李航统计学习方法 隐马尔科夫 用后向算法实现例10.2

import numpy as np

# 状态转移矩阵
A = np.array([[0.5,0.2,0.3],[0.3,0.5,0.2],[0.2,0.3,0.5]])
# 观测矩阵
B = np.array([[0.5,0.5],[0.4,0.6],[0.7,0.3]])
# 初始矩阵
pi = np.array([0.2,0.4,0.4])
# 序列
O = np.array([0,1,0])

def backward(A,B,pi,O):
    m=np.shape(A)[0]
    T=len(O)
    belta=np.zeros((T, m))
    # 假设最后一个时刻的值均为1
    for i in range(m):
        belta[0][i]=1
    # 递推
    for t in range(1,T):
        for i in range(m):
            sum=0
            for j in range(m):
                sum+=A[i][j]*B[j][O[T-t]]*belta[t-1][j]
            belta[t][i]=sum
    # 计算最后的概率值
    P=0
    for i in range(m):
       P+=pi[i]*B[i][O[0]]*belta[-1][i]
    return P


print(backward(A,B,pi,O))

后向算法理论部分在李航统计学习方法上有详细的介绍，实践部分采用python实现