High
i use pca as following but i cant see any changes in accuracy of my classifiers why?

a : labeld dataset
b: unlabeld dataset

[tr,test]=gendat(a,.5);
w1=pca([],.9);
w2=bpxnc([],5);
w3=a*(w1*w2); trained map
bb=b*w3*classc;
lab1=bb*labeld ;

if i want use w2=emclust(?,nmc,5)
how can i write?

thankyou

Depends on your data.

emclust is not an alternative for pca as it is a cluster routine and not
a feature reduction. Moreover, the first outpur parameter are the labels and
not a mapping.

Bob Duin

hello BobDuin
first i give thanks you.then
I mean about Emclust is ([d,w]=emclust(...));
and I want to use Emclust like as bpxnc for classifing of dataset, not instead of PCA
now befor Emclust how i can apply PCA?

another Question is the relationship between Emlust and Kmeans?
is Emclust(a,nmc,2) equivalent kmeans(a,2) ?
Best regard

Yes, sorry I misunderstood you, emclust can be used as a classifier,
in fact a semi-supervised classifier. I am not sure how good it is, but
you can try. In case all objects are labeled you can also consider to
use mogc, wich applies the em algorithm for finding a mixture of gaussians
per class.

emclust is a generalizationof kmeans. emclust(a,nmc,2) is thereby equivalent
to kmeans(a,2). I am not sure whether it is identical.

Bob Duin