说明:每个样本都会装入Data样本对象,决策树生成算法接收的是一个Array<Data>样本列表,所以构建测试数据时也要符合格式,最后生成的决策树是树的根节点,通过里面提供的showTree()方法可查看整个树结构,下面奉上源码。
Data.java
package ai.tree.data; import java.util.HashMap; /** * 样本类 * @author ChenLuyang * @date 2019/2/21 */ public class Data implements Cloneable{ /** * K是特征描述,V是特征值 */ private HashMap<String,String> feature = new HashMap<String, String>(); /** * 该样本结论 */ private String result; public Data(HashMap<String,String> feature,String result){ this.feature = feature; this.result = result; } public HashMap<String, String> getFeature() { return feature; } public String getResult() { return result; } private void setFeature(HashMap<String, String> feature) { this.feature = feature; } @Override public Data clone() { Data object=null; try { object = (Data) super.clone(); object.setFeature((HashMap<String, String>) this.feature.clone()); } catch (CloneNotSupportedException e) { e.printStackTrace(); } return object; } }
DecisionTree.java
package ai.tree.algorithm; import ai.tree.data.Data; import java.math.BigDecimal; import java.util.*; /** * @author ChenLuyang * @date 2019/2/21 */ public class DecisionTree { /** * 递归构建决策树 * * @param dataList 样本集合 * @return ai.tree.algorithm.DecisionTree.TreeNode 使用传入样本构建的决策节点 * @author ChenLuyang * @date 2019/2/21 16:05 */ public TreeNode createTree(List<Data> dataList) { //创建当前节点 TreeNode<String, String, String> nowTreeNode = new TreeNode<String, String, String>(); //当前节点的各个分支节点 Map<String, TreeNode> featureDecisionMap = new HashMap<String, TreeNode>(); //统计当前样本集中所有的分类结果 Set<String> resultSet = new HashSet<String>(); for (Data data : dataList) { resultSet.add(data.getResult()); } //如果当前样本集只有一种类别,则表示不用分类了,返回当前节点 if (resultSet.size() == 1) { String resultClassify = resultSet.iterator().next(); nowTreeNode.setResultNode(resultClassify); return nowTreeNode; } //如果数据集中特征为空,则选择整个集合中出现次数最多的分类,作为分类结果 if (dataList.get(0).getFeature().size() == 0) { Map<String, Integer> countMap = new HashMap<String, Integer>(); for (Data data : dataList) { Integer num = countMap.get(data.getResult()); if (num == null) { countMap.put(data.getResult(), 1); } else { countMap.put(data.getResult(), num + 1); } } String tmpResult = ""; Integer tmpNum = 0; for (String res : countMap.keySet()) { if (countMap.get(res) > tmpNum) { tmpNum = countMap.get(res); tmpResult = res; } } nowTreeNode.setResultNode(tmpResult); return nowTreeNode; } //寻找当前最优分类 String bestLabel = chooseBestFeatureToSplit(dataList); //提取最优特征的所有可能值 Set<String> bestLabelInfoSet = new HashSet<String>(); for (Data data : dataList) { bestLabelInfoSet.add(data.getFeature().get(bestLabel)); } //使用最优特征的各个特征值进行分类 for (String labelInfo : bestLabelInfoSet) { for (Data data : dataList) { } List<Data> branchDataList = splitDataList(dataList, bestLabel, labelInfo); //最优特征下该特征值的节点 TreeNode branchTreeNode = createTree(branchDataList); featureDecisionMap.put(labelInfo, branchTreeNode); } nowTreeNode.setDecisionNode(bestLabel, featureDecisionMap); return nowTreeNode; } /** * 计算传入数据集中的最优分类特征 * * @param dataList * @return int 最优分类特征的描述 * @author ChenLuyang * @date 2019/2/21 14:12 */ public String chooseBestFeatureToSplit(List<Data> dataList) { //目前数据集中的特征集合 Set<String> futureSet = dataList.get(0).getFeature().keySet(); //未分类时的熵 BigDecimal baseEntropy = calcShannonEnt(dataList); //熵差 BigDecimal bestInfoGain = new