【求大神指点】使用某个字典变量中的某个键值时出错但是没找到未返回的值

t464961268

新手求教，希望各位大神不要嫌弃。。。时间紧迫所以羞愧的当半个伸手党。。会好好学习的哭唧唧。。

错误如下：

1. /Library/Frameworks/Python.framework/Versions/3.5/bin/python3.5 "/Applications/PyCharm CE.app/Contents/helpers/pydev/pydevd.py" --multiproc --qt-support --client 127.0.0.1 --port 59067 --file /Users/tt/Desktop/baum-welch.py
   
2. warning: Debugger speedups using cython not found. Run '"/Library/Frameworks/Python.framework/Versions/3.5/bin/python3.5" "/Applications/PyCharm CE.app/Contents/helpers/pydev/setup_cython.py" build_ext --inplace' to build.
   
3. pydev debugger: process 26567 is connecting
   
4. 
   
5. Connected to pydev debugger (build 171.4694.38)
   
6. Traceback (most recent call last):
   
7. at t=0, scale is [ 0.4995  0.      0.      0.      0.      0.      0.      0.      0.      0.    ]
   
8.   File "/Applications/PyCharm CE.app/Contents/helpers/pydev/pydevd.py", line 1591, in <module>
   
9. 0.333333333333
   
10. 0.5
    
11.     globals = debugger.run(setup['file'], None, None, is_module)
    
12. 0.166666666667
    
13.   File "/Applications/PyCharm CE.app/Contents/helpers/pydev/pydevd.py", line 1018, in run
    
14. 前向一次的局部概率矩阵是:  [[ 0.33333333  0.5         0.16666667]
    
15. [ 0.36363636  0.40909091  0.22727273]
    
16. [ 0.36144578  0.43373494  0.20481928]
    
17. [ 0.36334405  0.42684887  0.20980707]
    
18. [ 0.3211513   0.12639652  0.55245219]
    
19. [ 0.3274895   0.51073717  0.16177333]
    
20. [ 0.30538114  0.11378704  0.58083182]
    
21. [ 0.32955045  0.17497844  0.49547111]
    
22. [ 0.33154192  0.16520629  0.5032518 ]
    
23. [ 0.33293248  0.16687277  0.50019475]]
    
24. 前向求出的scale is : None
    
25.     pydev_imports.execfile(file, globals, locals)  # execute the script
    
26.   File "/Applications/PyCharm CE.app/Contents/helpers/pydev/_pydev_imps/_pydev_execfile.py", line 18, in execfile
    
27.     exec(compile(contents+"\n", file, 'exec'), glob, loc)
    
28.   File "/Users/tt/Desktop/baum-welch.py", line 72, in <module>
    
29.     backwardWithScale(forward_scale, observation_sequence, partial_prob_backward , log_prob_backward)  
    
30.   File "/Users/tt/Desktop/baum-welch.py", line 55, in backwardWithScale
    
31.     partial_prob[T-1][i] = 1.0/scale[T-1]
    
32. TypeError: 'NoneType' object is not subscriptable
    
33. 
    
34. Process finished with exit code 1
    

复制代码

代码如下：

1. # -*- coding: utf-8 -*-
   
2. """
   
3. Spyder Editor
   
4. 
   
5. This is a temporary script file.
   
6. """
   
7. 
   
8. import numpy as np
   
9. import math
   
10. class hmm:
    
11.     M = 2
    
12.     N = 3
    
13.     A =  [[0.500, 0.375, 0.125],  
    
14.           [0.250, 0.125, 0.625],  
    
15.           [0.250, 0.375, 0.375]]
    
16.     B = [[0.50, 0.50],  
    
17.          [0.75, 0.25],  
    
18.          [0.25, 0.75]]
    
19.     PI = [0.333, 0.333, 0.333]
    
20. observation_sequence = [0, 0, 0, 0, 1, 0, 1, 1, 1, 1]
    
21. T = len(observation_sequence)
    
22. partial_prob_forward = np.zeros((T, hmm.N))
    
23. log_prob_forward = 0.
    
24. 
    
25. def forwardWithScale(observation_sequence, partial_prob, log_prob_forward):
    
26.     scale = np.zeros(T)
    
27.     for i in range(hmm.N):
    
28.         partial_prob[0][i] = hmm.PI[i] * hmm.B[i][observation_sequence[0]]  
    
29.         scale[0] += partial_prob[0][i]
    
30.     print ('at t=0, scale is',scale)
    
31.     for i in range(hmm.N):
    
32.         partial_prob[0][i] /= scale[0]  
    
33.         print (partial_prob[0][i])
    
34.     for t in range(1, T):
    
35.         for j in range(hmm.N):
    
36.             temp = 0.
    
