求决策树源代码。最好使用matlab实现。

Question

如题所述

Answer 1

function [Tree RulesMatrix]=DecisionTree(DataSet,AttributName)
%输入为训练集，为离散后的数字，如记录1：1 1 3 2 1；
%前面为属性列，最后一列为类标
if nargin<1
error('请输入数据集');
else
if isstr(DataSet)
[DataSet AttributValue]=readdata2(DataSet);
else
AttributValue=[];
end
end
if nargin<2
AttributName=[];
end
Attributs=[1:size(DataSet,2)-1];
Tree=CreatTree(DataSet,Attributs);
disp([char(13) 'The Decision Tree:']);
showTree(Tree,0,0,1,AttributValue,AttributName);
Rules=getRule(Tree);
RulesMatrix=zeros(size(Rules,1),size(DataSet,2));
for i=1:size(Rules,1)
rule=cell2struct(Rules(i,1),{'str'});
rule=str2num([rule.str([1:(find(rule.str=='C')-1)]) rule.str((find(rule.str=='C')+1):length(rule.str))]);
for j=1:(length(rule)-1)/2
RulesMatrix(i,rule((j-1)*2+1))=rule(j*2);
end
RulesMatrix(i,size(DataSet,2))=rule(length(rule));
end
end
function Tree=CreatTree(DataSet,Attributs) %决策树程序 输入为：数据集，属性名列表
%disp(Attributs);
[S ValRecords]=ComputEntropy(DataSet,0);
if(S==0) %当样例全为一类时退出，返回叶子节点类标
for i=1:length(ValRecords)
if(length(ValRecords(i).matrix)==size(DataSet,1))
break;
end
end
Tree.Attribut=i;
Tree.Child=[];
return;
end
if(length(Attributs)==0) %当条件属性个数为0时返回占多数的类标
mostlabelnum=0;
mostlabel=0;
for i=1:length(ValRecords)
if(length(ValRecords(i).matrix)>mostlabelnum)
mostlabelnum=length(ValRecords(i).matrix);
mostlabel=i;
end
end
Tree.Attribut=mostlabel;
Tree.Child=[];
return;
end
for i=1:length(Attributs)
[Sa(i) ValRecord]=ComputEntropy(DataSet,i);
Gains(i)=S-Sa(i);
AtrributMatric(i).val=ValRecord;
end
[maxval maxindex]=max(Gains);
Tree.Attribut=Attributs(maxindex);
Attributs2=[Attributs(1:maxindex-1) Attributs(maxindex+1:length(Attributs))];
for j=1:length(AtrributMatric(maxindex).val)
DataSet2=[DataSet(AtrributMatric(maxindex).val(j).matrix',1:maxindex-1) DataSet(AtrributMatric(maxindex).val(j).matrix',maxindex+1:size(DataSet,2))];
if(size(DataSet2,1)==0)
mostlabelnum=0;
mostlabel=0;
for i=1:length(ValRecords)
if(length(ValRecords(i).matrix)>mostlabelnum)
mostlabelnum=length(ValRecords(i).matrix);
mostlabel=i;
end
end
Tree.Child(j).root.Attribut=mostlabel;
Tree.Child(j).root.Child=[];
else
Tree.Child(j).root=CreatTree(DataSet2,Attributs2);
end
end
end
function [Entropy RecordVal]=ComputEntropy(DataSet,attribut) %计算信息熵
if(attribut==0)
clnum=0;
for i=1:size(DataSet,1)
if(DataSet(i,size(DataSet,2))>clnum) %防止下标越界
classnum(DataSet(i,size(DataSet,2)))=0;
clnum=DataSet(i,size(DataSet,2));
RecordVal(DataSet(i,size(DataSet,2))).matrix=[];
end
classnum(DataSet(i,size(DataSet,2)))=classnum(DataSet(i,size(DataSet,2)))+1;
RecordVal(DataSet(i,size(DataSet,2))).matrix=[RecordVal(DataSet(i,size(DataSet,2))).matrix i];
end

