Prev: Reading of Simulink block output value
Next: making a linear filter mask and apply to an image
From: Greg Heath on 25 Apr 2010 14:00
On Apr 25, 6:45 am, "Aadarsh " <imfromwa...(a)hotmail.com> wrote:
> Yes its
>
>  a problem with an implementation on file exchange. I have wrote a
> comment on the page where the file is, and hopefully the author will reply.
>
> Greg Heath <he...(a)alumni.brown.edu> wrote in message <ecfe8921-ade3-4b09-8041-b6e31b0aa...(a)x7g2000vbc.googlegroups.com>...
> > On Apr 24, 9:26 pm, Greg Heath <he...(a)alumni.brown.edu> wrote:
> > > On Apr 22, 9:01 am, "Aadarsh " <imfromwa...(a)hotmail.com> wrote:
>
> > > > If i have a n unknown points and n unknown clusters, is there any algorithm which will > try to group the points into the most appropriate groups.
>
> > > Did you really mean that you have the same number of clusters as you
> > > have points?
>
> > How many points do you have and what is their dimension?
>
> > Are there any restrictions on
> > 1. The maximum number of clusters
> > 2. The maximum number of members per cluster
> > 3. The maximum member to cluster center difference
> > 4. The maximum member to cluster root-mean-square distance.
>
> > Greg
>
> I am trying to cluster roughly 500 pairs of points - i.e. the data is 2 dimensional.
>
> Theoretically, there are no maximum number of clusters, although there definatley should not be the same number of clusters, as points to cluster - i know this because of how my algorithm works, prior to the clustering.
>
> There are no maximum number of members per cluster.
>
> There is no maximum member to cluster centre difference.
>
> I dont know what the maximum member to cluster root-mean-square distance.

These are all parameters that can be specified to keep
individual clusters from getting too large in volume and/or
density.

Since your data is only 2-D you can look at a visual
display and get a feel for practical limits on number
of clusters and one or more of the above parameters
as well a reasonable choices for intial center placement.

Then do one of the following
1. search for an algorithm that implements the
constraints you want
2. modify an existing algorithm to implement the
constraints you want
3. write an algorithm to implement the constraints
you want

I don't expect you will have to go as far as 3.

Hope this helps.

Greg

From: Greg Heath on 25 Apr 2010 14:38
On Apr 24, 3:41 pm, "Aadarsh " <imfromwa...(a)hotmail.com> wrote:
> "Ashish Uthama" <first.l...(a)mathworks.com> wrote in message
> >http://www.mathworks.com/matlabcentral/fileexchange/26431-qt-clusteri...
>
> Do you know how the algorithm works? Ive tried it several times
> but cant get it to work - have a look on the page, i left a comment.

The algorithm documentation:

input:
G-nxp data to cluster
d-diameter threshold
D-Euclidean distance for all 0<i<j<=n

Your code:

qtclusteuclid(new,5,4)

Observation:

"4" is not an upper triangular matrix containing
all of the between point positions.

Hope this helps.

Greg

First  |  Prev  | 
Pages: 1 2 3
Prev: Reading of Simulink block output value
Next: making a linear filter mask and apply to an image