[tr, tr_target] = load ('bibtex.csv')
[P,PN,M,MN,s,sN] = iMLNB_train(tr,tr_target,1)
m = iMLNB_model(tr,tr_target, P,PN,M,MN,s,sN)

function [P,PN,M,MN,s,sN]=iMLNB_train(tr,tr_target,S)
% imlnb training
    [nc,nt]=size(tr_target);
    [nt,D]=size(tr);
    for i=1:nc
        t=sum(tr_target(i,:)==ones(1,nt));
        P(i,1)=(S+t)/(S*2+nt);
        Pn(i,1)=1-Prior(i,1);
        [X,in]=sort(tr_target(i,:),'descend');
        in1=in(1:t);
        in2=in(t+1:nt);        
       tr_p=tr(in1,:);
        tr_n=tr(in2,:);
        for j=1:D
            [M(i,j),s(i,j)]=cartprod(normfit(tr_p(:,j)),binopdf(tr_p(:,j+1:n)));
            [MN(i,j),sN(i,j)]=cartprod(normfit(tr_n(:,j)), binopdf(tr_n(:,j+1:n)));
        end
    end

function model=iMLNB_model(td,td_target,P,PN,M,MN,s,sN)
% model 
    nr=sum(features);
    [temp,ind]=sort(features,'descend');
    ind=ind(1:nr);
    dim=size(td,1);
    model=0;
    for i=1:dim
        [HL,RL,OE,cover,AP,Y,pre]=iMLNB(td{i,1}(:,ind),td_target{i,1},P{i,1},PN{i,1},M{i,1}(:,ind),MN{i,1}(:,ind),s{i,1}(:,ind),sN{i,1}(:,index));
        model=model+(HL+RL)/2;
    end
    
    model=model/dim;

function HL=HL(pre,td_target)
% Hamming loss
    [nc,n]=size(pre);
    miss_samples=sum(sum(pre ~=td_target));
    HL=miss_samples/(nc*n);