维特比算法工具类

维特比算法工具类

Author: linyiqun

Others/DataMining_Viterbi/ViterbiTool.java

@@ -0,0 +1,240 @@
+package DataMining_Viterbi;
+
+import java.io.BufferedReader;
+import java.io.File;
+import java.io.FileReader;
+import java.io.IOException;
+import java.util.ArrayList;
+import java.util.HashMap;
+import java.util.Map;
+
+/**
+ * 维特比算法工具类
+ * 
+ * @author lyq
+ * 
+ */
+public class ViterbiTool {
+	// 状态转移概率矩阵文件地址
+	private String stmFilePath;
+	// 混淆矩阵文件地址
+	private String confusionFilePath;
+	// 初始状态概率
+	private double[] initStatePro;
+	// 观察到的状态序列
+	public String[] observeStates;
+	// 状态转移矩阵值
+	private double[][] stMatrix;
+	// 混淆矩阵值
+	private double[][] confusionMatrix;
+	// 各个条件下的潜在特征概率值
+	private double[][] potentialValues;
+	// 潜在特征
+	private ArrayList<String> potentialAttrs;
+	// 属性值列坐标映射图
+	private HashMap<String, Integer> name2Index;
+	// 列坐标属性值映射图
+	private HashMap<Integer, String> index2name;
+
+	public ViterbiTool(String stmFilePath, String confusionFilePath,
+			double[] initStatePro, String[] observeStates) {
+		this.stmFilePath = stmFilePath;
+		this.confusionFilePath = confusionFilePath;
+		this.initStatePro = initStatePro;
+		this.observeStates = observeStates;
+
+		initOperation();
+	}
+
+	/**
+	 * 初始化数据操作
+	 */
+	private void initOperation() {
+		double[] temp;
+		int index;
+		ArrayList<String[]> smtDatas;
+		ArrayList<String[]> cfDatas;
+
+		smtDatas = readDataFile(stmFilePath);
+		cfDatas = readDataFile(confusionFilePath);
+
+		index = 0;
+		this.stMatrix = new double[smtDatas.size()][];
+		for (String[] array : smtDatas) {
+			temp = new double[array.length];
+			for (int i = 0; i < array.length; i++) {
+				try {
+					temp[i] = Double.parseDouble(array[i]);
+				} catch (NumberFormatException e) {
+					temp[i] = -1;
+				}
+			}
+
+			// 将转换后的值赋给数组中
+			this.stMatrix[index] = temp;
+			index++;
+		}
+
+		index = 0;
+		this.confusionMatrix = new double[cfDatas.size()][];
+		for (String[] array : cfDatas) {
+			temp = new double[array.length];
+			for (int i = 0; i < array.length; i++) {
+				try {
+					temp[i] = Double.parseDouble(array[i]);
+				} catch (NumberFormatException e) {
+					temp[i] = -1;
+				}
+			}
+
+			// 将转换后的值赋给数组中
+			this.confusionMatrix[index] = temp;
+			index++;
+		}
+
+		this.potentialAttrs = new ArrayList<>();
+		// 添加潜在特征属性
+		for (String s : smtDatas.get(0)) {
+			this.potentialAttrs.add(s);
+		}
+		// 去除首列无效列
+		potentialAttrs.remove(0);
+
+		this.name2Index = new HashMap<>();
+		this.index2name = new HashMap<>();
+
+		// 添加名称下标映射关系
+		for (int i = 1; i < smtDatas.get(0).length; i++) {
+			this.name2Index.put(smtDatas.get(0)[i], i);
+			// 添加下标到名称的映射
+			this.index2name.put(i, smtDatas.get(0)[i]);
+		}
+
+		for (int i = 1; i < cfDatas.get(0).length; i++) {
+			this.name2Index.put(cfDatas.get(0)[i], i);
+		}
+	}
+
+	/**
+	 * 从文件中读取数据
+	 */
+	private ArrayList<String[]> readDataFile(String filePath) {
+		File file = new File(filePath);
+		ArrayList<String[]> dataArray = new ArrayList<String[]>();
+
+		try {
+			BufferedReader in = new BufferedReader(new FileReader(file));
+			String str;
+			String[] tempArray;
+			while ((str = in.readLine()) != null) {
+				tempArray = str.split(" ");
+				dataArray.add(tempArray);
+			}
+			in.close();
+		} catch (IOException e) {
+			e.getStackTrace();
+		}
+
+		return dataArray;
+	}
+
+	/**
+	 * 根据观察特征计算隐藏的特征概率矩阵
+	 */
+	private void calPotencialProMatrix() {
+		String curObserveState;
+		// 观察特征和潜在特征的下标
+		int osIndex;
+		int psIndex;
+		double temp;
+		double maxPro;
+		// 混淆矩阵概率值，就是相关影响的因素概率
+		double confusionPro;
+
+		this.potentialValues = new double[observeStates.length][potentialAttrs
+				.size() + 1];
+		for (int i = 0; i < this.observeStates.length; i++) {
+			curObserveState = this.observeStates[i];
+			osIndex = this.name2Index.get(curObserveState);
+			maxPro = -1;
+
+			// 因为是第一个观察特征，没有前面的影响，根据初始状态计算
+			if (i == 0) {
+				for (String attr : this.potentialAttrs) {
+					psIndex = this.name2Index.get(attr);
+					confusionPro = this.confusionMatrix[psIndex][osIndex];
+
+					temp = this.initStatePro[psIndex - 1] * confusionPro;
+					this.potentialValues[BaseNames.DAY1][psIndex] = temp;
+				}
+			} else {
+				// 后面的潜在特征受前一个特征的影响，以及当前的混淆因素影响
+				for (String toDayAttr : this.potentialAttrs) {
+					psIndex = this.name2Index.get(toDayAttr);
+					confusionPro = this.confusionMatrix[psIndex][osIndex];
+
+					int index;
+					maxPro = -1;
+					// 通过昨天的概率计算今天此特征的最大概率
+					for (String yAttr : this.potentialAttrs) {
+						index = this.name2Index.get(yAttr);
+						temp = this.potentialValues[i - 1][index]
+								* this.stMatrix[index][psIndex];
+
+						// 计算得到今天此潜在特征的最大概率
+						if (temp > maxPro) {
+							maxPro = temp;
+						}
+					}
+
+					this.potentialValues[i][psIndex] = maxPro * confusionPro;
+				}
+			}
+		}
+	}
+
+	/**
+	 * 根据同时期最大概率值输出潜在特征值
+	 */
+	private void outputResultAttr() {
+		double maxPro;
+		int maxIndex;
+		ArrayList<String> psValues;
+
+		psValues = new ArrayList<>();
+		for (int i = 0; i < this.potentialValues.length; i++) {
+			maxPro = -1;
+			maxIndex = 0;
+
+			for (int j = 0; j < potentialValues[i].length; j++) {
+				if (this.potentialValues[i][j] > maxPro) {
+					maxPro = potentialValues[i][j];
+					maxIndex = j;
+				}
+			}
+
+			// 取出最大概率下标对应的潜在特征
+			psValues.add(this.index2name.get(maxIndex));
+		}
+
+		System.out.println("观察特征为：");
+		for (String s : this.observeStates) {
+			System.out.print(s + ", ");
+		}
+		System.out.println();
+
+		System.out.println("潜在特征为：");
+		for (String s : psValues) {
+			System.out.print(s + ", ");
+		}
+		System.out.println();
+	}
+
+	/**
+	 * 根据观察属性，得到潜在属性信息
+	 */
+	public void calHMMObserve() {
+		calPotencialProMatrix();
+		outputResultAttr();
+	}
+}