Hierarchical Clustering - Extract Distances/Clusters for each step

btibert

Is it possible to extract the height/distance added at each step of a hierarchical clustering step?  Ideal state is the number of clusters and height/distance added at each step.  It would be helpful to be able to extract this, somehow, as an ExampleSet to explore where the cut should be relative to the balance distance/clusters.

It is 100% possible that these tools exist, but they are not jumping out to me.  When combining the Agglomerative Clustering and Flatten Cluster operator, I am not sure where I can extract information to make a data-driven decision on how many clusters to extract.

MartinLiebig

Hi,

to the best of my knowledge it is not possible to get an extract this with operators.

br,

martin

sgenzer

cc @mschmitz @yyhuang

MartinLiebig

Hi,

to the best of my knowledge it is not possible to get an extract this with operators.

br,

martin

btibert

Thanks everyone.  Short of fitting the model and associating the clusters in this context, any other suggestions on how to "review" or "validate" cluster assignments? Validating KMeans is straight forward, but wondering about hierarchical methods in a teaching context.

MartinLiebig

https://towardsdatascience.com/understanding-clustering-cf0117148ef4#b7ae is what i like to do, and this is fairly method-independend.

btibert

Thanks, I love showing my students how to profile with Trees (something that we already covered), but I was hoping that there was something I could show them around how to select the cut in a more data-driven way.  I can totally make it work with smaller datasets to make the intuition easier to grasp, but I just wanted to make sure I wasn't missing any operators along the way.