-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathstan_result_postprocessing.py
211 lines (183 loc) · 6.61 KB
/
stan_result_postprocessing.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
import numpy as np
import pandas as pd
import os
import pickle
import time
import matplotlib.pyplot as plt
from utils import timeit
import utils
from multiprocessing import Pool, Manager
def load_stan_models(mcmc_dir, max_num_class):
'''
Args:
mcmc_dir ():
max_num_class ():
Return:
dict_stan_models ():
'''
dict_stan_models = {}
for k in range(1, max_num_class+1):
if os.path.exists(mcmc_dir + '/stan_model_{}.pkl'.format(k)):
dict_stan_models[k] = pickle.load(open(mcmc_dir + '/stan_model_{}.pkl'.format(k),'rb'))
return dict_stan_models
def load_fitted_models(mcmc_dir, max_num_class):
'''
Args:
mcmc_dir ():
max_num_class ():
Return:
dict_fitted_models ():
'''
dict_fitted_models = {}
for k in range(1, max_num_class+1):
if os.path.exists(mcmc_dir + '/fitted_model_{}.pkl'.format(k)):
dict_fitted_models[k] = pickle.load(open(mcmc_dir + '/fitted_model_{}.pkl'.format(k),'rb'))
return dict_fitted_models
def load_ex(result_ex_dir, max_num_class):
'''
Args:
result_ex_dir ():
max_num_class ():
Returns:
dict_emp_bayes ():
dict_log_liks ():
'''
dict_emp_bayes = {}
dict_log_liks = {}
for k_class in range(1, max_num_class+1):
if os.path.exists(result_ex_dir + '/dict_emp_bayes_{}.pkl'.format(k_class)):
dict_emp_bayes[k_class] = pickle.load(open(result_ex_dir + '/dict_emp_bayes_{}.pkl'.format(k_class),'rb'))
dict_log_liks[k_class] = pickle.load(open(result_ex_dir + '/dict_log_liks_{}.pkl'.format(k_class),'rb'))
return dict_emp_bayes, dict_log_liks
@timeit
def multi_cont_extract_fitted_model(dict_stan_models,
dict_fitted_models,
size,
result_ex_dir,
max_num_class,
multiprocess_mode_flag):
'''
Args:
dict_stan_models ():
dict_fitted_models ():
size ():
result_ex_dir ():
max_num_class ():
multiprocess_mode_flag ():
Return:
'''
if multiprocess_mode_flag:
k_classes = list(range(1, max_num_class+1))
n_proc = max_num_class
pool = Pool(processes=n_proc)
list_args_ex = []
for k_class in k_classes:
list_args_ex.append((dict_stan_models, dict_fitted_models, size, result_ex_dir, k_class))
pool.map(wrap_extract_fitted_model, list_args_ex)
else:
for k_class in range(1, max_num_class+1):
extract_fitted_model(dict_stan_models, dict_fitted_models, size, result_ex_dir, k_class)
return
def wrap_extract_fitted_model(args_ex):
'''
Wrapper of fitted model extraction for multiprocessing mode.
Args:
args_ex
Return:
'''
extract_fitted_model(*args_ex)
return
def extract_fitted_model(dict_stan_models, dict_fitted_models, size, result_ex_dir, k_class):
'''
Entity of fitted model extraction
Args:
dict_stan_models ():
dict_fitted_models ():
size ():
result_ex_dir ():
k_class ():
Return:
'''
ex_flag = os.path.exists(result_ex_dir + '/dict_log_liks_{}.pkl'.format(k_class))
if (k_class in dict_fitted_models.keys())&(ex_flag==False):
stan_model = dict_stan_models[k_class]
fitted_model = dict_fitted_models[k_class]
fitted_model_ex = fitted_model.extract()
dict_emp_bayes = {}
for param in fitted_model_ex.keys():
if param not in ['class_lp', 'log_lik', 'lp__']:
dict_emp_bayes[param] = fitted_model_ex[param].mean(axis=0)
dict_log_liks = {}
dict_log_liks['log_lik'] = fitted_model_ex['log_lik']
dict_log_liks['class_lp'] = fitted_model_ex['class_lp']
np_gamma = np.zeros((size, k_class))
for k in range(k_class):
if k_class==1:
np_gamma += 1
else:
epsilon = 1e-6
np_gamma[:,k] = ((np.exp(dict_log_liks['class_lp'][:,:,k]) + epsilon / k_class) /\
(np.exp(dict_log_liks['log_lik']) + epsilon)).mean(axis=0)
dict_log_liks['gamma'] = np_gamma
pickle.dump(dict_emp_bayes, open(result_ex_dir + '/dict_emp_bayes_{}.pkl'.format(k_class),'wb'))
pickle.dump(dict_log_liks, open(result_ex_dir + '/dict_log_liks_{}.pkl'.format(k_class),'wb'))
return
@timeit
def wbic(dict_log_liks, max_num_class):
'''
wbic用のMCMCが必要
Args:
dict_log_liks ():
max_num_class ():
Return:
list_wbic ():
'''
list_wbic = []
for k in range(1, max_num_class+1):
if k in dict_log_liks.keys():
log_lik = dict_log_liks[k]['log_lik']
wbic = - np.mean(log_lik.sum(axis=1))
list_wbic.append(round(wbic, 3))
else:
list_wbic.append(np.nan)
return list_wbic
@timeit
def param_num_count(dict_emp_bayes, max_num_class):
'''
Args:
dict_emp_bayes ():
max_num_class ():
Return:
list_num_param ():
'''
list_num_param = []
for k in range(1, max_num_class+1):
if k in dict_emp_bayes.keys():
num_count = 0
for param in dict_emp_bayes[k].keys():
if param not in ['phi_sigma_X_cont', 'beta1_0']:
if param == 'pi':
num_count += dict_emp_bayes[k][param].size - 1
elif 'phi_X_disc' in param:
num_count += dict_emp_bayes[k][param].size - k
else:
num_count += dict_emp_bayes[k][param].size
list_num_param.append(num_count)
else:
list_num_param.append(np.nan)
return list_num_param
def plot_by_class(list_y, y_name, max_num_class, result_summary_dir, fontsize=36):
xdata = list(range(1,max_num_class+1,1))
plt.figure(figsize=(8, 8), dpi=300)
plt.plot(xdata, list_y, linewidth=5)
# plt.title('{} vs num_class'.format(y_name), fontsize=fontsize)
plt.xlabel('Number of mixture components', fontsize=fontsize)
plt.ylabel(y_name, fontsize=fontsize)
plt.xticks(fontsize=fontsize*0.75)
plt.yticks(fontsize=fontsize*0.75)
plt.grid(True)
for axis in ['top', 'bottom', 'left', 'right']:
plt.gca().spines[axis].set_linewidth(4)
plt.savefig(result_summary_dir + '/num_class_vs_{}.png'.format(y_name))
# plt.show()
return