统计学习方法 第十章习题答案

10.1

题目：

给定盒子和球组成的隐马尔可夫模型$\lambda =(A,B,\pi )$，其中，$A=\left[\begin{array}{ccc}0.5& 0.2& 0.3\\ 0.3& 0.5& 0.2\\ 0.2& 0.3& 0.5\end{array}\right],B=\left[\begin{array}{cc}0.5& 0.5\\ 0.4& 0.6\\ 0.7& 0.3\end{array}\right],\pi =(0.2,0.4,0.4{)}^T$
设$T=4,O=(红，白，红，白)$，试用后向算法计算$P(O|\lambda )$

解答

按照算法10.3
（1）计算初值
${\beta }_4(i)=1i=1,2,3$
（2）递推计算
${\beta }_3(1)={\sum }_{j=1}^3{a}_{1j}{b}_j(O_4){\beta }_4(j)=0.25+0.12+0.09=0.46$
${\beta }_3(2)={\sum }_{j=1}^3{a}_{2j}{b}_j(O_4){\beta }_4(j)=0.15+0.3+0.06=0.51$
${\beta }_3(3)={\sum }_{j=1}^3{a}_{3j}{b}_j(O_4){\beta }_4(j)=0.1+0.18+0.15=0.43$

${\beta }_2(1)={\sum }_{j=1}^3{a}_{1j}{b}_j(O_3){\beta }_3(j)=0.25\ast 0.46+0.08\ast 0.51+0.21\ast 0.43=0.2461$
${\beta }_2(2)={\sum }_{j=1}^3{a}_{2j}{b}_j(O_3){\beta }_3(j)=0.15\ast 0.46+0.2\ast 0.51+0.14\ast 0.43=0.2312$
${\beta }_2(3)={\sum }_{j=1}^3{a}_{3j}{b}_j(O_3){\beta }_3(j)=0.1\ast 0.46+0.12\ast 0.51+0.35\ast 0.43=0.2577$

${\beta }_1(1)={\sum }_{j=1}^3{a}_{1j}{b}_j(O_2){\beta }_2(j)=0.25\ast 0.2461+0.12\ast 0.2312+0.09\ast 0.2577=0.112462$
${\beta }_1(2)={\sum }_{j=1}^3{a}_{2j}{b}_j(O_2){\beta }_2(j)=0.15\ast 0.2461+0.3\ast 0.2312+0.06\ast 0.2577=0.121737$
${\beta }_1(3)={\sum }_{j=1}^3{a}_{3j}{b}_j(O_2){\beta }_2(j)=0.1\ast 0.2461+0.18\ast 0.2312+0.15\ast 0.2577=0.104881$
（3）终止
$P(O|\lambda )={\sum }_{i=1}^3{\pi }_i{b}_i(O_1){\beta }_1(i)=0.2\ast 0.5\ast 0.112462+0.4\ast 0.4\ast 0.121737+0.4\ast 0.7\ast 0.104881=0.0600908$

10.2

题目

考虑盒子和球组成的隐马尔可夫模型$\lambda =(A,B,\pi )$，其中，$A=\left[\begin{array}{ccc}0.5& 0.1& 0.4\\ 0.3& 0.5& 0.2\\ 0.2& 0.2& 0.6\end{array}\right],B=\left[\begin{array}{cc}0.5& 0.5\\ 0.4& 0.6\\ 0.7& 0.3\end{array}\right],\pi =(0.2,0.3,0.5{)}^T$
设$T=8,O=(红，白，红，红，白，红，白，白)$，试用前向后向算法计算$P(i_4=q_3|O,\lambda )$

解答

这题就偷个小懒，让python帮忙解决叭(╹▽╹)
代码：

import numpy as np

def fore_algorithm(A, B, p_i, o, T, N):
    # 设初值，alpha_1(i) = pt_(i)b_i(o(i))
    alpha = np.zeros((T, N))
    for i in range(N):
        h = o[0]
        alpha[0][i] = p_i[i] * B[i][h]
    #递推
    for t in range(T-1):
        h = o[t+1]
        for i in range(N):
            a = 0
            for j in range(N):
                a += (alpha[t][j] * A[j][i])
            alpha[t+1][i] = a * B[i][h]
    #终止
    P = 0
    for i in range(N):
        P += alpha[T-1][i]
    return P, alpha

