-
Notifications
You must be signed in to change notification settings - Fork 1
/
baselines.py
208 lines (199 loc) · 6.88 KB
/
baselines.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
import numpy as np
import scipy.stats as stats
import pickle
import pandas as pd
from collections import OrderedDict
from statsmodels.distributions.empirical_distribution import ECDF
import matplotlib.pyplot as plt
def plot_scores(seq, score):
vt_event = seq['time_target']
lambda_x = seq['lambda_x']
t_x = seq['t_x']
scale = 0.25 * np.max(lambda_x)
vlabel = seq['label_test'].astype(np.int)
vt = seq['time_test']
colors = np.array(['black', 'red'])
plt.figure()
plt.plot(t_x, lambda_x)
plt.stem(vt_event, scale*np.ones_like(vt_event), 'k-', 'ko')
plt.scatter(vt[1:], score, c=colors[vlabel])
plt.show()
def detect_rand(test_set):
result = []
for seq_i, seq in enumerate(test_set):
vt = seq['time_test']
vlabel = seq['label_test']
score = np.random.rand(len(vt)-1)
result.append(pd.DataFrame(OrderedDict({
'seq': seq_i,
'time': vt[1:],
'score_omiss': score,
'score_commiss': -score,
'label': vlabel,
})))
result = pd.concat(result)
return result
def fit_len(train_set):
lens = []
for seq in train_set:
vt = seq['time_target']
t_min = seq['start']
if vt[0] != t_min:
vt = np.insert(vt, 0, t_min)
lens.extend(np.diff(vt))
return ECDF(lens)
def detect_len(test_set, ecdf):
result = []
for seq_i, seq in enumerate(test_set):
vt = seq['time_test']
vlabel = seq['label_test']
dt = np.diff(vt)
score_omiss = dt
p_left = ecdf(dt)
p_right = 1 - p_left
score_commiss = -np.minimum(p_left, p_right)
result.append(pd.DataFrame(OrderedDict({
'seq': seq_i,
'time': vt[1:],
'score_omiss': score_omiss,
'score_commiss': score_commiss,
'label': vlabel,
})))
result = pd.concat(result)
return result
def detect_model_true(test_set):
result = []
for seq_i, seq in enumerate(test_set):
vt = seq['time_test']
vlabel = seq['label_test']
t_x = seq['t_x']
lambda_x = seq['lambda_x']
lambda_ = np.interp(vt, t_x, lambda_x)
Lambda = []
for i in range(len(vt)-1):
idx = (t_x > vt[i]) & (t_x < vt[i+1])
y = np.concatenate(([lambda_[i]], lambda_x[idx], [lambda_[i+1]]))
x = np.concatenate(([vt[i]], t_x[idx], [vt[i+1]]))
Lambda.append(np.trapz(y, x))
Lambda = np.array(Lambda)
result.append(pd.DataFrame(OrderedDict({
'seq': seq_i,
'time': vt[1:],
'score_omiss': Lambda,
'score_commiss': -lambda_[1:],
'label': vlabel,
})))
result = pd.concat(result)
return result
def detect_model_pois(test_set, param):
result = []
for seq_i, seq in enumerate(test_set):
vt = seq['time_test']
vlabel = seq['label_test']
t_max = seq['stop']
vt_z = np.append(seq['time_context'], t_max)
vz = seq['mark_context']
n = len(vt)
score = np.zeros((n-1,2))
for i in range(n-1):
# find z's inbetween
t_beg = vt[i]
t_end = vt[i+1]
j_beg = np.nonzero(vt_z > t_beg)[0][0] - 1
j_end = np.nonzero(vt_z >= t_end)[0][0]
Lambda = 0
for j in range(j_beg, j_end, 1):
t_span = np.min([vt_z[j+1], t_end]) - np.max([vt_z[j], t_beg])
Lambda = Lambda + param[vz[j]] * t_span
lambda_ = param[vz[j_end-1]]
score[i, 0] = Lambda
score[i, 1] = -lambda_
result.append(pd.DataFrame(OrderedDict({
'seq': seq_i,
'time': vt[1:],
'score_omiss': score[:, 0],
'score_commiss': score[:, 1],
'label': vlabel,
})))
result = pd.concat(result)
return result
def fit_model_pois(train_set, n_mode):
data = [None] * n_mode
for seq in train_set:
vt_event = seq['time_target']
t_max = seq['stop']
vt_z = np.append(seq['time_context'], t_max)
vz = seq['mark_context']
for i in range(len(vz)):
vt = vt_event[(vt_event > vt_z[i]) & (vt_event <= vt_z[i+1])]
new_data = np.diff(vt)
if data[vz[i]] is None:
data[vz[i]] = new_data
else:
data[vz[i]] = np.append(data[vz[i]], new_data)
param = np.zeros(n_mode)
for k in range(n_mode):
param[k] = 1/data[k].mean()
return param
def detect_model_gam(test_set, param):
result = []
for seq_i, seq in enumerate(test_set):
vt_event = seq['time_target']
vt = seq['time_test']
vlabel = seq['label_test']
t_max = seq['stop']
vt_z = np.append(seq['time_context'], t_max)
vz = seq['mark_context']
n = len(vt)
Lambda = np.zeros(n-1)
score = np.zeros((n-1,2))
vt_ref = np.concatenate(([0], vt_event))
for i in range(n-1):
# find z's inbetween
t_beg = vt[i]
t_end = vt[i+1]
j_beg = np.nonzero(vt_z > t_beg)[0][0] - 1
j_end = np.nonzero(vt_z >= t_end)[0][0]
t_ref = vt_ref[vt_ref <= t_beg][-1]
Lambda[i] = 0
for j in range(j_beg, j_end, 1):
t_s_beg = np.max([vt_z[j], t_beg]) - t_ref
t_s_end = np.min([vt_z[j+1], t_end]) - t_ref
Lambda[i] = Lambda[i] \
- stats.gamma.logsf(t_s_end, param[vz[j],0], scale=1/param[vz[j],1]) \
+ stats.gamma.logsf(t_s_beg, param[vz[j],0], scale=1/param[vz[j],1])
a = param[vz[j_end-1],0]
b = 1/param[vz[j_end-1],1]
lambda_ = stats.gamma.pdf(t_end - t_beg, a, scale=b) / stats.gamma.sf(t_end - t_beg, a, scale=b)
score[i, 0] = Lambda[i]
score[i, 1] = -lambda_
assert(not np.any(np.isnan(score)))
result.append(pd.DataFrame(OrderedDict({
'seq': seq_i,
'time': vt[1:],
'score_omiss': score[:, 0],
'score_commiss': score[:, 1],
'label': vlabel,
})))
result = pd.concat(result)
return result
def fit_model_gam(train_set, n_mode):
data = [None] * n_mode
for seq in train_set:
vt_event = seq['time_target']
t_max = seq['stop']
vt_z = np.append(seq['time_context'], t_max)
vz = seq['mark_context']
for i in range(len(vz)):
vt = vt_event[(vt_event > vt_z[i]) & (vt_event <= vt_z[i+1])]
new_data = np.diff(vt)
if data[vz[i]] is None:
data[vz[i]] = new_data
else:
data[vz[i]] = np.append(data[vz[i]], new_data)
K = len(data)
param = np.zeros((K, 2))
for k in range(K):
param[k,0], _, param[k,1] = stats.gamma.fit(data[k], floc=0)
param[:,1] = 1/param[:,1]
return param