Stop k-means clustering

User 596 | 11/26/2014, 11:16:20 AM

Hello to everyone :smile: I'am running the k-means clustering using the k-means executable in release/toolkits/clustering. Since is several days that is running, but i did't set the maximum number of iteration, I'd like to know if there is some way to stop the algorithm and save the partial results obtained from the iteration. The best will be to obtain partial results, and let the iterations go on. There is some way to do that?

Thanks in advance for the help.

Comments

User 6 | 11/26/2014, 2:44:55 PM

As I often recommend here, it is much easier and recommended to use our newer GraphLab Create. You can find kmeans explanation here; http://graphlab.com/products/create/docs/generated/graphlab.kmeans.create.html

In PowerGraph, you can use the --max-iteration=XX flag to tell kmeans for how many iterations to run.


User 596 | 11/28/2014, 12:07:43 PM

I can't use GraphLabCreate because I've implemented the tanimoto distance measure on PowerGraph, and I need it for my clustering operation. I will wait until convergence. Thanks you very much for the help.


User 6 | 11/28/2014, 1:36:18 PM

We are working on an SDK where you could implement your own distance measure using Graphlab Create. Stay tuned!