def back_algorithm(A, B, p_i, o, T, N):
    #设置初值，beta_t(i)=1
    beta = np.ones((T, N))
    #递推
    for t in range(T-1):
        t = T - t - 2
        h = o[t + 1]
        h = int(h)

        for i in range(N):
            beta[t][i] = 0
            for j in range(N):
                beta[t][i] += A[i][j] * B[j][h] * beta[t+1][j]
    #终止
    P = 0
    for i in range(N):
        h = o[0]
        h = int(h)
        P += p_i[i] * B[i][h] * beta[0][i]
    return P, beta

if __name__ == "__main__":
    T = 8
    N = 3
    A = [[0.5, 0.1, 0.4], [0.3, 0.5, 0.2], [0.2, 0.2, 0.6]]
    B = [[0.5, 0.5], [0.4, 0.6], [0.7, 0.3]]
    pi = [0.2, 0.3, 0.5]
    O = ['红', '白', '红', '红', '白', '红', '白', '白']
    o = np.zeros(T, np.int)
    for i in range(T):
        if O[i] == '白':
            o[i] = 1
        else:
            o[i] = 0
    PF, alpha = fore_algorithm(A, B, pi, o, T, N)
    PB, beta = back_algorithm(A, B, pi, o, T, N)
    print("PF:", PF, "PB:", PB)
    #P(i_4=q_3|O,\lambda) = alpah_4(3)* beta_4(3)
    P = alpha[4-1][3-1] * beta[4-1][3-1]
    print("前向后向概率计算可得 P(i4=q3|O,lambda)=", P / PF)

运行结果：
PF为前向算法得到的$P(O|\lambda )$
PB为后向算法得到的$P(O|\lambda )$
（改一下main函数中的输出值，可以验算一下第一题的结果）

10.3

题目

在习题10.1中，试用维特比算法求最优路径

解答

（1）初始化
${\delta }_1(1)={\pi }_1\ast b_1(o_1)=0.2\ast 0.5=0.1$
${\delta }_1(2)={\pi }_2\ast b_2(o_1)=0.4\ast 0.4=0.16$
${\delta }_1(3)={\pi }_3\ast b_3(o_1)=0.4\ast 0.7=0.28$

（2）递推
${\delta }_t(i)=ma{x}_{1\le j\le N}[{\delta }_{t-1}(j)a_{ji}]b_i(o_t)$

t = 2
δ 2 ( 1 ) = m a x { 0.1 ∗ 0.5 , 0.16 ∗ 0.3 , 0.28 ∗ 0.2 } ∗ 0.5 = 0.028 ; ψ 2 ( 1 ) = 3 \delta_2(1) = max\{0.1*0.5, 0.16*0.3, 0.28*0.2\}*0.5 = 0.028; \quad\psi_2(1) = 3 δ2​(1)=max{0.1∗0.5,0.16∗0.3,0.28∗0.2}∗0.5=0.028;ψ2​(1)=3
δ 2 ( 2 ) = m a x { 0.1 ∗ 0.2 , 0.16 ∗ 0.5 , 0.28 ∗ 0.3 } ∗ 0.6 = 0.0504 ; ψ 2 ( 2 ) = 3 \delta_2(2) = max\{0.1*0.2, 0.16*0.5, 0.28*0.3\}*0.6 = 0.0504;\quad\psi_2(2) = 3 δ2​(2)=max{0.1∗0.2,0.16∗0.5,0.28∗0.3}∗0.6=0.0504;ψ2​(2)=3
δ 2 ( 3 ) = m a x { 0.1 ∗ 0.3 , 0.16 ∗ 0.2 , 0.28 ∗ 0.5 } ∗ 0.3 = 0.042 ; ψ 2 ( 3 ) = 3 \delta_2(3) = max\{0.1*0.3, 0.16*0.2, 0.28*0.5\}*0.3 = 0.042;\quad\psi_2(3) = 3 δ2​(3)=max{0.1∗0.3,0.16∗0.2,0.28∗0.5}∗0.3=0.042;ψ2​(3)=3

