Nesting parameter optimization within counting loops

[acito]

Hello,
RunSilhouetteInLoop.knwf (40.7 KB)

I am trying to find the optimum value of the silhouette coefficient with random starting centroids in k-means clustering. The idea is to explore the stability of the number of clusters with different centroids. As soon as I add the outer Counting Loop, the k-means node throws an error. Can this not be done?
Frank

[ScottF]

Hi @acito -

I downloaded your workflow and tried it out. It seems there is a conflict between workflow variables with identical names being introduced by the two loop start nodes - they both want to create a flow variable called Loop (0).

After playing around a bit, I was able to get it to work by adding a flow variable connector between the Normalizer and Parameter Optimization Loop Start nodes, like this:

This forces the second loop start node to create a flow variable called Loop (1), since the Loop (0) variable already exists on its branch. And then there’s no conflict for it to complain about. A bit hacky, but it works.

This is maybe more motivation for us to have a “Variable Filter” node or similar, to more easily resolve these types of conflicts in more complicated workflow arrangements

Anyway, hope it helps!

[acito]

Scott,
Thanks so much for your quick response. The analysis now runs perfectly.
The purpose of this workflow is to show the effects of different random starts when using k-means. The number of clusters reported by the Silhouette coefficient varied between 5 and 11.
Frank