Open cimentadaj opened 4 years ago
Students asked for this. And I found that the idea of subjectively choosing the k
parameter made the method unattractive from a substantive point of view. I was thinking that aside from adding a section on hierarchichal clustering, it would be very handy to workout some strategies combining both things. That is, taking out a real world example, use HC for identifying clusters, cross-check this with k-means clustering and also with k-nearest neighbor, always using theoretical descriptions of the problem to assess whether the clusters make sense.
You didn't explain hierarchical clustering because there isn't enough time to teach it at the course. We still need a good section on the topic with a real world example.