Get denormalized k-Means centroids

gulyasg

Hi everybody,

my problem is the following. I read a database, and then I normalize all attributes because they are not in the same scale.
For this modified example set I run a k-Means clustering and I get the cluster labeled example set and the centroid table.
But the centroids are also normalized, so I can't use them to characterize the original clusters. The "De-normalize" operator
works just for ExampleSets, and not for ClusterModel.

What can I do to get the centroids so that the attributes are in their original scales?

Thanks,
Gergely Gulyas

earmijo

Suppose you have a process like the following:

Read Data  =>  Normalize  =>  K-Means

1) Now connect the port "Preprocess Model" from the operator Normalize to the Operator "DeNormalize".

2) Connect the Example Set coming out of the K-means operator and the Model coming out of the DeNormalize operator to an Operator ApplyModel.

3) Use the operator Aggregate to obtain the means grouped by Cluster.

Hope this helps,

E.



                                                   

earmijo

You can watch a video tutorial (in Spanish though) here:

http://www.screencast.com/t/1HO47i0w2y00

gulyasg

Hi,

if I apply de-normalization for the Example Set it does not transform the cluster table of the k-Means. I do not understand exactly the third step, I will be watching your video...

Thanks,
G.

earmijo wrote:

Suppose you have a process like the following: Read Data  =>  Normalize   =>   K-Means 1) Now connect the port "Preprocess Model" from the operator Normalize to the Operator "DeNormalize". 2) Connect the Example Set coming out of the K-means operator and the Model coming out of the DeNormalize operator to an Operator ApplyModel. 3) Use the operator Aggregate to obtain the means grouped by Cluster. Hope this helps, E.

Unknown

Hello

You might find the answer here...

http://rapid-i.com/rapidforum/index.php/topic,3686.0.html

Andrew