testphase.m

for MLP neural network.

function[per confmat DExperts]=testPhase(WI,WO,PT,TT,NTmp,NumNet)
% PT=NormInData(PT);
for i=1:size(PT,1)
    if NTmp(i,1)==NTmp(i,2)
        PT(i,:)=(-NTmp(i,1)/(NTmp(i,1)))+PT(i,:)./(NTmp(i,1));            
    else
        PT(i,:)=(-NTmp(i,1)/(NTmp(i,2)-NTmp(i,1))+PT(i,:)./(NTmp(i,2)-NTmp(i,1)));
    end
end

PT(size(PT,1)+1,:)=1;%+1 is for bias
for NC=1:NumNet
    for i=1:size(PT,2)
        Y1=(USF(PT(:,i)'*WI{NC,1}));        
        Y1=[Y1 1];
        Yout(:,i)=(USF(Y1*WO{NC,1}))';
    end
    [m,nn]=max(Yout);
    [m1,nn1]=max(TT);
    counter1=0;
    for i=1:size(TT,2)
        if nn(1,i)==nn1(1,i) 
            counter1=counter1+1;
        end
%%   confusion matrix
        try
            confmat{nn1(1,i),nn(1,i)}(size(confmat{nn1(1,i),nn(1,i)},2)+1)=i-fix(i/(size(PT,2)/10))*(size(PT,2)/10); %i;
        catch
            confmat{nn1(1,i),nn(1,i)}(1)=i-fix(i/(size(PT,2)/10))*(size(PT,2)/10); %i;        
        end
%%   Learning Rate oF Each Expert
        for TNC=1:NumNet
            if nn(1,i)==nn1(1,i) 
                try
                    DExperts{TNC,nn1(1,i)}(size(DExperts{TNC,nn1(1,i)},2)+1)=i-fix(i/(size(PT,2)/10))*(size(PT,2)/10); %i;
                catch
                    DExperts{TNC,nn1(1,i)}(1)=i-fix(i/(size(PT,2)/10))*(size(PT,2)/10); %i;        
                end
            end
        end
    end
    per(NC)=100*counter1/size(TT,2)
end