Prev: Reading in Text files
Next: pcolor not working on my PC only.
From: murat on 22 Jan 2010 12:19
Hello everyone,
These days are so hard for me. :) Climate change changes me every point of me. :)
Anyway, now i have a problem, i am already problem for life may be. :)

Now, let us think a space or matrix which is dimension is infinite.
And, datas have a similarity a bit or too close. Now, in my mind, optimal number of cluster in fuzzyclusteringtoolbox, is it work for that? or what is your opinion?
Another approach, are there any of significant cluster? How can we obtain clusters?
Thanks for sharing your mind! :)
From: murat on 23 Jan 2010 04:47
"murat " <hungocan(a)hotmail.com> wrote in message <hjcmm9$oa4$1(a)fred.mathworks.com>...
> Hello everyone,
> These days are so hard for me. :) Climate change changes me every point of me. :)
> Anyway, now i have a problem, i am already problem for life may be. :)
>
> Now, let us think a space or matrix which is dimension is infinite.
> And, datas have a similarity a bit or too close. Now, in my mind, optimal number of cluster in fuzzyclusteringtoolbox, is it work for that? or what is your opinion?
> Another approach, are there any of significant cluster? How can we obtain clusters?
> Thanks for sharing your mind! :)


I've been waiting for your helping...
From: Walter Roberson on 23 Jan 2010 13:05
murat wrote:

> Now, let us think a space or matrix which is dimension is infinite. And,
> datas have a similarity a bit or too close. Now, in my mind, optimal
> number of cluster in fuzzyclusteringtoolbox, is it work for that? or
> what is your opinion?

My opinion is that there is no way in Matlab to represent a matrix which
is infinite in any dimension, and thus that there is no way to create
such a matrix to pass it to the fuzzy clustering toolbox.

> Another approach, are there any of significant cluster? How can we
> obtain clusters?


The "optimal" number of clusters in any problem is the same as the
number of points involved, if you measure "optimal" by Euclidean
distance to the cluster centroids. If you are expecting a smaller
number, then your definition of "optimal" needs to be explicitly stated
and has to include a penalty that increases as the number of clusters
increases.

Similarly you have not given us any definition of "significant". I could
say that every possible cluster is significant -- or that none are --
and still be within the bounds of your question. But considering that
your matrix is infinite, unless you adopt the definition that NO cluster
will be significant, it will be impossible to determine the
"significant" clusters of the infinite matrix in any finite amount of time.

"Infinity is like a stuffed walrus I can hold in the palm of my hand.
Don't do anything with infinity you wouldn't do with a stuffed walrus."
-- Dr. Fletcher, Va. Polytechnic Inst. and St. Univ.
 | 
Pages: 1
Prev: Reading in Text files
Next: pcolor not working on my PC only.