I would like to run a K -means loop where I have to use the following nodes:

- Partitioning


-K means node

-Cluster Assigner


I would like to know when in the loop I have to do the partitioning and to which other node I have to connect thereafter. My second question is about the interpretation of the cluster model. Is there a way to get information about the means of every centroids computed by the K-means node and information about the sum of error squares for each cluster ?

By the way, can you tell me when do I have to use the normalizer. I think it is important to use the denormalizer once the cluster centers have been found to interprete what means each cluster.

Thanks in advance for your help



Where to put Partitioning? It depends on what you want to achieve. If you want to use the same partinioning with different parameters for k-Means, put it outside the loop, if you want different test sets (like in cross-validation) put it inside. (But in that case it might be a better idea to use the nodes designed for that purpose.)

Normalizing before k-Means seems to be a good idea (I assume you want to use the validation set also normalized with the same normalization settings). I am not sure why you would need to denormalize the model unless you want to (human-)interpret the model in the original context.

Cheers, gabor


I read that it was crucial to normalize data before using the k-means node. I wanted to denormalize the data with the same normalization settings to interpret the results of the k-means node, to extract meaningful information.

Regarding the partitioning, I used the partitioning node. To be honest with you, i do not understand how the partioning node works. By looking at my meta node in the following link https://www.dropbox.com/s/eks9byw6gh2amlq/Clustering%20Algorithm.JPG?m , could you tell me where I can use the partitioning node or the X-partitioner/X-aggregator in the meta node that I constructed?


Thank aborg for your help

Sorry, I was busy recently, but I am afraid there is no easy way to denormalize the k-Means model. Maybe others have more experience in this regard.

PS.: It seems the image is no longer available.