Looking for optimal k in K-Medoids using the Cluster Distance Performance operator

tonyboy9

which is supposed to produce the Davies-Bouldin index for k in K-Medoids.

https://docs.rapidminer.com/9.1/studio/operators/validation/performance/segmentation/cluster_distance_performance.html

Used the Tutorial to try to duplicate the process. My results in sequence are:

1 Process looking for k produced by the Davies-Bouldin method.

Result Folder view

Result Graph



Result Centroid table



Result Annotations



AutoModel Results k-means-summary. K-medoids is not a model choice.



What am I missing here looking for the Davies-Bouldin method?

Caperez

Hi @tonyboy9, 

I understand that you miss the Davies-Bouldin Index in Automodel clustering models. You can apply the Cluster Distance Performance Operator to the resulting  model from Automodel.  I hope that can help you.

Best

tonyboy9

Many thanks to ceaperez for suggesting I try the Cluster Distance Performance Operator.
I used the steps in the tutorial.



This was my first clue Davies-Boudin could be found.



I used the tutorial to set up the process in RapidMiner Studio.



After I ran the above process, on Cluster Model I clicked on PerformanceVector.



On this result, Davies-Bouldin is the choice in the left bar.



And here you have it, Davies-Bouldin. It's only a start, now that I understand how to play with the k parameter inside the operator.

Caperez

Hi @tonyboy9,

The K-means is a very useful and common used technique.

Both K-Means and K-medoids are partitional methods but the K-means is centered into minimize the total square error and the K-medoids is centered into minimizes the sum of dissimilarities between points.

 Best

Marco_Barradas

Hi @tonyboy9,

Maybe this thread could help you understand and find that optimal number.

https://community.rapidminer.com/discussion/comment/61654#Comment_61654

Caperez

Hi @tonyboy9, 

I understand that you miss the Davies-Bouldin Index in Automodel clustering models. You can apply the Cluster Distance Performance Operator to the resulting  model from Automodel.  I hope that can help you.

Best

tonyboy9

Many thanks to ceaperez for suggesting I try the Cluster Distance Performance Operator.
I used the steps in the tutorial.



This was my first clue Davies-Boudin could be found.



I used the tutorial to set up the process in RapidMiner Studio.



After I ran the above process, on Cluster Model I clicked on PerformanceVector.



On this result, Davies-Bouldin is the choice in the left bar.



And here you have it, Davies-Bouldin. It's only a start, now that I understand how to play with the k parameter inside the operator.