机器学习:Viterbi(维特比)实现程序

nianzai — Fri, 07 Sep 2012 08:43:00 GMT

wikipedia上有个java版的Viterbi(维特比)实现程序（http://en.wikipedia.org/wiki/Viterbi_algorithm），但是3个观察序列会标注出4个状态序列。
下面本人写的这个Viterbi(维特比)实现程序就没这个问题，3个观察序列就只标注出3个状态序列。

public class Viterbi
{
    public static void main(String[] args)
    {
        String[] states = {"Rainy", "Sunny"};
        String[] observations = {"walk", "shop", "clean"};
        double[] start_probability = {0.6, 0.4};
        double[][] transition_probability = {{0.7, 0.3}, {0.4, 0.6}};
        double[][] emission_probability = {{0.1, 0.4, 0.5}, {0.6, 0.3, 0.1}};
        forward_viterbi(observations,states,start_probability,transition_probability,emission_probability);
    }

    public static void  forward_viterbi(String[] observations, String[] states,double[] start_probability, double[][] transition_probability, double[][] emission_probability)
    {
        int[][] path=new int[observations.length][states.length];
        double[][] r=new double[observations.length][states.length];
        for(int j=0;j<states.length;j++)
        {
            r[0][j]=start_probability[j]*emission_probability[j][0];
            path[0][j]=0;
        }

        for(int t=1;t<observations.length;t++)
        {
            for(int i=0;i<states.length;i++)
            {
                double tmp=0;int m=0;
                for(int j=0;j<states.length;j++)
                {
                    double tem=r[t-1][j]*transition_probability[j][i]*emission_probability[i][t];
                    if(tem>tmp)
                    {
                        tmp=tem;
                        m=j;
                    }
                }
                r[t][i]=tmp;
                path[t][i]=m;
            }
        }

        double p=0;int m=0;
        for(int i=0;i<r[0].length;i++)
        {
            if(r[r.length-1][i]>p)
            {
                p=r[r.length-1][i];
                m=i;
            }
        }

        System.out.println("p="+p);

        int[] trace=new int[observations.length];
        trace[observations.length-1]=m;
        for(int t=observations.length-1;t>0;t--)
        {
            trace[t-1]=path[t][m];
            m=path[t][m];
        }

        for(int i=0;i<trace.length;i++)
            System.out.println(states[trace[i]]);
    }
}

。

nianzai 2012-09-07 16:43 发表评论

BlogJava-lucene + hadoop 分布式并行计算搜索框架-随笔分类-机器学习

机器学习:Viterbi(维特比)实现程序