-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathbatch_iterators.py
251 lines (211 loc) · 6.81 KB
/
batch_iterators.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
from nolearn.lasagne import BatchIterator
import numpy as np
import math
import scipy
import global_vars as g
import random
def normalize(x):
return x/2000*255
def _sldict(arr, sl):
if isinstance(arr, dict):
return {k: v[sl] for k, v in arr.items()}
else:
return arr[sl]
class BI_train_bal_complete(BatchIterator):
no_trials = 20
def __init__(self, batch_size, seed=42):
self.batch_size = batch_size
self.random = np.random.RandomState(seed)
preprocess_params = g.cfg['preprocess']
self.m_window = preprocess_params['magnitude_window']
self.ceil = preprocess_params['ceil']
self.floor = preprocess_params['floor']
self.shuffle = False
def __call__(self, X, y=None):
self.y = y
#print "BI_train_bal_complete called with len(X) =",len(X)
self.X = np.arange(len(X))
return self
def transform(self, Xb, yb):
#print 'transforming batch'
# Select and normalize:
bs = Xb.shape[0]
Xb_new = []
yb_new = []
for idx in range(bs):
hour = Xb[idx]
slice = None
if yb[idx]:
hour = hour % (g.ms_seizure_train.shape[0]+g.ms_xtra_seizure_train.shape[0]) #print 'seizure', hour
if hour >= g.ms_seizure_train.shape[0]:
#select extra samples
hour = hour - g.ms_seizure_train.shape[0]
r = range(g.ms_xtra_seizure_train.shape[2]-self.m_window)
start = np.random.choice(r)
slice = g.ms_xtra_seizure_train[hour, :, start:start+self.m_window]
Xb_new.append(slice)
yb_new.append(yb[idx])
else:
r = xrange(self.m_window)
start = np.random.choice(r)
for pos in range(start,g.ms_seizure_train.shape[2]-self.m_window+1,self.m_window):
slice = g.ms_seizure_train[hour, :, pos:pos+self.m_window]
Xb_new.append(slice)
yb_new.append(yb[idx])
else:
hour = hour % g.ms_normal_train.shape[0]
#print 'normal', hour
# if not g.blacklist_normal_train[hour]:
r = xrange(self.m_window)
start = np.random.choice(r)
for pos in range(start,g.ms_normal_train.shape[2]-self.m_window+1,self.m_window):
slice = g.ms_normal_train[hour, :, pos:pos+self.m_window]
Xb_new.append(slice)
yb_new.append(yb[idx])
# else:
# print 'blacklist_normal_train blocked', hour
Xb_new = np.stack(Xb_new)
Xb_new = normalize(Xb_new)
Xb_new = Xb_new.astype(np.float32)
return Xb_new, np.stack(yb_new)
class BI_test_bal_complete(BatchIterator):
no_trials = 20
def __init__(self, batch_size, seed=42):
self.batch_size = batch_size
self.random = np.random.RandomState(seed)
preprocess_params = g.cfg['preprocess']
self.m_window = preprocess_params['magnitude_window']
self.ceil = preprocess_params['ceil']
self.floor = preprocess_params['floor']
self.shuffle = False
def __call__(self, X, y=None):
self.y = y
#print "BI_test_bal_complete called with len(X) =",len(X)
self.X = np.arange(len(X))
self.X_orig = X
return self
def __iter__(self):
bs = self.batch_size
for i in range((self.n_samples + bs - 1) // bs):
sl = slice(i * bs, (i + 1) * bs)
Xb = _sldict(self.X, sl)
Xb_orig = _sldict(self.X_orig, sl)
if self.y is not None:
yb = _sldict(self.y, sl)
else:
yb = None
yield self.transform(Xb, yb, Xb_orig)
def transform(self, Xb, yb, Xb_orig):
# Select and normalize:
bs = Xb.shape[0]
Xb_new = []
yb_new = []
for idx in range(bs):
