Orange Forum • View topic - K-Means initialization

K-Means initialization

A place to ask questions about methods in Orange and how they are used and other general support.

K-Means initialization

Postby flolli » Fri Oct 12, 2012 17:03

Hello,
I'm looking for some discussion on agglomerative and diversity initialization for K-Means algorithm because they are not clear in my mind. I understand that the final solution of the algorithm is affected by the initial clustering, so I set the number of restarts to the maximum (i.e.100) in order to report the best solution achieved. However, the solution still depends on the initialization, i.e. random, agglomerative and diversity. How to make the solution as robust as possible?
Many thanks,
Francesco

Re: K-Means initialization

Postby flolli » Tue Oct 16, 2012 11:05

In other words, it is sufficient to set at 100 the number of restarts in order to obtain a robust solution?


Return to Questions & Support