BigDecimal("0"); //最优特征 String bestFeature = ""; //按照各特征分类 for (String future : futureSet) { //该特征分类后的熵 BigDecimal futureEntropy = new BigDecimal("0"); //该特征的所有特征值去重集合 Set<String> futureInfoSet = new HashSet<String>(); for (Data data : dataList) { futureInfoSet.add(data.getFeature().get(future)); } //按照该特征的特征值一一分类 for (String futureInfo : futureInfoSet) { List<Data> splitResultDataList = splitDataList(dataList, future, futureInfo); //分类后样本数占总样本数的比例 BigDecimal tmpProb = new BigDecimal(splitResultDataList.size() + "").divide(new BigDecimal(dataList.size() + ""), 5, BigDecimal.ROUND_HALF_DOWN); //所占比例乘以分类后的样本熵,然后再进行熵的累加 futureEntropy = futureEntropy.add(tmpProb.multiply(calcShannonEnt(splitResultDataList))); } BigDecimal subEntropy = baseEntropy.subtract(futureEntropy); if (subEntropy.compareTo(bestInfoGain) >= 0) { bestInfoGain = subEntropy; bestFeature = future; } } return bestFeature; } /** * 计算传入样本集的熵值 * * @param dataList 样本集 * @return java.math.BigDecimal 熵 * @author ChenLuyang * @date 2019/2/22 9:41 */ public BigDecimal calcShannonEnt(List<Data> dataList) { //样本总数 BigDecimal sumEntries = new BigDecimal(dataList.size() + ""); //香农熵 BigDecimal shannonEnt = new BigDecimal("0"); //统计各个分类结果的样本数量 Map<String, Integer> resultCountMap = new HashMap<String, Integer>(); for (Data data : dataList) { Integer dataResultCount = resultCountMap.get(data.getResult()); if (dataResultCount == null) { resultCountMap.put(data.getResult(), 1); } else { resultCountMap.put(data.getResult(), dataResultCount + 1); } } for (String resultCountKey : resultCountMap.keySet()) { BigDecimal resultCountValue = new BigDecimal(resultCountMap.get(resultCountKey).toString()); BigDecimal prob = resultCountValue.divide(sumEntries, 5, BigDecimal.ROUND_HALF_DOWN); shannonEnt = shannonEnt.subtract(prob.multiply(new BigDecimal(Math.log(prob.doubleValue()) / Math.log(2) + ""))); } return shannonEnt; } /** * 根据某个特征的特征值,进行样本数据的划分,将划分后的样本数据集返回 * * @param dataList 待划分的样本数据集 * @param future 筛选的特征依据 * @param info 筛选的特征值依据 * @return java.util.List<ai.tree.data.Data> 按照指定特征值分类后的数据集 * @author ChenLuyang * @date 2019/2/21 18:26 */ public List<Data> splitDataList(List<Data> dataList, String future, String info) { List<Data> resultDataList = new ArrayList<Data>(); for (Data data : dataList) { if (data.getFeature().get(future).equals(info)) { Data newData = (Data) data.clone(); newData.getFeature().remove(future); resultDataList.add(newData); } } return resultDataList; } /** * L:每一个特征的描述信息的类型 * F:特征的类型 * R:最终分类结果的类型 */ public class TreeNode<L, F, R> { /** * 该节点的最优特征的描述信息 */ private L label; /** * 根据不同的特征作出响应的决定。 * K为特征值,V为该特征值作出的决策节点 */ private Map<F, TreeNode> featureDecisionMap; /** * 是否为最终分类节点 */ private boolean isFinal; /** * 最终分类结果信息 */ private R resultClassify; /** * 设置叶子节点 * * @param resultClassify 最终分类结果 * @return void * @author ChenLuyang * @date 2019/2/22 18:31 */ public void setResultNode(R resultClassify) { this.isFinal = true; this.resultClassify = resultClassify; } /** * 设置分支节点 * * @param label 当前分支节点的描述信息(特征) * @param featureDecisionMap 当前分支节点的各个特征值,与其对应的子节点 * @return void * @author ChenLuyang * @date 2019/2/22 18:31 */ public void setDecisionNode(L label, Map<F, TreeNode> featureDecisionMap) { this.isFinal = false; this.label = label; this.featureDecisionMap = featureDecisionMap; } /** * 展示当前节点的树结构 * * @return void * @author ChenLuyang * @date 2019/2/22 16:54 */ public String showTree() { HashMap<String, String> treeMap = new HashMap<String, String>(); if (isFinal) { String key = "result"; R value = resultClassify; treeMap.put(key, value.toString()); } else { String key = label.toString(); HashMap<F, String> showFutureMap = new HashMap<F, String>(); for (F f : featureDecisionMap.keySet()) { showFutureMap.put(f, featureDecisionMap.get(f).showTree()); } String value = showFutureMap.toString(); treeMap.put(key, value); } return treeMap.toString(); } public L getLabel() { return label; } public Map<F, TreeNode> getFeatureDecisionMap() { return featureDecisionMap; } public R getResultClassify() { return resultClassify; } public boolean getFinal() { return isFinal; } } }
Start.java
package ai.tree.algorithm; import ai.tree.data.Data; import java.util.ArrayList; import java.util.HashMap; import java.util.List; /** * @author ChenLuyang * @date 2019/2/22 */ public class Start { /** * 构建测试样本集,测试样本如下: 样本特征:{头发长短=短发, 身材=胖, 是否戴眼镜=有眼镜} 分类:男 样本特征:{头发长短=长发, 身材=瘦, 是否戴眼镜=有眼镜} 分类:女 样本特征:{头发长短=短发, 身材=胖, 是否戴眼镜=有眼镜} 分类:女 样本特征:{头发长短=长发, 身材=胖, 是否戴眼镜=没眼镜} 分类:男 样本特征:{头发长短=短发, 身材=瘦, 是否戴眼镜=没眼镜} 分类:男 样本特征:{头发长短=长发, 身材=瘦, 是否戴眼镜=有眼镜} 分类:女 样本特征:{头发长短=长发, 身材=胖, 是否戴眼镜=有眼镜} 分类:男 * @author ChenLuyang * @date 2019/2/21 15:34 * @return java.util.List<ai.tree.data.DecisionTreeTestData.Data> 样本集 */ public static List<Data> createDataList(){ /** * 样本特征描述 * @author ChenLuyang * @date 2019/2/22 18:55 * @return java.util.List<ai.tree.data.Data> */ String[] labels = new String[]{"是否戴眼镜", "头发长短", "身材"}; List<Data> dataList = new ArrayList<Data>(); HashMap<String,String> feature1 = new HashMap<String, String>(); feature1.put(labels[0],"有眼镜"); feature1.put(labels[1].toString(),"短发"); feature1.put(labels[2].toString(),"胖"); dataList.add(new Data(feature1,"男")); HashMap<String,String> feature2 = new HashMap<String, String>(); feature2.put(labels[0],"有眼镜"); feature2.put(labels[1],"长发"); feature2.put(labels[2],"瘦"); dataList.add(new Data(feature2,"女")); HashMap<String,String> feature3 = new HashMap<String, String>(); feature3.put(labels[0],"有眼镜"); feature3.put(labels[1],"短发"); feature3.put(labels[2],"胖"); dataList.add(new Data(feature3,"女")); HashMap<String,String> feature4 = new HashMap<String, String>(); feature4.put(labels[0],"没眼镜"); feature4.put(labels[1],"长发"); feature4.put(labels[2],"胖"); dataList.add(new Data(feature4,"男")); HashMap<String,String> feature5 = new HashMap<String, String>(); feature5.put(labels[0],"没眼镜"); feature5.put(labels[1],"短发"); feature5.put(labels[2],"瘦"); dataList.add(new Data(feature5,"男")); HashMap<String,String> feature6 = new HashMap<String, String>(); feature6.put(labels[0],"有眼镜"); feature6.put(labels[1],"长发"); feature6.put(labels[2],"瘦"); dataList.add(new Data(feature6,"女")); HashMap<String,String> feature7 = new HashMap<String, String>(); feature7.put(labels[0],"有眼镜"); feature7.put(labels[1],"长发"); feature7.put(labels[2],"胖"); dataList.add(new Data(feature7,"男")); return dataList; } public static void main(String[] args) { DecisionTree decisionTree = new DecisionTree(); //使用测试样本生成决策树 DecisionTree.TreeNode tree = decisionTree.createTree(createDataList()); //展示决策树 System.out.println(tree.showTree()); } }
生成树结构:{是否戴眼镜={没眼镜={result=男}, 有眼镜={身材={胖={头发长短={长发={result=男}, 短发={result=女}}}, 瘦={result=女}}}}}