Java实现的隐马尔科夫模型中前向、后项和维特比算法.docx

package com.HMM.test;public class HMM protected int N;/状态数protected int M;/观察符号数protected double A;/状态转移概率protected double B;/符号观测概率protected double PI;/初始状态概率分布public HMM();/参数1 状态数目 ；参数2 观察符号数目public HMM(int stateNum,int observationSymbolNum)N=stateNum;M=observationSymbolNum;A=new doubleNN;B=new doubleNM;PI=new doubleN;/前向算法package com.HMM.test;public class Forward extends HMMpublic Forward(int stateNum,int observationSymbolNum)super(stateNum,observationSymbolNum);/ob是已知的观察序列；返回结果是观察序列的概率public double forward(int ob)double alpha = null;return forward(ob,alpha);/ob:已知的观察序列；alpha：输出中间结果，局部概率；返回观察序列的概率public double forward(int ob,double alpha)alpha=new doubleob.lengthN;/1.初始化，计算初始时刻所有状态的局部概率System.out.println(1.初始化：);for(int i=0;iN;i+)alpha0i=PIi*Biob0;System.out.println(alpha0+i+:+alpha0i);/2.归纳，递归计算每个时间点的局部概率System.out.println(2.归纳：);for (int i = 1; i ob.length; i+)/从第一个观测值开始循环， for (int j = 0; j N; j+)/对于每个状态 double sum = 0; for (int k = 0; k N; k+) sum += alphai - 1k * Akj; alphaij = sum * Bjobi; System.out.println(alpha+i+j+:+alphaij); /3.终止，观察序列的概率等于最终时刻所有局部概率之和String s=P(red,yellow,blue)=;System.out.println(3.终止，求和：);double probability = 0; for (int i = 0; i N; i+) probability += alphaob.length - 1i; s+=alpha+(ob.length - 1)+i+; if(i!=N-1) s+=+; System.out.println(s+=+probability); return probability;/后项算法package com.HMM.test;public class Backward extends HMMpublic Backward(int stateNum,int observationSymbolNum)super(stateNum,observationSymbolNum);/ob 已知的观察序列public double backward(int ob) double beta = null; / 只声明，不定义 return backward(ob, beta); /ob 已知的观察序列；beta 后向变量；返回观测序列的概率public double backward(int ob,double beta) beta = new doubleob.lengthN; / 初始化 System.out.println(1.初始化：); for (int i = 0; i = 0; t-) for (int j = 0; j N; j+) double Sum = 0; for (int i = 0; i N; i+) Sum += Aji * Biobt + 1 * betat + 1i; betatj = Sum; System.out.println(beta+t+j+:+betatj); / 终止 String s=P(red,yellow,blue)=;System.out.println(3.终止，求和：); double probability = 0; for (int i = 0; i N; i+) probability += beta0i; s+=beta0+i+; if(i!=N-1) s+=+; System.out.println(s+=+probability); return probability; /维特比算法package com.HMM.test;import java.util.ArrayList;import java.util.List;public class Viterbi extends HMMpublic Viterbi(int stateNum,int observationSymbolNum)super(stateNum,observationSymbolNum);/ob 已知的观察序列；probability 可能性最大的隐藏状态序列的概率；返回 可能性最大的隐藏状态序列public List viterbi(int ob, double probability) double delta = null; int psi = null; return viterbi(ob, delta, psi, probability); /delta 输出中间结果，局部最大概率；psi 输出中间结果，反向指针指示最可能路径；返回可能性最大的隐藏状态序列的概率public List viterbi(int ob, double delta, int psi,double probability) delta = new doubleob.lengthN; / 局部概率 psi = new intob.lengthN; / 反向指针 System.out.println(1.初始化：); / 1. 初始化 for (int j = 0; j N; j+) delta0j = PIj * Bjob0; System.out.println(delta0+j+:+delta0j); / 2. 