Dear Karsten,
On Sunday 12 March 2006 21:11, K. Weinert wrote:
> Hello Jonas,
>
> the code in the RC 0.2 branch seems to work fine. Set it free!
>
> Considering the numerical problems in the higher-dimensional case in the
> soft-k-means algorithm, I'm afraid I cannot contribute anything. Can you
> point me to a paper or textbook describing the algorithm? I'm interested
> in learning about it...
If you can get your hands onto "Information Theory, Inference and Learning
Algorithms" by David J.C. MacKay, take a look at it! This is the book that we
read in a university course. I think it is quite nice read - I even bought it
myself. It is also mentioned in our CREDITS file.
Chapter 20 details the k-means algorithm. Soft k-means is in 20.2 and what we
call mixture models in out implementation is found in 22.3 as "Enhancements
to soft K-Means". The problems that we run into are detailed in 22.4 "A fatal
flaw of maximum likelihood".
Regards, Jonas.
|
