This question already has an answer here:

There are different methods which are proposed for initialization of K means, but is there any literature that lists the merits and demerits of each one.(some sort of survey)

Most popular one is i guess k-means++ but why is it better than farthest point method ?


marked as duplicate by ttnphns, Michael Chernick, kjetil b halvorsen, Peter Flom Jun 7 '18 at 11:05

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.

  • $\begingroup$ author himself has said this "I'm not discussing presently which method is "better" and in what circumstance" which is exactly what i am looking for and specifically in the case of farthest point vs k means ++ $\endgroup$ – Siddharth Shakya Jun 6 '18 at 17:20

I believe I have seen such surveys.

K-means++ is better than farthest points if you want to do more than a single run. Farthest points tends to produce almost the same initial conditions every time. K-means++ is well randomized, so if you run it 10 times, you have a better chance of getting a good result at least once.

Among the best initializations are those that sample a subset and cluster the subset, such as Bradley and Fayyad.


Not the answer you're looking for? Browse other questions tagged or ask your own question.