文章标题

function accuracy = myAdaboostMulticlass( Dtrain,Dtest,baseLearner,T,tag)
%myAdaboostMulticlass Summary of this function goes here
%   input： Dtrain——training dataset
%           Dtest——testing dataset（with label）
%           baseLearner——base learner(0 represent decision stump, 1 represnet Naive Bayes)
%           T——training rounds
%           tag—— attribute discrete or continuous（0 represent discrete, 1 represent continuous）

%   output：accuracy on the training set

%adaboost这个算法我是照着紫色的数据挖掘书写的，区别是在第4行不会break掉

pT=ones(size(Dtest,1),T); %every colomn is generated by a base learner
p=ones(size(Dtest,1),1); %every sample's final result

alphaT=zeros(T,1)*(-1); %weight of each base learner. Original value is -1.
weight_sample=ones(size(Dtrain,1),1)/size(Dtrain,1); %Originate the sampling weight of each training sample

for t=1:T
    %sample from training set with withdrawal
    sampleIndex = randsample(size(Dtrain,1),size(Dtrain,1),'true',weight_sample);
    DtrainSample=Dtrain(sampleIndex,:);   %traing set after sampling for this round

    %compute the tth colomn of pT
    train_predict=zeros(size(Dtrain,1),1);

    if baseLearner==0   %base learner is dicision dump
        pT(:,t)=computeDeision(DtrainSample,Dtest(:,1:(end-1)),tag); %remove the last attribute of Dtest
        train_predict=computeDeision(DtrainSample,Dtrain(:,1:(end-1)),tag); 
    elseif baseLearner==1  %base learner is Naive Bayes
         pT(:,t)=computeNB(DtrainSample,Dtest(:,1:(end-1)),tag); 
         train_predict=computeNB(DtrainSample,Dtrain(:,1:(end-1)),tag); 
    end

    train_predict=train_predict(:); %transform to column vector

    errorate= sum( weight_sample.* (Dtrain(:,end) ~=train_predict ) );

    if errorate> 0.5  %threshold for originating weights of training samples
        alphaT(t)=0;
        %originate the sampling weight of each training sample
        weight_sample=ones(size(Dtrain,1),1)/size(Dtrain,1); 
        continue;
    end

    alphaT(t)=0.5*log((1-errorate)/errorate );

    %renew weight_sample
    weight_sample=weight_sample.* exp(alphaT(t)*( (-2)*( Dtrain(:,end)==train_predict)+1));
    weight_sample=weight_sample/sum(weight_sample);   
end

%accumulate the result from each base learner. Add up the vote for each label.
for i=1:size(Dtest)
    lable_vote=zeros(max(Dtrain(:,end)),1);
    for t=1:T
        lable_vote( pT(i,t) )=lable_vote(pT(i,t))+alphaT(t);
    end
    [~,p(i)]=max(lable_vote);
end

accuracy=sum((p==Dtest(:,end)))/size(Dtest,1);
end