a = prnist([0:9],[1:10:1000])
preproc = im_box([],0,1)*im_rotate*im_resize([],[20 20])*im_box([],1,0);
obj=a*preproc
obj1=seldat(obj);
nistmeas11=dataset(im_mean(obj1))
nistmeas22=dataset(im_measure(obj1,obj1,{'size','perimeter'}))
nistmeas33 = dataset(im_moments(obj1,'hu'))
nistmeas44 = dataset(im_harris(obj1,4))
nistdata=[nistmeas11 nistmeas22 nistmeas33 nistmeas44]

%%
b = nistdata*scalem(nistdata,'variance');
[trn,tst] = gendat(b,0.5)
[w,list] = pca(trn,2);
j=feateval(trn,'NN',tst)
figure (3);scatterd(trn*w)

%%
[trn,tst] = gendat(nistdata,0.5)
w = fisherm(trn,2);
j=feateval(trn,'NN',tst)
figure (4);scatterd(trn*w)
%%
[trn,tst] = gendat(nistdata,0.5)
[w,list] = featself(trn,'NN',2);
j=feateval(trn,'NN',tst)
figure (5);scatterd(trn*w)
%% best features from featself
nistdat2=nistdata(:,[1 4])
%%
w = fisherm(nistdata,8);
nistdat8=nistdata*w
[train8,test8] = gendat(nistdat8,0.5)


error = zeros(1,8) % make space for errors
for dim = 1:8 % run over all dimensionalities
w = ldc(train8(:,[1:dim])); % train a classifier
error(dim) = testc(test8(:,[1:dim]),w); % test the classifier, store error
end
figure(6);plot(error)