-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathnode_model.py
266 lines (236 loc) · 9.56 KB
/
node_model.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
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
#%%
import pystan
import matplotlib.pyplot as plt
import numpy as np
import json
from scipy import stats
from sklearn.metrics import *
import sys
class Node_model:
# def __init__(self,prior=(0,3),good=(0,1),bad=(5,3),pbad=0.1): # TODO: N(0,1) assumption
# cycle_code = """
# data {{
# int<lower=1> N; // Total number of nodes
# int<lower=1> M; // Total number of edges
# int<lower=1> src[M]; // Edge connectivity
# int<lower=1> dst[M];
# real y[M]; // measured edge data
# }}
# parameters {{
# vector[N] v; // value at each node
# }}
# model {{
# real pBad={pbad};
# v ~ normal({pu}, {ps});
# for (e in 1:M) {{
# target += log_sum_exp( log(pBad/2) + normal_lpdf( v[dst[e]] - v[src[e]] - y[e] | {nbu}, {bs}), // bad noise assumption
# log_sum_exp(
# log(1-pBad) + normal_lpdf( v[dst[e]] - v[src[e]] -y[e] | {gu}, {gs}), // good noise assumption
# log(pBad/2) + normal_lpdf( v[dst[e]] - v[src[e]] -y[e] | {pbu}, {bs}) ) ); // bad noise assumption
# }}
# }}
# """.format(pbad=pbad,pu=prior[0],ps=prior[1],gu=good[0],gs=good[1],nbu=-bad[0],pbu=bad[0],bs=bad[1])
# self.prior = prior
# self.good = good
# self.bad = bad
# self.pbad = pbad
# self.model = pystan.StanModel(model_code=cycle_code)
def __init__(self,prior=(0,3),noise=(0,1)): # TODO: N(0,1) assumption
if prior:
cycle_code = """
data {{
int<lower=1> N; // Total number of nodes
int<lower=1> M; // Total number of edges
int<lower=1> src[M]; // Edge connectivity
int<lower=1> dst[M];
real y[M]; // measured edge data
}}
parameters {{
vector[N] v; // value at each node
}}
model {{
v ~ normal({pu}, {ps});
for (e in 1:M) {{
target += normal_lpdf( v[dst[e]] - v[src[e]] - y[e] | {gu}, {gs});
}}
}}
""".format(pu=prior[0],ps=prior[1],gu=noise[0],gs=noise[1])
else:
cycle_code = """
data {{
int<lower=1> N; // Total number of nodes
int<lower=1> M; // Total number of edges
int<lower=1> src[M]; // Edge connectivity
int<lower=1> dst[M];
real y[M]; // measured edge data
}}
parameters {{
vector[N] v; // value at each node
}}
model {{
for (e in 1:M) {{
target += normal_lpdf( v[dst[e]] - v[src[e]] - y[e] | {gu}, {gs});
}}
}}
""".format(gu=noise[0],gs=noise[1])
self.prior = prior
self.model = pystan.StanModel(model_code=cycle_code)
def read_graph(self, graph_dir):
start = []
end = []
y = []
graph = open(graph_dir)
idx = 0
self.node2idx = dict()
self.idx2node = dict()
for line in graph:
s,e,v = line.rstrip().split(' ')
if s not in self.node2idx:
self.node2idx[s] = idx
self.idx2node[idx] = s
idx += 1
if e not in self.node2idx:
self.node2idx[e] = idx
self.idx2node[idx] = e
idx += 1
start.append(self.node2idx[s]+1)
end.append(self.node2idx[e]+1)
y.append(float(v))
self.cycle_dat = {
'N':len(self.node2idx),
'M':len(y),
'src':start,
'dst':end,
'y': y }
def sample(self, it=1000,chains=4,verbose=True):
fit = self.model.sampling(data=self.cycle_dat, iter=it, chains=chains,verbose=verbose)
self.result = fit
self.sample = True
return fit
def optimize(self):
fit = self.model.optimizing(data=self.cycle_dat)
self.result = fit
self.sample = False
return fit
def pBgV(self, pred_diff,obs_diff):
noise = obs_diff-pred_diff
pB = 0.5*stats.norm.pdf(noise,*self.bad)+0.5*stats.norm.pdf(noise,-self.bad[0],self.bad[1])
pB *= self.pbad
pG = stats.norm.pdf(noise,*self.good)*(1-self.pbad)
return pB/(pB+pG)
def predict(self):
samples = self.result.extract()['v']
start = self.cycle_dat['src']
end = self.cycle_dat['dst']
y = self.cycle_dat['y']
edge_pred = dict()
for i in range(len(start)):
s = start[i]-1
e = end[i]-1
