ranking_loss.m

BPMLL is a package for training multi-label BP neural networks. The package includes the MATLAB code

function RankingLoss=Ranking_loss(Outputs,test_target)
%Computing the hamming loss
%Outputs: the predicted outputs of the classifier, the output of the ith instance for the jth class is stored in Outputs(j,i)
%test_target: the actual labels of the test instances, if the ith instance belong to the jth class, test_target(j,i)=1, otherwise test_target(j,i)=-1

    [num_class,num_instance]=size(Outputs);
    temp_Outputs=[];
    temp_test_target=[];
    for i=1:num_instance
        temp=test_target(:,i);
        if((sum(temp)~=num_class)&(sum(temp)~=-num_class))
            temp_Outputs=[temp_Outputs,Outputs(:,i)];
            temp_test_target=[temp_test_target,temp];
        end
    end
    Outputs=temp_Outputs;
    test_target=temp_test_target;     
    [num_class,num_instance]=size(Outputs);
    
    Label=cell(num_instance,1);
    not_Label=cell(num_instance,1);
    Label_size=zeros(1,num_instance);
    for i=1:num_instance
        temp=test_target(:,i);
        Label_size(1,i)=sum(temp==ones(num_class,1));
        for j=1:num_class
            if(temp(j)==1)
                Label{i,1}=[Label{i,1},j];
            else
                not_Label{i,1}=[not_Label{i,1},j];
            end
        end
    end
    
    rankloss=0;
    for i=1:num_instance
        temp=0;
        for m=1:Label_size(i)
            for n=1:(num_class-Label_size(i))
                if(Outputs(Label{i,1}(m),i)<=Outputs(not_Label{i,1}(n),i))
                    temp=temp+1;
                end
            end
        end
        rl_binary(i)=temp/(m*n);
        rankloss=rankloss+temp/(m*n);
    end
    RankingLoss=rankloss/num_instance;