SVM Classification:

num=ceil(5000*rand);
traindat=[rand(2,num)-0.2 rand(2,num)+0.2];
trainlab=[-ones(1,num) ones(1,num)];

sg('set_features', 'TRAIN', traindat);
sg('set_labels', 'TRAIN', trainlab);
sg('send_command', 'set_kernel GAUSSIAN REAL 10 0.1');
sg('send_command', 'init_kernel TRAIN');
sg('send_command', 'new_svm LIBSVM');
sg('send_command','svm_epsilon 1e-5')
sg('send_command', 'c 10');
sg('send_command', 'svm_train');

mi=min(traindat');
ma=max(traindat');
[x,y]=meshgrid(linspace(mi(1),ma(1),50), linspace(mi(2),ma(2),50));
testdat=[x(:),y(:)]';
sg('set_features', 'TEST', testdat);
sg('send_command', 'init_kernel TEST');
out=sg('svm_classify');

figure(1)
clf
out=reshape(out,50,50);
surf(out)

SVR Regression - see Matlab examples (the very same example works)

Hidden Markov Model - very similar to the Matlab version (cf. Matlab examples):

clear all;
leng=50;
rep=5;
weight=1;
num_hmms=30;

% generate a sequence with characters 1-6 drawn from 3 loaded cubes
for i = 1:3,
    a{i}= [  ones(1,ceil(leng*rand)), ...
            2*ones(1,ceil(leng*rand)), ...
            3*ones(1,ceil(leng*rand)), ...
            4*ones(1,ceil(leng*rand)), ...
            5*ones(1,ceil(leng*rand)), ...
            6*ones(1,ceil(leng*rand)) ];
    a{i}= a{i}(randperm(length(a{i})));
end

s=[];
for i = 1:size(a,2),
    s= [ s i*ones(1,ceil(rep*rand)) ];
end
s=s(randperm(length(s)));
sequence='';
for i = 1:length(s),
    rn=rand;
    f(i)=ceil(((1-weight)*rand+weight)*length(a{s(i)}));
    t=randperm(length(a{s(i)}));
    r=a{s(i)}(t(1:f(i)));
    sequence=[sequence char(r+'0')];
end

% train 30 hmms on sequence with pseudocount 1e-10
sg('send_command','loglevel ERROR');
sg('send_command','pseudo 1e-10');
hmms=[];
liks=[];
for i=1:num_hmms,
  sg('send_command','new_hmm 3 6');
  sg('set_features','TRAIN',sequence','CUBE');
  sg('send_command', 'convert TRAIN SIMPLE CHAR STRING CHAR');
  sg('send_command', 'convert TRAIN STRING CHAR STRING WORD 1');
  sg('send_command', 'bw');
  [hmms(i).p, hmms(i).q, hmms(i).a, hmms(i).b]=sg('get_hmm');
  hmms(i).lik=sg('hmm_likelihood');
end

% take the most likely one
[v,idx]=max([hmms.lik]);
sg('set_hmm', hmms(idx).p,hmms(idx).q,hmms(idx).a,hmms(idx).b);

% compute viterbi path
sg('set_features','TEST',sequence','CUBE');
sg('send_command', 'convert TEST SIMPLE CHAR STRING CHAR');
sg('send_command', 'convert TEST STRING CHAR STRING WORD 1');
[path,lik]=sg('get_viterbi_path',0);
path=path+1; %path is zero based in shogun but one based in matlab

% fancy output when which cube was estimated to be drawn
f_est=[]; s_est=[]; f_est=1; s_est=path(1); i=1;

for j=2:length(path),
  if s_est(i)==path(j)
      f_est(i)=f_est(i)+1;
  else
      i=i+1;
      f_est = [ f_est 1 ];
      s_est = [ s_est path(j) ];
  end
end

fprintf('estimated:\n');
fprintf('========\n\n');
fprintf('sequence:\n');
for i = 1:length(s_est),
    fprintf('\tcube: %dx%d\n',f_est(i), s_est(i));
end
for i= 1:size(a,2),
    fprintf('Estimated Distribution: Dice %d:', i);
    fprintf('%f ', exp(hmms(idx).b(i,:)));
    fprintf('\n');
end

fprintf('solution:\n');
fprintf('========\n\n');
fprintf('sequence:\n');
for i = 1:length(s),
    fprintf('\tcube: %dx%d\n',f(i), s(i));
end
for i= 1:size(a,2),
    fprintf('Distribution: Cube %d:', i);
  nn=hist(a{i},1:6)/length(a{i});
    fprintf('%f ', nn);
    fprintf('\n');
end

%plot the segmentation
truepath=[];
for i = 1:length(s),
  truepath = [truepath ones(1,f(i))*s(i)];
end
figure(1);
clf
plot(path,'r-')
hold on
plot(truepath+0.02,'b-')
legend('estimated path', 'true path');