Entropy=0;
for j=1:length(classnum)
P=classnum(j)/size(DataSet,1);
if(P~=0)
Entropy=Entropy+(-P)*log2(P);
end
end
else
valnum=0;
for i=1:size(DataSet,1)
if(DataSet(i,attribut)>valnum) %防止参数下标越界
clnum(DataSet(i,attribut))=0;
valnum=DataSet(i,attribut);
Valueexamnum(DataSet(i,attribut))=0;
RecordVal(DataSet(i,attribut)).matrix=[]; %将编号保留下来，以方便后面按值分割数据集
end
if(DataSet(i,size(DataSet,2))>clnum(DataSet(i,attribut))) %防止下标越界
Value(DataSet(i,attribut)).classnum(DataSet(i,size(DataSet,2)))=0;
clnum(DataSet(i,attribut))=DataSet(i,size(DataSet,2));
end
Value(DataSet(i,attribut)).classnum(DataSet(i,size(DataSet,2)))= Value(DataSet(i,attribut)).classnum(DataSet(i,size(DataSet,2)))+1;
Valueexamnum(DataSet(i,attribut))= Valueexamnum(DataSet(i,attribut))+1;
RecordVal(DataSet(i,attribut)).matrix=[RecordVal(DataSet(i,attribut)).matrix i];
end
Entropy=0;
for j=1:valnum
Entropys=0;
for k=1:length(Value(j).classnum)
P=Value(j).classnum(k)/Valueexamnum(j);
if(P~=0)
Entropys=Entropys+(-P)*log2(P);
end
end
Entropy=Entropy+(Valueexamnum(j)/size(DataSet,1))*Entropys;
end
end
end
function showTree(Tree,level,value,branch,AttributValue,AttributName)
blank=[];
for i=1:level-1
if(branch(i)==1)
blank=[blank ' |'];
else
blank=[blank ' '];
end
end
blank=[blank ' '];
if(level==0)
blank=[' (The Root):'];
else
if isempty(AttributValue)
blank=[blank '|_____' int2str(value) '______'];
else
blank=[blank '|_____' value '______'];
end
end
if(length(Tree.Child)~=0) %非叶子节点
if isempty(AttributName)
disp([blank 'Attribut ' int2str(Tree.Attribut)]);
else
disp([blank 'Attribut ' AttributName{Tree.Attribut}]);
end
if isempty(AttributValue)
for j=1:length(Tree.Child)-1
showTree(Tree.Child(j).root,level+1,j,[branch 1],AttributValue,AttributName);
end
showTree(Tree.Child(length(Tree.Child)).root,level+1,length(Tree.Child),[branch(1:length(branch)-1) 0 1],AttributValue,AttributName);
else
for j=1:length(Tree.Child)-1
showTree(Tree.Child(j).root,level+1,AttributValue{Tree.Attribut}{j},[branch 1],AttributValue,AttributName);
end
showTree(Tree.Child(length(Tree.Child)).root,level+1,AttributValue{Tree.Attribut}{length(Tree.Child)},[branch(1:length(branch)-1) 0 1],AttributValue,AttributName);
end
else
if isempty(AttributValue)
disp([blank 'leaf ' int2str(Tree.Attribut)]);
else
disp([blank 'leaf ' AttributValue{length(AttributValue)}{Tree.Attribut}]);
end
end
end
function Rules=getRule(Tree)
if(length(Tree.Child)~=0)
Rules={};
for i=1:length(Tree.Child)
content=getRule(Tree.Child(i).root);
%disp(content);
%disp([num2str(Tree.Attribut) ',' num2str(i) ',']);
for j=1:size(content,1)
rule=cell2struct(content(j,1),{'str'});
content(j,1)={[num2str(Tree.Attribut) ',' num2str(i) ',' rule.str]};
end
Rules=[Rules;content];
end
else
Rules={['C' num2str(Tree.Attribut)]};
end
end