isrish / vmm Goto Github PK

I think that according to "Clustering on the unit hypersphere using von Mises-Fisher distributions" the update of rbar in Maximization function missing a scaling of N.

I suspect that
rbar = normMu/W(k);
shoud be
rbar = normMu/(W(k)*N);
according to Algorithm 1 in the referred paper.

Hi! I just started using your code and it's working great for the most part on my data! Thanks for sharing.

I just have a quick question - how can I interpret the obj.Kappa value coming from fitVMM_CEM? For my data, it is a value in the 10s to 100s. I expected 0<=obj.Kappa<=1 based on the von Mises Fisher distribution. Is there a scaling factor or some other way to interpret this value? Any help on this would be great!

Edit
After looking at it some more, it seems maybe if I divide by 1,000, I'll get the result I'm after? Still unsure though.

I'm a little bit confused
In obj = fitVMM_EM(X,Kmax,varargin)
data is in radian or in degree.

and it should be spherical but its values vary between -1 to 1, indicating that it is in cartesian.