edge_pred[str((s,e))] = self.pBgV(samples[:,e]-samples[:,s],y[i]).mean()
self.pred = edge_pred
return edge_pred
# def predict_val(self):
# samples = self.result.extract()['v']
# start = self.cycle_dat['src']
# end = self.cycle_dat['dst']
# y = self.cycle_dat['y']
# edge_pred = dict()
# for i in range(len(start)):
# s = start[i]-1
# e = end[i]-1
# edge_pred[str((s,e))] = (samples[:,e]-samples[:,s]).mean()
# self.pred = edge_pred
# return edge_pred
def predict_val(self):
if self.sample:
samples = self.result.extract()['v']
else:
samples = self.result['v']
start = self.cycle_dat['src']
end = self.cycle_dat['dst']
y = self.cycle_dat['y']
edge_pred = dict()
for i in range(len(start)):
s = self.idx2node[start[i]-1]
e = self.idx2node[end[i]-1]
if not self.sample:
edge_pred[(s,e)] = samples[end[i]-1]-samples[start[i]-1]
else:
edge_pred[(s,e)] = (samples[:,end[i]-1]-samples[:,start[i]-1]).mean()
self.pred = edge_pred
return edge_pred
def auc(self,truth):
order = truth.keys()
edge_truth_arr = np.array([truth[k][0] for k in order])
edge_pred_arr = np.array([self.pred[k] for k in order])
return roc_auc_score(edge_truth_arr,edge_pred_arr)
def auc_plot(self,truth):
order = truth.keys()
edge_truth_arr = np.array([truth[k][0] for k in order])
edge_pred_arr = np.array([self.pred[k] for k in order])
fpr, tpr, thresholds = roc_curve(edge_truth_arr, edge_pred_arr)
plt.plot(fpr,tpr)
plt.plot([0, 1], [0, 1], linestyle='--',color='r')
def recall(self,truth,thres=0.5):
order = truth.keys()
edge_truth_arr = np.array([truth[k][0] for k in order])
edge_pred_arr = np.array([self.pred[k] for k in order])
pos_pred = edge_pred_arr[edge_truth_arr==1]
plt.hist(pos_pred,density=True,color='r',histtype='step')
print(f'TP/TP+FN: {(pos_pred>thres).sum()/len(pos_pred)}')
def precision(self,truth,thres=0.5):
order = truth.keys()
edge_truth_arr = np.array([truth[k][0] for k in order])
edge_pred_arr = np.array([self.pred[k] for k in order])
print(f'TP/TP+FP: {precision_score(edge_truth_arr,edge_pred_arr>thres)}')
def confusion_matrix(self,truth,thres=0.5,normalize=None):
order = truth.keys()
edge_truth_arr = np.array([truth[k][0] for k in order])
edge_pred_arr = np.array([self.pred[k] for k in order])
ConfusionMatrixDisplay(confusion_matrix(edge_truth_arr,edge_pred_arr>thres,normalize=normalize),['Good','Bad']).plot()
return confusion_matrix(edge_truth_arr,edge_pred_arr>thres,normalize=normalize)
class Node_model_Uniform(Node_model):
def __init__(self, prior=(-1,1), noise=(0,1)):
cycle_code = """
data {{
int<lower=1> N; // Total number of nodes
int<lower=1> M; // Total number of edges
int<lower=1> src[M]; // Edge connectivity
int<lower=1> dst[M];
real y[M]; // measured edge data
}}
parameters {{
vector<lower={pu},upper={ps}>[N] v; // value at each node
}}
model {{
v ~ uniform({pu}, {ps});
for (e in 1:M) {{
target += normal_lpdf( v[dst[e]] - v[src[e]] - y[e] | {gu}, {gs}); // bad noise assumption
}}
}}
""".format(pu=prior[0],ps=prior[1],gu=noise[0],gs=noise[1])
self.prior = prior
self.model = pystan.StanModel(model_code=cycle_code)
class Node_model_Gamma(Node_model):
def __init__(self, prior=(2.,2.), noise=(0,1)):
cycle_code = """
data {{
int<lower=1> N; // Total number of nodes
int<lower=1> M; // Total number of edges
int<lower=1> src[M]; // Edge connectivity
int<lower=1> dst[M];
real y[M]; // measured edge data
}}
parameters {{
vector<lower=0>[N] v; // value at each node
}}
model {{
v ~ gamma({pu}, {ps});
for (e in 1:M) {{
target += normal_lpdf( v[dst[e]] - v[src[e]] - y[e] | {gu}, {gs}); // bad noise assumption
}}
}}
""".format(pu=prior[0],ps=prior[1],gu=noise[0],gs=noise[1])
self.prior = prior
self.model = pystan.StanModel(model_code=cycle_code)