generate_decision_tree.m

ID3+C4.5的源程序。用于数据挖掘决策算法的一个实例。

function Result=Generate_decision_tree(DataName,WhereSen,ForecastSen,attributName,i,j)
%DataName为数表名称,ForecastSen预测属性名称,attributName为现有的属性名称,i为结点位置标记,为
%全局变量
%WhereSen为筛选语句名称,也就是从决策树根到这个结点的筛选条件 由j来记录parent的位置为局部变量
global Node;
global i;
logintimeout(15);
conn = database('DecisionTreeTest', '', '');
%exec(conn,'use PAKDDCompetition2007');
%所建决策树结点结构 对中间结点 1 属性值attributSen 2 筛选语句whereSen 3父结点ID 4 子结点ID 根结点父结点为NULL
%对叶子结点 1 属性值 2 筛选语句 3 分类divSen 4 父结点ID 5 子结点为NULL

%把属性的单元数组转为字符串 这里的attributName是一个1*n的单元数组

if i==1
    %第一次循环的时候得到需要进行操作的列名
    KK=['select b.name from   sysobjects   a   inner   join   syscolumns   b   on   a.id=b.id  where a.name=''',DataName,'''']
    curs1=exec(conn,['select b.name from   sysobjects   a   inner   join   syscolumns   b   on   a.id=b.id  where a.name=''',DataName,'''']);
%    curs1=exec(conn,'select b.name from   sysobjects   a   inner   join   syscolumns   b   on   a.id=b.id  where a.name=''DecisionTreeTest''');
    curs1=fetch(curs1);
    attributeNameArray=curs1.data;
    close(curs1);
    %为适应DecisionTreeTestCo翻转了90度
    x=size(attributeNameArray);
    attributName=reshape(attributeNameArray,x(2),x(1));
else
    attributNameList='';
    for I=1:length(attributName)
        if ~isempty(attributName{1,I})
            if I==length(attributName)
            attributNameList=[attributNameList,attributName{1,I}]
            else
            attributNameList=[attributNameList,attributName{1,I},','];
            end
        end
    end
end

curs2=exec(conn,['select',' ',ForecastSen,' from',' ',DataName,' ',WhereSen,' ','group by',' ',ForecastSen]);
setdbprefs('DataReturnFormat','cellarray');
KK=['select',' ',ForecastSen,' from',' ',DataName,' ',WhereSen,' ','group by',' ',ForecastSen]
curs2=fetch(curs2);
DiffClus=curs2.data;
%close(curs2);
%如果Sample都在同一个类C上,则产生叶子结点如下,并用类C标记
if length(DiffClus)==1
    Node(i).attributSen=attributNameList;
    Node(i).whereSen=WhereSen
    Node(i).divSen=DiffClus(1);
        if i==1
        Node(i).parent=0;
        else
        Node(i).parent=j;
        end
    return;
end


%如果剩下的输入属性为空,则按多数表决决定类别
numnoempty=0;
for I=1:length(attributName)
    if isempty(attributName{1,I})
        numnoempty=numnoempty+1;
    end
end     
if numnoempty==length(attributName)-2
    divflag=MostChoose(DataName,WhereSen,ForecastSen);
    %由子函数得到的divflag即为多数表决的结果 产生叶子结点如下
    Node(i).attributSen=attributNameList;
    Node(i).whereSen=WhereSen;
    Node(i).divSen=divflag;
        if i==1
        Node(i).parent=0;
        else
        Node(i).parent=j;
        end
    return;
end
    


%若是其它情况则调用信息增益比较算法来确定信息增益最高属性
    sz=DecisionTreeTestCon(DataName,WhereSen,ForecastSen,attributName);
    if sz==0%如果信息增益过小，则直接产生叶子结点
        divflag=MostChoose(DataName,WhereSen,ForecastSen);
        Node(i).attributSen=attributNameList;
        Node(i).whereSen=WhereSen
        Node(i).divSen=divflag;
        if i==1
            Node(i).parent=0;
        else
            Node(i).parent=j;
        end          
    else
    Node(i).attributSen=DecisionTreeTestCon(DataName,WhereSen,ForecastSen,attributName);
    Node(i).whereSen=WhereSen;
    if i==1
        Node(i).parent=0;
    else
        Node(i).parent=j;
    end
    %找出新结点的分类情况并递归求解
    KK=['select ',' ',Node(i).attributSen,' from DecisionTreeTest group by',' ',Node(i).attributSen]
    curs=exec(conn,['select ',' ',Node(i).attributSen,' from',' ',DataName,' group by',' ',Node(i).attributSen]);
    setdbprefs('DataReturnFormat','cellarray');
    curs=fetch(curs,3);
    NewDiv=curs.data;
    close(curs);
    WhereSenOld=WhereSen;
    %WhereSenOld用来保存上一层的where条件
    if i==1
        WhereSenOld=WhereSen;
        WhereSen=['where',' ',Node(i).attributSen,'='];
    else
        WhereSen=[WhereSen,' and',' ',Node(i).attributSen,'='];
    end
    %进入循环递归之前对一些变量进行保存,以便出递归时调用
    %WhereSenOld1用来保存进循环前的where条件
    WhereSenOld1=WhereSen;
    attributNameOld=attributName;
    j=i;
    %取各条件分枝做递归运算
    for I=1:length(NewDiv)
        WhereSen=[WhereSenOld1,'''',NewDiv{I,1},''''];
        i
        for II=1:length(attributNameOld)
        if strcmp(attributNameOld{1,II},Node(i).attributSen)
                attributName{1,II}={};
            end
        end
        i=i+1;
        KK=['select count(*) from',' ',DataName,' ',WhereSen]
        curs=exec(conn,['select count(*) from',' ',DataName,' ',WhereSen]);
        setdbprefs('DataReturnFormat','numeric');
        curs=fetch(curs,3);
        RecordNum=curs.data;
        if RecordNum(1)<2
            divflag=MostChoose(DataName,WhereSenOld,ForecastSen);
            Node(i).attributSen=attributNameList;
            Node(i).whereSen=WhereSen
            Node(i).divSen=divflag;
            if i==1
            Node(i).parent=0;
            else
            Node(i).parent=j;
            
            end                      
        else
            Generate_decision_tree(DataName,WhereSen,ForecastSen,attributName,i,j);
        end
    end
    KK=['select b.name from   sysobjects   a   inner   join   syscolumns   b   on   a.id=b.id  where a.name=''',DataName,'''']
    Result=Node;
    end

    function modivflag=MostChoose(DataName,WhereSen,ForecastSen)
    logintimeout(15);
    conn = database('DecisionTreeTest', '', '');
%    exec(conn,'use PAKDDCompetition2007');
    KK=['select ',' ',ForecastSen,',count(',ForecastSen,') from',' ',DataName,' ',WhereSen,' group by',' ',ForecastSen]
        curs3=exec(conn,['select ',' ',ForecastSen,',count(',ForecastSen,') from',' ',DataName,' ',WhereSen,' group by',' ',ForecastSen]);
        setdbprefs('DataReturnFormat','cellarray');
        curs3=fetch(curs3,3);
        test=curs3.data;
%        close(curs3);
        divnum=0;
        for I=1:length(test)
            if divnum<test{I,2}
                divnum=test{I,2};
                divflag=I;
            end
        end
        modivflag=test{divflag,1};