classify.m

有监督自组织映射-偏最小二乘算法(A supervised self-organising map–partial least squares algorithm),可以用语多变量数据的回归分析

function [Class, PercentageCorrect] = Classify(Ytest,Ypred,MLKP);

[Nobj,Nvar]=size(Ytest);
if (upper(MLKP.OutputCriterion) == 'MAX')
    OffSet=Nvar;
    Class=zeros(Nvar+1,Nvar+1);
else
    OffSet=Nvar+2;
    Class=zeros(Nvar+1,Nvar+3);
end
%
% calculate confusion matrix
%
for iobj=1:Nobj
    Ytst=Ypred(iobj,:);
    [Value, PredClass]=max(Ytst);
    if (upper(MLKP.OutputCriterion) == 'MAX')
        DefClass=0;
    else
        Ytst(PredClass) = -1;
        if (Value > MLKP.Threshold)
            if (Value-max(Ytst) > MLKP.Diff)
                DefClass = 0; % classification unambiguously
            else
                DefClass = 2; % classification undecided
            end
        else
            DefClass = 1; % classification unknown
        end
    end
    [Dummy, RealClass]=max(Ytest(iobj,:));
    if (DefClass == 0)
        Class(RealClass,PredClass) = Class(RealClass,PredClass)+1;
    else
        Class(RealClass,Nvar+DefClass) = Class(RealClass,Nvar+DefClass)+1;
    end
end    
%
% calculate alpha and beta errors
%
for irow=1:Nvar
    NFalseClass=0;
    for icol=1:OffSet
        if (icol == irow)
            NTrueClass=Class(irow,icol);
        else
            NFalseClass=NFalseClass+Class(irow,icol);
        end
    end
    if (NTrueClass+NFalseClass > 0)
        Class(irow,OffSet+1)=round(100*(1-NTrueClass/(NTrueClass+NFalseClass)));
    else
        Class(irow,OffSet+1)=0;
    end
end
for icol=1:Nvar
    NFalseClass=0;
    for irow=1:Nvar
        if (icol == irow)
            NTrueClass=Class(irow,icol);
        else
            NFalseClass=NFalseClass+Class(irow,icol);
        end
    end
    if (NTrueClass > 0)
        Class(Nvar+1,icol)=round(100*(1-NTrueClass/(NTrueClass+NFalseClass)));
    else
        Class(Nvar+1,icol)=0;
    end
end   
PercentageCorrect=100*sum(diag(Class))/Nobj;