Hey all,
does the node for k-means have a way to display the Silhouttes of the Clusters and the Model itself?
Regards,
Robert
Hi Robert,
Not directly, but you can use the color manager to visualize the results.
Regards,
Aaron
Hello Aron,
thanks a lot. I know that. But I'm trying to compare a couple of algorithms inbetween data mining tools and not having the silhouette or any values like the average distance to calculate it, makes it hard to do so.
Sure the vizualisation can show me the clusters, but it's far from having the one value in clustering directly giving me an indication of the models and the clusters quality. Unfortunately it seems this issue has been unresolved for about four years now (I've seen another post about this from 2011), so I doubt it will ever get properly implemented. Which is really a pity, since the results of the k-means node seem to be just as good in Knime as they are in other tools which cost a significant amount of money.
Regards,
Robert
Znakebite,
I don't even think this should be implemented in the K-means node. I think this functionality should be in their own set of nodes. One node that can transform any suitable data into a set of silhouettes, one or more for statistics, and maybe even one or more dedicated viewing nodes. This way, every clustering node is "extended" in one go. And with some clever tricks, this functionality could be exploited for some other crazy stunts, too.
You know what that also means? This means this is the perfect opportunity for a community extension! If you really want this functionality, you could implement it yourself, or hire a freelancer to do it for you. Should I mention that I'm a freelancer with a love for geometric computations and Knime? ;)