t=3
δ 3 ( 1 ) = m a x { 0.028 ∗ 0.5 , 0.0504 ∗ 0.3 , 0.042 ∗ 0.2 } ∗ 0.5 = 0.00756 ; ψ 3 ( 1 ) = 2 \delta_3(1) = max\{0.028*0.5, 0.0504*0.3, 0.042*0.2\}*0.5 = 0.00756; \quad\psi_3(1) = 2 δ3​(1)=max{0.028∗0.5,0.0504∗0.3,0.042∗0.2}∗0.5=0.00756;ψ3​(1)=2
δ 3 ( 2 ) = m a x { 0.028 ∗ 0.2 , 0.0504 ∗ 0.5 , 0.042 ∗ 0.3 } ∗ 0.4 = 0.01008 ; ψ 3 ( 2 ) = 2 \delta_3(2) = max\{0.028*0.2, 0.0504*0.5, 0.042*0.3\}*0.4 = 0.01008; \quad\psi_3(2) = 2 δ3​(2)=max{0.028∗0.2,0.0504∗0.5,0.042∗0.3}∗0.4=0.01008;ψ3​(2)=2
δ 3 ( 3 ) = m a x { 0.028 ∗ 0.3 , 0.0504 ∗ 0.2 , 0.042 ∗ 0.5 } ∗ 0.7 = 0.0147 ; ψ 3 ( 3 ) = 3 \delta_3(3) = max\{0.028*0.3, 0.0504*0.2, 0.042*0.5\}*0.7 = 0.0147; \quad\psi_3(3) = 3 δ3​(3)=max{0.028∗0.3,0.0504∗0.2,0.042∗0.5}∗0.7=0.0147;ψ3​(3)=3

t=4
δ 4 ( 1 ) = m a x { 0.00756 ∗ 0.5 , 0.01008 ∗ 0.3 , 0.0147 ∗ 0.2 } ∗ 0.5 = 0.00189 ; ψ 4 ( 1 ) = 1 \delta_4(1) = max\{0.00756*0.5, 0.01008*0.3, 0.0147*0.2\}*0.5 = 0.00189; \quad\psi_4(1) = 1 δ4​(1)=max{0.00756∗0.5,0.01008∗0.3,0.0147∗0.2}∗0.5=0.00189;ψ4​(1)=1
δ 4 ( 2 ) = m a x { 0.00756 ∗ 0.2 , 0.01008 ∗ 0.5 , 0.0147 ∗ 0.3 } ∗ 0.6 = 0.003024 ; ψ 4 ( 2 ) = 2 \delta_4(2) = max\{0.00756*0.2, 0.01008*0.5, 0.0147*0.3\}*0.6 = 0.003024;\quad\psi_4(2) = 2 δ4​(2)=max{0.00756∗0.2,0.01008∗0.5,0.0147∗0.3}∗0.6=0.003024;ψ4​(2)=2
δ 4 ( 3 ) = m a x { 0.00756 ∗ 0.3 , 0.01008 ∗ 0.2 , 0.0147 ∗ 0.5 } ∗ 0.3 = 0.002205 ; ψ 4 ( 3 ) = 3 \delta_4(3) = max\{0.00756*0.3, 0.01008*0.2, 0.0147*0.5\}*0.3 = 0.002205;\quad\psi_4(3) = 3 δ4​(3)=max{0.00756∗0.3,0.01008∗0.2,0.0147∗0.5}∗0.3=0.002205;ψ4​(3)=3

（3）终止
$P^{\ast }=ma{x}_{1\le i\le 3}{\delta }_4(i)=0.003024$
$i_4^{\ast }=argma{x}_i[{\delta }_4(i)]=2$

（4）最优路径回溯
$i_3^{\ast }={\psi }_4(i_4^{\ast })=2$
$i_2^{\ast }={\psi }_3(i_3^{\ast })=2$
$i_1^{\ast }={\psi }_2(i_2^{\ast })=3$

$I^{\ast }=(i_1^{\ast },i_2^{\ast },i_3^{\ast },i_4^{\ast })=(3,2,2,2)$
用代码验算一下~~

import numpy as np

def viterbi(A, B, pai, o, T, N):
    a = []
    i_star = np.zeros(T, np.int)
    # 初始化
    delta = np.zeros((T, N))
    psi = np.zeros((T, N), np.int)

    for i in range(N):
        delta[0][i] = pai[i] * B[i][o[0]]
        psi[0][i] = 0

