k-Means convergence

[rmx <ru...@hotmail.com>]

Hi, just a basic question.

I am using  Mahout k-Means with the same initial centroids file, and enough
iterations to converge.
I was expecting the algorithm would produce the same results every time I
run it over the same dataset with the same settings.
Most of the times it does. But sometimes it doesn't.
I am running the experiments with an EC2 10 machines cluster.

Is this behavior expected with Mahout k-means?
Or is it something wrong?

Thanks
-- 
View this message in context: http://lucene.472066.n3.nabble.com/k-Means-convergence-tp1965749p1965749.html
Sent from the Mahout User List mailing list archive at Nabble.com.

Re: k-Means convergence

[Ted Dunning <te...@gmail.com>]

Starting conditions can affect the results of k-means so this is probably
expected behavior.

On Thu, Nov 25, 2010 at 1:06 AM, rmx <ru...@hotmail.com> wrote:

> Is this behavior expected with Mahout k-means?
> Or is it something wrong?
>