<机器学习>ID3决策树javaDemo

       近期在看周志华老师的机器学习，对第四章的西瓜好坏预测（西瓜数据2.0）写了一个决策树的Demo，写的不好的地方欢迎指正！

数据说明：

       在TrainingData.java中，有学习数据及测试数据，数据内容全部为拼音（书上P80页的表4.2）

package ID3_demo;public class TrainData {public Object[] array_outside;public Object[] test_data;//定义构造函数，初始化对象时即调用public TrainData(){//初始化数组,数据以字符串的形式保存array_outside = new Object[]{new String[]{"qinglv","quansuo","zhuoxiang","qingxi","aoxian","yinghua","Yes"},new String[]{"wuhei","quansuo","chenmen","qingxi","aoxian","yinghua","Yes"},new String[]{"wuhei","quansuo","zhuoxiang","qingxi","aoxian","yinghua","Yes"},new String[]{"qinglv","shaoquan","zhuoxiang","qingxi","shaoao","ruannian","Yes"},new String[]{"wuhei","shaoquan","zhuoxiang","shaohu","shaoao","ruannian","Yes"},new String[]{"qinglv","yingting","qingcui","qingxi","pingtan","ruannian","No"},new String[]{"qianbai","shaoquan","chenmen","shaohu","aoxian","yinghua","No"},new String[]{"wuhei","shaoquan","zhuoxiang","qingxi","shaoao","ruannian","No"},new String[]{"qianbai","quansuo","zhuoxiang","mohu","pingtan","yinghua","No"},new String[]{"qinglv","quansuo","chenmen","shaohu","shaoao","yinghua","No"},};test_data = new Object[]{new String[]{"qinglv","quansuo","chenmen","qingxi","aoxian","yinghua","Yes"},new String[]{"qianbai","quansuo","zhuoxiang","qingxi","aoxian","yinghua","Yes"},new String[]{"wuhei","shaoquan","zhuoxiang","qingxi","shaoao","yinghua","Yes"},new String[]{"wuhei","shaoquan","chenmen","shaohu","shaoao","yinghua","No"},new String[]{"qianbai","yingting","qingcui","mohu","pingtan","yinghua","No"},new String[]{"qinglv","shaoquan","zhuoxiang","shaohu","aoxian","yinghua","No"},new String[]{"qianbai","quansuo","zhuoxiang","mohu","pingtan","ruannian","No"},};}}

决策树的节点类：

package ID3_demo;public class TreeNode {boolean isLeaft;TreeNode parent;TreeNode[] childNodes;String[] attributes;int attribute_id;String parentAttribute;String type = null;public TreeNode(){isLeaft = false;parent = null;childNodes = null;}}

