisation, whereas Matlab
> > finds an initial set of clusters by sampling k random points. This
> > shouldn't be much work to implement, though.
> >
> > The Matlab version also has a bunch of options that my version does not
> > have. I don't think it's much work to implement these options, but I
> > don't have the time right now.
> >
> > Søren
> 
> Should we add it to the statistics package anyway? Other users can
> work on top of it if it is made available first. What do you think?

My current version works like

  clusters = kmeans (data, initial_clusters);

whereas Matlab's work like

  clusters = kmeans (data, number_of_clusters);

I think we should at least match this before putting it into the
statistics package. But otherwise, I see no problems with not having all
the options available that Matlab has.

Søren



reply via email to

[Prev in Thread] Current Thread [Next in Thread]