k-means clustering

How do i use Inter-class factor and Intra class factor in k-means algorithm to find out the number of clusters which can be formed in the given data.

Answer 1

[Walter Roberson]

The maximum number of clusters that can be formed is equal to the number of points.

You are again trying to determine the number of clusters that you "should" use for a blind data source, and that is something that simply cannot be done without deciding ahead of time which measurements and parameters you will use to decide whether a certain number of clusters is "better" than a different number of clusters.

k-means has absolutely no idea of what the "right" number of clusters is: k-means is strictly an algorithm that deals with finding the best clustering given a specific number of clusters ahead of time. Comparing results between different numbers of clusters is something that there is no one "right" answer for: it depends upon your purposes and your definitions.

[ramya]

How can i find Timing analysis for the given data .The data given to me is 1*20480. Now for this given data i have to indentify the number of neurons by forming clusters and also have to find timing analysis. k-means algorithm is been used to form the clusters but i dont know the value of "k"..i have run the the code by assuming k=5 ...

[Oleg Komarov]

You can assume a trial and error approach. Use different k, say k = 2:20, and run your analysis. I would generally try to identify the k which keeps results robust to cluster ordering and composition.

[Walter Roberson]

If you were to generate the timing analyses for several different k, then is there some means by which you could determine which was a "better" timing analysis?