label = None
hour = None
if yb != None:
label = yb[idx]
hour = Xb[idx]
else:
label = Xb_orig[idx,1]
hour = Xb[idx]
if label:
hour = hour % g.ms_seizure_val.shape[0]
# if not g.blacklist_seizure_val[hour]:
r = xrange(self.m_window)
start = np.random.choice(r)
for pos in range(start,g.ms_seizure_val.shape[2]-self.m_window+1,self.m_window):
slice = g.ms_seizure_val[hour, :, pos:pos+self.m_window]
Xb_new.append(slice)
yb_new.append(yb[idx])
# else:
# print 'blacklist_seizure_val blocked', hour
else:
hour = hour % g.ms_normal_val.shape[0]
# if not g.blacklist_normal_val[hour]:
r = xrange(self.m_window)
start = np.random.choice(r)
for pos in range(start,g.ms_normal_val.shape[2]-self.m_window+1,self.m_window):
slice = g.ms_normal_val[hour, :, pos:pos+self.m_window]
Xb_new.append(slice)
yb_new.append(yb[idx])
# else:
# print 'blacklist_normal_val blocked', hour
Xb_new = normalize(np.stack(Xb_new))
Xb_new = Xb_new.astype(np.float32)
return Xb_new, np.stack(yb_new)
class BI_new(BatchIterator):
def __init__(self, batch_size, seed=42):
self.batch_size = batch_size
self.random = np.random.RandomState(seed)
preprocess_params = g.cfg['preprocess']
self.m_window = preprocess_params['magnitude_window']
self.ceil = preprocess_params['ceil']
self.floor = preprocess_params['floor']
self.shuffle = False
def __call__(self, X, y=None):
if y == None:
self.y = y
else:
self.y = y[0]
self.X = X
return self
def __iter__(self):
bs = self.batch_size
n_normal = len(self.X['normal'])
n_seizure = len(self.X['seizure'])
n_xtra_seizure = len(self.X['xtra_seizure'])
n_max = max([n_normal, n_seizure, n_xtra_seizure])
if bs > n_max:
print 'Warning: batch size is larger than the no. normal, seizure and xtra_seizure files.\nSo it does not make sense to increase batch size further'
for i in range((n_max + bs - 1) // bs):
Xb = np.empty((3,bs),dtype=np.int32)
yb = np.empty((3,bs),dtype=np.int32)
for j in range(bs):
idx = i*bs+j
if idx >= n_normal:
Xb[0,j] = np.random.choice(self.X['normal'])
else:
Xb[0,j] = self.X['normal'][idx]
yb[0,j] = 0
if idx >= n_seizure:
Xb[1,j] = np.random.choice(self.X['seizure'])
else:
Xb[1,j] = self.X['seizure'][idx]
yb[1,j] = 1
if idx >= n_xtra_seizure:
Xb[2,j] = np.random.choice(self.X['xtra_seizure'])
else:
Xb[2,j] = self.X['xtra_seizure'][idx]
yb[2,j] = 1
yield self.transform(Xb, yb)
def transform(self, Xb, yb):
bs = Xb.shape[0]
Xb_new = []
yb_new = []
for idx in range(bs):
# Normal samples
r = xrange(self.m_window)
start = np.random.choice(r)
for pos in range(start,g.ms_normal.shape[2]-self.m_window+1,self.m_window):
slice = g.ms_normal[Xb[0,idx], :, pos:pos+self.m_window]
Xb_new.append(slice)
yb_new.append(0)
# Seizure samples
r = xrange(self.m_window)
start = np.random.choice(r)
for pos in range(start,g.ms_seizure.shape[2]-self.m_window+1,self.m_window):
slice = g.ms_seizure[Xb[1,idx], :, pos:pos+self.m_window]
Xb_new.append(slice)
yb_new.append(1)
# Extra seizure samples
r = xrange(g.ms_xtra_seizure.shape[2]-self.m_window)
start = np.random.choice(r)
slice = g.ms_xtra_seizure[Xb[2,idx], :, start:start+self.m_window]
Xb_new.append(slice)
yb_new.append(1)
Xb_new = np.stack(Xb_new)
yb_new = np.stack(yb_new)
Xb_new = Xb_new.astype(np.float32)
yb_new = yb_new.astype(np.int32)
return Xb_new, yb_new