decisionstumpforrealboost.m

好东西啊 人脸识别用的 希望大家多支持这个

%%% Patrick Etyngier
%%% [e-mail]
%%% http://cermics.enpc.fr/~etyngier
%%% ENPC - Ecole Nationale des Ponts et Chaussees 
%%%     http://www.enpc.fr
%%% CERTIS - Centre d'Enseignement en Recherche en Technologie de
%%%     l'Information et Systemes
%%%     http://cermics.enpc.fr/~certis
%%% Copyright CERTIS
%%%
%%% V.9 decisionStumpForRealBoost.m

function [c_,d_,w_,classOne_,classTwo_]  = decisionStumpForRealBoost(X,y,weight,h)
%%
%%      | X1^T |        | y1 |               | w1 |
%%  X = |      | /  y = |    | /    weight = |    |
%%      | XN^T |        | yN |               | wN |
%%
if nargin < 2
    error('[c_,d_]  = decisionStump(X,y,weight) : Argument error');
end

[N , dimFS] = size(X);

if nargin < 3
    weight = ones(N,1)/N;
end

sign_ = [1 -1];

c_ = -1;
d_ = -1;

trueClassOne = find(y==1);
trueClassTwo = find(y==-1);

maxLog = Inf;
old = N;
indi = 1;
condi = 0;
condi = 0;
if condi
    figure(h)
    clf
end

fprintf('\t');
fprintf('Decision Stump / dimension : ')

for d = 1:dimFS
    fprintf('%.4d',d);
    vs  = sort(X(:,d));
    mtt = mean([vs,circshift(vs,1)],2);
    thresholds = mtt(2:end);

    TC  = circshift(thresholds,1);
    thresholds = thresholds(find(thresholds-TC~=0));
    di = floor(length(thresholds));
    temp = randperm(length(thresholds));
    thresholds = thresholds(temp(1:di));
    for c = 1:length(thresholds)
        
        %( c==66 ) & ( d == 1 ) ;
        s=1;
   %     for s = 1:length(sign_)
            line = [];
            line(d,:) = sign_(s)*1;
            line(dimFS+1,:) = - sign_(s)*thresholds(c);
            predY = pksign( [X ones(N,1)] * line );
            
            tmp(find(y.*predY~=-1),:) = 1;
            tmp(find(y.*predY==-1),:) = 0;
            
            n11 = sum(weight(trueClassOne).*tmp(trueClassOne));
            n21 = sum(weight(trueClassTwo).*~tmp(trueClassTwo));
            n12 = sum(weight(trueClassOne).*~tmp(trueClassOne));
            n22 = sum(weight(trueClassTwo).*tmp(trueClassTwo));
         
            logL = sqrt(n22*n12)  + sqrt(n11*n21);
% fprintf('logl %d ',logL);
% fprintf('\n')
% fprintf('maxlog %d ',maxLog);
% fprintf('\n')
% fprintf('sum %d ',sum(~tmp));
% fprintf('\n')
% fprintf('old %d ' ,old);
% fprintf('\n')
            if logL < maxLog | (logL == maxLog & sum(~tmp) < old )
                maxLog = (logL);
                old = sum(~tmp);
                if d ~= -1
                    c_ = thresholds(c);
                    d_ = d;
                    w_ = sign_(s);
                    classOne_ = find(predY ~= -1);
                    classTwo_ = find(predY == -1);
                end
            end
           
        indi = indi+1;
    end
    fprintf('\b\b\b\b');
end
fprintf('\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b');