| Computers Forum Index » Computer Artificial Intelligence - Neural Nets » Self Organizing Maps's->is it possibke ?... |
|
Page 1 of 1 |
|
| Author |
Message |
| tguclu... |
 Posted: Fri Nov 06, 2009 10:02 am |
|
|
|
Guest
|
Hi
Summary: I'm trying to cluster of some images depending on the angles
between body parts.
It's like angle1:torso-torso angle2:torso-upper left arm ..
angle10:torso-lower right foot
Facts: Input data:1057x10 ,1057 stands for number of image.10 stands
for angles of body parts with torso. TestSet:821x10
I want all the rows in input data to be clustered with 88 clusters.
Then I will use these clusters to find which clusters does TestData
fall into ?
Previous work I did it in k-means which is very straight forward. Just
ask k-means to cluster data to 88 clusters. And implement another
method that calculates the distance between each row in test data and
center of each cluster and pick the smallest values.This is the
cluster of corresponding input data row.
Question My question is: (1) Is it possible to do it with SOM ? AFAIK
SOM's are for visual clustering.But I need the know the character of
each cluster so that I can later calculate whic cluster that my test
data row will hold.
(2)Do you have a better solution ?
Thanks.. |
|
|
| Back to top |
|
|
|
| Cagdas Ozgenc... |
| Posted: Fri Nov 06, 2009 12:18 pm |
|
|
|
Guest
|
I think 88 clusters are too many for a data set of 1057 elements. It
means 12 samples per cluster on average.
My experience is to match SOM topology dimension with the dimension of
the data set. In this case it is 10. With such a high dimension I
don't think that topology will properly converge to anything useful.
SOM nodes also move to cluster centers. So you can use euclidean
distance for membership test.
My recommendation is to use k-means with a lower number of clusters. I
like SOM with 2-3 dimensions where I can actually see the topology
visually. Sometimes the mesh twist on itself and don't converge
properly. For higher dimensions you won't even know that this
happened.
On Nov 6, 1:02 pm, tguclu <tugrul.gu... at (no spam) gmail.com> wrote:
Quote: Hi
Summary: I'm trying to cluster of some images depending on the angles
between body parts.
It's like angle1:torso-torso angle2:torso-upper left arm ..
angle10:torso-lower right foot
Facts: Input data:1057x10 ,1057 stands for number of image.10 stands
for angles of body parts with torso. TestSet:821x10
I want all the rows in input data to be clustered with 88 clusters.
Then I will use these clusters to find which clusters does TestData
fall into ?
Previous work I did it in k-means which is very straight forward. Just
ask k-means to cluster data to 88 clusters. And implement another
method that calculates the distance between each row in test data and
center of each cluster and pick the smallest values.This is the
cluster of corresponding input data row.
Question My question is: (1) Is it possible to do it with SOM ? AFAIK
SOM's are for visual clustering.But I need the know the character of
each cluster so that I can later calculate whic cluster that my test
data row will hold.
(2)Do you have a better solution ?
Thanks.. |
|
|
| Back to top |
|
|
|
|
