function [pred_label,pred_value,ind_value]=svm_manifold_predict(testing_com, training_com, training_label, list, sigma, gama, model)

%get the number of subjects
  [tt_n,m,v]=size(testing_com);
  [tr_n,m,v]=size(training_com);

%get the number of components of each individual
  m=length(list);

%normalizing
 for i=1:tt_n
    if(m==1)
      tmp=(squeeze(testing_com(i,list,:)));
      tt_ncom(i,:)=tmp*(tmp'*tmp)^(-1/2);
    else
      tmp=(squeeze(testing_com(i,list,:)))';
      tt_ncom(i,:,:)=tmp*(tmp'*tmp)^(-1/2);
    end
 end

 for i=1:tr_n
    if(m==1)
      tmp=(squeeze(training_com(i,list,:)));
      tr_ncom(i,:)=tmp*(tmp'*tmp)^(-1/2);
    else
      tmp=(squeeze(training_com(i,list,:)))';
      tr_ncom(i,:,:)=tmp*(tmp'*tmp)^(-1/2);
    end
 end

%size(tt_ncom)
%size(tr_ncom)
%computing the distance between testing and training components
 dist=zeros(tt_n,tr_n);
 for i=1:tt_n
   for j=1:tr_n
     if(m==1)
       s=svd(tt_ncom(i,:)*tr_ncom(j,:)');
     else
       s=svd(squeeze(tt_ncom(i,:,:))'*squeeze(tr_ncom(j,:,:)));
     end
     dist(i,j)=sum(s(:).^2)/m;
      %dist(i,j)=real(sqrt(m-sum(s(:).^2)));
      %dist(i,j)=real(sqrt(sum(s(:).^2)/m));
   end
 end
 
kernel_=tanh(gama.*dist+sigma);

for i=1:tt_n
  test_list=i;
  for j=1:tr_n
    
    [tpred(j), accuracy_P, dec_values_P] = svmpredict(1, [i, kernel_(test_list,:)], model{j},'-b 1');
    [tmp,idx]=max(dec_values_P);
    tdec_values_P(j)=max(dec_values_P)*model{j}.Label(idx);

  end
  
  ind_value(i,:)=tdec_values_P(:)';
  pred_value(i)=median(tdec_values_P);
  pred_label(i)=sign(pred_value(i));
  if(pred_value(i)==0)
    pred_label(i)=1;
  end
end
pred_label=pred_label(:);
pred_value=pred_value(:);