递归 System.out.println(2.归纳：); for (int t = 1; t ob.length; t+) for (int j = 0; j N; j+) double MaxValue = deltat - 10 * A0j;/初始，设第0个状态处的值为最大值 int MaxValueIndex = 0;/存储取得最大值处的状态的索引 for (int i = 1; i MaxValue) MaxValue = Value; MaxValueIndex = i; deltatj = MaxValue * Bjobt; System.out.println(delta+t+j+:+deltatj); psitj = MaxValueIndex; / 记录下最有可能到达此状态的上一个状态 / 3. 终止 System.out.println(3.终止，回溯求最佳路径); int q= new intob.length; / 定义最佳路径 probability = deltaob.length - 10;/将第0个定义为最大的权值 qob.length - 1 = 0; for (int i = 1; i probability) probability = deltaob.length - 1i; qob.length - 1 = i;/最后一个时间点的最优状态 / 4. 路径回溯 for (int t = ob.length - 2; t = 0; t-) qt = psit + 1qt + 1; List list=new ArrayList(); list.add(q); list.add(probability); return list; /前向 后项和维特比算法测试package com.HMM.test;import java.util.List;public class TestHMM2 enum Box one,two,three ; / 隐藏状态（箱子编号） enum Color red,yellow,blue,green ; / 观察状态（观测到的颜色值） public static void main(String args) / 测试前向算法和后向算法/ CheckForwardAndBackward();/ Console.WriteLine(); / 测试维特比算法 CheckViterbi();/ Console.WriteLine(); / 测试前向算法和后向算法 static void CheckForwardAndBackward() / 状态转移矩阵 double A = 0.500, 0.375, 0.125, 0.250, 0.125, 0.625, 0.250, 0.375, 0.375 ; / 混淆矩阵 double B = 0.60, 0.20, 0.15, 0.05, 0.25, 0.25, 0.25, 0.25, 0.05, 0.10, 0.35, 0.50 ; / 初始概率向量 double PI = 0.63,0.17,0.20; / 观察序列 int OB = Color.red.ordinal(), Color.yellow.ordinal(), Color.blue.ordinal(); System.out.println(状态转移概率矩阵：); for(int i=0;iA.length;i+) for(int j=0;jA0.length;j+) System.out.print(Aij+t); System.out.println(); System.out.println(符号观测概率矩阵：); for(int i=0;iB.length;i+) for(int j=0;jB0.length;j+) System.out.print(Bij+t); System.out.println(); System.out.println(初始概率向量：+PI0+ +PI1+ +PI2+); System.out.println(隐藏状态序列：+Box.one+ +Box.two+ +Box.three+); System.out.println(观测序列：+Color.red+ +Color.yellow+ +Color.blue+); / 初始化HMM模型 Forward forward = new Forward(A.length, B0.length); forward.A=A; forward.B=B; forward.PI=PI; / 观察序列的概率 System.out.println(-前向算法-); double probability = forward.forward(OB); Backward backWard = new Backward(A.length, B0.length); backWard.A=A; backWard.B=B; backWard.PI=PI; / 观察序列的概率 System.out.println(-后向算法-); probability = backWard.backward(OB); / 测试维特比算法 static void CheckViterbi() / 状态转移矩阵 double A = 0.500, 0.250, 0.250, 0.375, 0.125, 0.375, 0.125, 0.675, 0.375 ; / 混淆矩阵 double B = 0.60, 0.20, 0.15, 0.05, 0.25, 0.25, 0.25, 0.25, 0.05, 0.10, 0.35, 0.50 ; / 初始概率向量 double PI = 0.63, 0.17, 0.20 ; / 观察序列 int OB = Color.red.ordinal(),Color.yellow.ordinal(), Color.blue.ordinal(), Color.yellow.ordinal(),Color.green.ordinal() ; System.out.println(状态转移概率矩阵：); for(int i=0;iA.length;i+) for(int j=0;jA0.length;j+) System.out.print(Aij+t); System.out.println(); System.out.println(符号观测概率矩阵：); for(int i=0;iB.length;i+) for(int j=0;jB0.length;