k-means empty cluster

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

k-means empty cluster

codierknecht
Hi,

i want to use the k-means implementation included in the pcl. However, it
runs almost always into empty clusters with centroids set to (inf, inf,
inf). Is this a known problem or am I using it wrong?
I'm also wondereing why the centroids are initialized with [0, 0, 0] in
initialClusterPoints()-function of kmeans.cpp.

I hope someone has an idea what to do or experience with the usage of this
implementation.

Thanks in advance








--
Sent from: http://www.pcl-users.org/
_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users
Reply | Threaded
Open this post in threaded view
|

Re: k-means empty cluster

Sérgio Agostinho
The ml module is not stable, to the point that it has it even has non
implemented functionalities.

Cheers



_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users

signature.asc (836 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: k-means empty cluster

codierknecht
Thanks for the fast response. I will have this in mind when I'll use it. Is
there a version or date that is expected to get a stable ml module?

The kmeans-class performs a clustering almost as expected, but often
produces these empty clusters. I think it could be a problem with the
initial centroids. I already added a few things to the class, like several
clustering attempts to result in the most compact clustering and a
interchangeable distance-function. I'll also try other initialization
strategies and plan to provide an implementation here, that works for me.
Presupposed I'll be able produce one

Kind regards



--
Sent from: http://www.pcl-users.org/
_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users
Reply | Threaded
Open this post in threaded view
|

Re: k-means empty cluster

Sérgio Agostinho


On 29-01-2018 08:20, codierknecht wrote:
> Thanks for the fast response. I will have this in mind when I'll use it. Is
> there a version or date that is expected to get a stable ml module?

There's no one actively developing it, therefore there's no progress
towards stabilizing it.
> The kmeans-class performs a clustering almost as expected, but often
> produces these empty clusters. I think it could be a problem with the
> initial centroids. I already added a few things to the class, like several
> clustering attempts to result in the most compact clustering and a
> interchangeable distance-function. I'll also try other initialization
> strategies and plan to provide an implementation here, that works for me.
> Presupposed I'll be able produce one
That would be appreciated.

Cheers



_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users

signature.asc (836 bytes) Download Attachment