决策树主类（目前只写了后剪枝，也只能对离散数据进行预测，针对 不同的问题，需要修改init调整参数既可 )：

package ID3_demo;import java.util.ArrayList;import java.util.LinkedList;import java.util.Queue;import java.util.Stack;public class ID3 {TreeNode root;Object[] training_data;Object[] test_data;int y,attribute_num,test_num;String[][] attributes;String[] results;public ID3(){init();TrainData t = new TrainData();root = new TreeNode();training_data = t.array_outside;test_data = t.test_data;}public void cutBranch(){System.out.println("=====beforeCutBranch====="+"  "+getAccuracy());double bf=0.0,af=0.0;TreeNode node;Queue<TreeNode> q = new LinkedList<TreeNode>();Stack<TreeNode> p = new Stack<TreeNode>();q.add(root);p.add(root);while(!q.isEmpty()){node = q.poll();if(node.isLeaft){}else{p.add(node);for(int i=0;i<node.childNodes.length;i++){q.add(node.childNodes[i]);}}}System.out.println("p.size:  "+p.size());while(!p.isEmpty()){node = p.pop();bf = getAccuracy();node.isLeaft = true;af = getAccuracy();if(bf > af){node.isLeaft = false;}System.out.println("node: "+node.attribute_id+"  bf: "+bf+" af: "+af);}System.out.println("=====afterCutBranch====="+"  "+getAccuracy());}public double getAccuracy(){double accuracy = 0.0;int good=0,bad=0;String rlt;for(int i=0;i<test_num;i++){rlt = testData((String[])(test_data[i]));if(rlt.equals(((String[])(test_data[i]))[attribute_num])){good++;}else{bad++;}}//System.out.println("good: "+good+" bad: "+bad);accuracy = (double)good/(double)(good+bad);return accuracy;}public String testData(String[] data){int i;String result = "";boolean flag = false;TreeNode t = root;while(true){for(i=0;i<attributes[t.attribute_id].length;i++){if(attributes[t.attribute_id][i].equals(data[t.attribute_id])){if(!t.isLeaft)t = t.childNodes[i];else{flag = true;result = t.type;//System.out.println("判断结果："+t.type);break;}}}if(flag)break;}return result;}public void printTree(TreeNode root){Queue<TreeNode> s = new LinkedList<TreeNode>();TreeNode t;s.add(root);while(!s.isEmpty()){t = s.poll();if(t!=null){if(t.isLeaft){System.out.println("叶子节点：     "+t.type + " 父属性  "+t.parent.attribute_id);}else{if(t.parent!=null)System.out.println("不是叶子节点：       "+t.attribute_id + " 父属性  "+t.parent.attribute_id);elseSystem.out.println("不是叶子节点：       "+t.attribute_id);for(int i=0;i<t.childNodes.length;i++){System.out.println(t.attributes[i]);s.add(t.childNodes[i]);}}}}}public void TreeGenerate(ArrayList<Integer>D,ArrayList<Integer>a,TreeNode father){boolean flag = true;int i,j,selectnum;if(father == root){root.type = getMostInD(D);}for(i=1;i<D.size();i++){if(((String[])(training_data[D.get(i)]))[attribute_num].equals(((String[])(training_data[D.get(0)]))[attribute_num])){}else{flag = false;}}//D中所有样本全属于同一类别Cif(flag){//将node标记为C类叶节点，returnfather.isLeaft = true;father.type = ((String[])(training_data[D.get(0)]))[attribute_num];return;}if(a.size() == 0 || isSameD(D,a)){//将 node标记为叶节点，其类别标记为D中样本数最多的类；returnfather.isLeaft = true;father.type = getMostInD(D);return;}selectnum = selectAttr(D,a);ArrayList<Integer> av = new ArrayList<Integer>();for(i=0;i<a.size();i++){if(a.get(i)!=selectnum){av.add(a.get(i));}}father.attribute_id = selectnum;father.isLeaft = false;int len = attributes[selectnum].length;father.childNodes = new TreeNode[len];father.attributes = new String[len];for(i=0;i<len;i++){TreeNode tree = new TreeNode();father.childNodes[i] = tree;father.attributes[i] = attributes[selectnum][i];tree.parent = father;tree.isLeaft = false;//为node生成一个分支；令Dv表示D中在a*上取值为av*的样本子集ArrayList<Integer>Dv = new ArrayList<Integer>();for(j=0;j<D.size();j++){if(((String[])(training_data[D.get(j)]))[selectnum].equals(attributes[selectnum][i])){Dv.add(D.get(j));}}if(Dv.isEmpty()){//将分支节点标记为叶节点，其类别标记为D中样本最多的类；return;tree.isLeaft = true;tree.type = getMostInD(D);}else{//递归调用TreeGenerate()tree.type = getMostInD(Dv);TreeGenerate(Dv,av,tree);}}}public String getMostInD(ArrayList<Integer>D){String result = "";int ans = -100,i,j;for(i=0;i<results.length;i++){int tmp = 0;for(j=0;j<D.size();j++){if(((String[])(training_data[D.get(j)]))[attribute_num].equals(results[i])){tmp++;}}if(tmp > ans){result = results[i];}}return result;}public boolean isSameD(ArrayList<Integer>D,ArrayList<Integer>a){boolean flag = true;int i,j;for(i=0;i<a.size();i++){for(j=1;j<D.size();j++){if(((String[])(training_data[D.get(j)]))[a.get(i)].equals(((String[])(training_data[D.get(0)]))[a.get(i)])){}else{flag = false;break;}}if(!flag)break;}return flag;}public int selectAttr(ArrayList<Integer>D,ArrayList<Integer>a){int max_a = 0,i;double max = -1000000.0;for(i=0;i<a.size();i++){double tmp = getGain(D,a.get(i));if(tmp >= max){max = tmp;max_a = a.get(i);}}return max_a;}//a_num为从0~5public double getGain(ArrayList<Integer> D,int a_num){double gain = 0.0,ent = 0.0;ArrayList<Integer> di = new ArrayList<Integer>();int len = attributes[a_num].length;int i,j,k;int[] ans = new int[len];for(i=0;i<len;i++)ans[i] = 0;for(i=0;i<len;i++){for(j=0;j<D.size();j++){if(((String[])training_data[D.get(j)])[a_num].equals(attributes[a_num][i])){di.add(j);}}ent += (double)di.size()/D.size()*(getEnt(di));di.clear();}gain = getEnt(D) - ent;return gain;}public double getEnt(ArrayList<Integer> D){double ent = 0.0;int[] ans = new int[y];int total = 0;for(int i=0;i<y;i++)ans[i] = 0;for(int i=0;i<D.size();i++){for(int j=0;j<y;j++){if(((String[])training_data[D.get(i)])[attribute_num].equals(results[j])){ans[j]++;total ++;break;}}}for(int i=0;i<y;i++){if(ans[i]!=0)ent += (double)ans[i]/(double)total * (Math.log((double)ans[i]/total)/Math.log(2));}ent = -ent;return ent;}public void init(){y = 2;test_num = 7;results = new String[]{"Yes","No"};attribute_num = 6;attributes = new String[attribute_num][];attributes[0] = new String[]{"qinglv","wuhei","qianbai"};attributes[1] = new String[]{"quansuo","shaoquan","yingting"};attributes[2] = new String[]{"zhuoxiang","chenmen","qingcui"};attributes[3] = new String[]{"qingxi","shaohu","mohu"};attributes[4] = new String[]{"aoxian","shaoao","pingtan"};attributes[5] = new String[]{"yinghua","ruannian"};}public static void main(String[] args){ID3 id3 = new ID3();ArrayList<Integer> d = new ArrayList<Integer>();ArrayList<Integer> a = new ArrayList<Integer>();for(int i = 0;i<10;i++){d.add(i);}for(int i = 0;i<6;i++){a.add(i);}id3.TreeGenerate(d, a, id3.root);id3.printTree(id3.root);id3.cutBranch();}}