37.             for i in range(hmm.N):
    
38.                 temp += partial_prob[t-1][i] * hmm.A[i][j]
    
39.             partial_prob[t][j] = temp*hmm.B[j][observation_sequence[t]]  
    
40.             scale[t] += partial_prob[t][j]
    
41.         for j in range(hmm.N):
    
42.             partial_prob[t][j] /= scale[t]
    
43.     for t in range(T):
    
44.         log_prob_forward += math.log(scale[t])
    
45. 
    
46. forward_scale = forwardWithScale(observation_sequence, partial_prob_forward, log_prob_forward)  
    
47. print  ("前向一次的局部概率矩阵是: ", partial_prob_forward )
    
48. plogprobinit = log_prob_forward  #log P(O |intial model)
    
49. 
    
50. partial_prob_backward = np.zeros((T, hmm.N))  
    
51. log_prob_backward = 0.  
    
52.   
    
53. def backwardWithScale(scale, observation_sequence, partial_prob, log_prob_backward):
    
54.     for i in range(hmm.N):
    
55.         partial_prob[T-1][i] = 1.0/scale[T-1]
    
56.     t = T-2
    
57.     while t >= 0:
    
58.         for j in range(hmm.N):
    
59.             temp = 0.
    
60.             for i in range(hmm.N):
    
61.                 temp += partial_prob[t+1][i] * hmm.A[i][j] * hmm.B[j][observation_sequence[t+1]]
    
62.             partial_prob[t][j] = temp / scale[t]
    
63.         t = t-1
    
64.     for t in  range(T):
    
65.         log_prob_backward += partial_prob[1][i]
    
66.         
    
67. print ("前向求出的scale is :",  forward_scale)
    
68. backwardWithScale(forward_scale, observation_sequence, partial_prob_backward , log_prob_backward)  
    
69. print ("后向的局部概率矩阵是: ", partial_prob_backward )
    
70. 
    
71. gamma = np.zeros((T,hmm.N))  
    
72. def computeGamma(gamma):
    
73.     for t in range(T):
    
74.         denominator = 0.0
    
75.         for j in range(hmm.N):
    
76.             gamma[t][j] = partial_prob_forward[t][j] * partial_prob_backward[t][j]  
    
77.             denominator += gamma[t][j]
    
78.         for i in range(hmm.N):
    
79.             gamma[t][i] = gamma[t][i] / denominator
    
80.             
    
81. computeGamma(gamma)
    
82. print ("最后得到的gamma矩阵是: ", gamma )
    
83.   
    
84. xi = np.zeros((T, hmm.N, hmm.N))  
    
85. def computeXi(xi, observation_sequence):
    
86.     for t in range(T-1):
    
87.         temp = 0.  
    
88.         for i in range(hmm.N):
    
89.             for j in range(hmm.N):
    
90.                 #
    
91.                 xi[t][i][j] = partial_prob_forward[t][i]*hmm.A[i][j]*hmm.B[j][observation_sequence[t+1]]*partial_prob_backward[t+1][j]
    
92.                 temp += xi[t][i][j]
    
93.         for i in range(hmm.N):
    
94.             for j in range(hmm.N):
    
95.                 xi[t][i][j] = xi[t][i][j] / temp
    
96. computeXi(xi, observation_sequence)  
    
97. print ("最后得到的 \epsilon 矩阵是: ", xi )
    
98. 
    
99. print ("开始更新PI,  A ,  B,  动量是0.999 ")
    
100. for i in range(hmm.N):
     
101.     hmm.PI[i] = 0.001 + 0.999 * gamma[0][i]
     
102. print ("更新后的初始状态矩阵是: ", hmm.PI)
     
103. 
     
104. for i in range(hmm.N):
     
105.     denominatorA =  0.
     
106.     for t in range(T-1):
     
107.         denominatorA += gamma[t][i]
     
108.     for j in range(hmm.N):
     
109.         numeratorA = 0.
     
110.         for t in range(T-1):
     
111.             numeratorA += xi[t][i][j]
     
112.         hmm.A[i][j] = 0.001 + 0.999 * numeratorA / denominatorA
     
113.     denominatorB = denominatorA + gamma[T][i]  
     
114.     for k in range(hmm.M):
     
115.         numeratorB = 0.
     
116.         for t in range(T):
     
117.             if observation_sequence[t] == k:
     
118.                 numeratorB += gamma[t][i]
     
119.         hmm.B[i][k] = 0.001 + 0.999 * numeratorB / denominatorB
     
120.         
     
121. print( "更新后的状态转移矩阵A是: ", hmm.A)
     
122. print( "更新后的生成概率矩阵B是: ", hmm.B)
     

复制代码

万分感谢！！！！！！！！

sunslj

第46行，forward_scale 变量内容是None啊，forwardWithScale函数没有返回值啊