adaptlearnparams.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 [AlphaX,AlphaY,RadiusX,RadiusY,DecayFract] = AdaptLearnParams(iter,MLKP);

AlphaX=MLKP.AlphaMaxX-(MLKP.AlphaMaxX-MLKP.AlphaMinX)*iter/MLKP.MaxIter;
AlphaY=MLKP.AlphaMaxY-(MLKP.AlphaMaxY-MLKP.AlphaMinY)*iter/MLKP.MaxIter;
F = 1.0 / MLKP.ShrinkCutOff;
if upper(MLKP.Shrink)=='EXP'
    if MLKP.RadiusMaxX == MLKP.RadiusMinX
        RadiusX = MLKP.RadiusMaxX;
    else
        ValX=F*log(MLKP.RadiusMaxX-MLKP.RadiusMinX+1);
        RadiusX=MLKP.RadiusMinX+floor(exp(-iter*ValX/MLKP.MaxIter)*(MLKP.RadiusMaxX-MLKP.RadiusMinX+1));
    end
    if MLKP.RadiusMaxY == MLKP.RadiusMinY
        RadiusY = MLKP.RadiusMaxY;
    else
        ValY=F*log(MLKP.RadiusMaxY-MLKP.RadiusMinY+1);
        RadiusY=MLKP.RadiusMinY+floor(exp(-iter*ValY/MLKP.MaxIter)*(MLKP.RadiusMaxY-MLKP.RadiusMinY+1));
    end
else
    RadiusX=max(0,ceil(MLKP.RadiusMaxX-(MLKP.RadiusMaxX-MLKP.RadiusMinX)*F*iter/MLKP.MaxIter));
    RadiusY=max(0,ceil(MLKP.RadiusMaxY-(MLKP.RadiusMaxY-MLKP.RadiusMinY)*F*iter/MLKP.MaxIter));
end
if (upper(MLKP.WeightDecay) == 'Y')
    F = 1.0 / MLKP.WeightDecayCutOff;
    if (upper(MLKP.WeightDecayMode) == 'EXP')
        DecayFract=1-exp(-MLKP.WeightDecayFactorStart-...
        (MLKP.WeightDecayFactorEnd-MLKP.WeightDecayFactorStart)*...
        F*(iter-1)/(MLKP.MaxIter-1));
    else
        FractStart=exp(-MLKP.WeightDecayFactorStart);
        FractEnd=exp(-MLKP.WeightDecayFactorEnd);
        DecayFract=1-max(0,(FractStart-(FractStart-FractEnd)*F*(iter-1)/(MLKP.MaxIter-1)));
    end
else
    DecayFract=1;
end