Power K-Means
Data Skeptic
00:00
Clustering Context
When you reach a local optimum of the objective function, you simply come a solution that is sort of better than other alternatives in some vicinity. But there could be another one kind of far away which would be even better. Getting out of one and jumping to a different one involves maybe changing the larger scale configuration of your clusters. From different starting points, i could potentially arrive at wildely different clustering results, depending on my data set. Are you going to help me get a little bit more lucky initially, or do you help me evolve it along way? So we are going to do the latter, we're going to try to evolve it along the way.
Transcript
Play full episode