    # 递推
    for t in range(T-1):
        t = t + 1
        for i in range(N):
            for j in range(N):
                a.append(delta[t-1][j] * A[j][i])
            delta[t][i] = np.max(a) * B[i][o[t]]
            psi[t][i] = np.argmax(a, axis=0)
            a = []
    psi = psi + 1

    # 终止
    P_star = np.max(delta[T-1])
    i_star[T-1] = np.argmax(delta[T-1], axis=0) +1

    # 最优路径回溯
    for t in range(T-1):
        t = T - t - 2
        a = t + 1
        b = i_star[t+1]-1
        i_star[t] = psi[a][b]

    return i_star, P_star

if __name__ == "__main__":
    T = 4
    N = 3
    O = ['红', '白', '红', '白']
    A = [[0.5, 0.2, 0.3],
         [0.3, 0.5, 0.2],
         [0.2, 0.3, 0.5]]
    B = [[0.5, 0.5],
         [0.4, 0.6],
         [0.7, 0.3]]
    p_i = [0.2, 0.4, 0.4]
    o = np.zeros(T, np.int)
    for i in range(T):
        if O[i] == '白':
            o[i] = 1
        else:
            o[i] = 0
    I, P = viterbi(A, B, p_i, o, T, N)
    print('最优路径： ')
    print(I)
    print('最优路径的概率：')
    print(P)

运行结果：

10.4

题目

试用前向概率和后向概率推导

$P(O|\lambda )={\sum }_{i=1}^N{\sum }_{j=1}^N{\alpha }_t(i)a_{ij}{b}_j(o_{t+1}){\beta }_{t+1}(j),t=1,2,...,T-1$
这个式子是在书中的公式(10.21)，书上是用前向概率和后向概率文字推导得出，${\alpha }_t(i)a_{ij}{b}_j(o_{t+1}){\beta }_{t+1}(j)$等价于$p(o_1,o_2,...,o_T,i_t=q_i,i_{t+1}=q_j|\lambda )$，两次求和则$q_i,q_j$取到所有可能的情况。
详细的数学证明可以参考：https://blog.csdn.net/weixin_35479108/article/details/87907429
这里补充一个数学归纳法的证明
（1）t = 1
$$\begin{gathered} P(O|\lambda ) \\ ={\sum }_{i=1}^N{\sum }_{i=1}^N{\alpha }_1(i)a_{ij}{b}_j(o_2){\beta }_2(j) \\ ={\sum }_{i=1}^N{\alpha }_1(i){\sum }_{i=1}^N{a}_{ij}{b}_j(o_2){\beta }_2(j) \\ ={\sum }_{i=1}^N{\alpha }_1(i){\beta }_1(i) \\ ={\sum }_{i=1}^N{\pi }_i{b}_i(o_1){\beta }_1(i) \end{gathered}$$
这个即为公式（10.21），后向算法的定义式，满足条件
（2）假设t=k时等式成立（t = 1,2,…T-2)
即$P(O|\lambda )={\sum }_{i=1}^N{\sum }_{j=1}^N{\alpha }_k(i)a_{ij}{b}_j(o_{k+1}){\beta }_{k+1}(j)$
那么对于t=k+1，有
$$\begin{gathered} {\sum }_{i=1}^N{\sum }_{j=1}^N{\alpha }_{k+1}(i)a_{ij}{b}_j(o_{k+2}){\beta }_{k+2}(j) \\ ={\sum }_{i=1}^N{\alpha }_{k+1}(i){\sum }_{j=1}^N{a}_{ij}{b}_j(o_{k+2}){\beta }_{k+2}(j) \\ ={\sum }_{i=1}^N{\alpha }_{k+1}(i){\beta }_{k+1}(i) \\ ={\sum }_{i=1}^N{\beta }_{k+1}(i)[{\sum }_{j=1}^N{\alpha }_k(j)a_{ij}]b_i(o_{k+1}) \\ ={\sum }_{i=1}^N{\sum }_{j=1}^N{\alpha }_k(i)a_{ij}{b}_j(o_{k+1}){\beta }_{k+1}(j) \\ =P(O|\lambda ) \end{gathered}$$

得证

解答

10.5

题目

比较维特比算法中变量$\delta$的计算和前向算法中变量$\alpha$的计算的主要区别.

解答

计算变量$\alpha$的时候直接对上个的结果进行数值计算，而计算变量$\delta$需要在上个结果计算的基础上选择最大值。