Cut dendogram at a certain similarity

danieldaniel Member Posts: 12 Contributor II
edited November 2018 in Help
Hello everyone,

I am in the process of evaluating some data and I wanted to compare different types of clustering. I found the 'Map Clustering on Labels' operator that lets me compare the results of the clustering with my sample data. This all works fine but I am concerned about the following scenario.

Let's say I have an example set that has examples from three different classes. I realize that using kmeans with an initial number of 2 expected clusters will obviously end up with two result clusters and probably with a higher failure in the performance evaluation. However the agglomerative clustering has the potential at performing better, as the number of clusters doesn't have to be known before the process. To make this work the resulting dendogram has to be cut at a certain point of similarity.

I have searched the list of operators but have only found the 'Flatten Clustering' operator that lets me choose the number of clusters. Searching the forums I have found this thread from 2010: http://rapid-i.com/rapidforum/index.php/topic,1734.0.html

Has there been any progress on this since 2010? Am I missing an operator to achieve this? Do I have to install a plugin?

Thanks in advance for your help,

Daniel

Answers

  • amanahR_miamanahR_mi Member Posts: 1 Contributor I
    So, anyone know the answer already?
  • marcin_blachnikmarcin_blachnik Member Posts: 55  Maven
    You can find it in extension called:
    Instance selection and Prototype based rules

    This operator is available in ISPR -> Clustering and is called Flatten Clusters By Distance. This operator has a parameter which is a distance/similarity threshold.
    As an input it requires the model of Agglomerative Clustring and an ExampleSet

    In case of any problems please contact me.

    Best
    Marcin
Sign In or Register to comment.