/
BO_BCI.py
326 lines (243 loc) · 16.1 KB
/
BO_BCI.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
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
import spearmint_lite
import os
import sys
os.chdir('../bci_framework')
sys.path.append('./BCI_Framework')
import Main
import Configuration_BCI
import Single_Job_runner as SJR
import numpy as np
import itertools
import time
from time import sleep
import sklearn
def generate_window_list_freq(window_start_list, window_length_list, limit):
window_list = []
for window_start_point in window_start_list:
for window_length in window_length_list:
if window_start_point + window_length <= limit:
window_list.append([window_start_point, window_start_point + window_length])
return window_list
def generate_window_list_time(window_start_list, window_length_list, limit):
window_list = []
for window_start_point in window_start_list:
for window_length in window_length_list:
if window_start_point + window_length <= limit:
window_list.append([window_start_point, limit - (window_start_point + window_length)])
return window_list
def generate_channel_indices(n_channels):
indices_list = []
for i in range(1,5):#9 is ok for ubc-pc
indices_list = indices_list + list(itertools.combinations(range(n_channels),i))
return indices_list
def revise_candidates(raw_candidates, BO_type):
#if elemetn >1 and type 5 binary
candidates = []
for raw_candidate in raw_candidates:
candidate = []
for element_ind, element in enumerate(raw_candidate):
if element_ind >= 1 and BO_type == 5:
cand = np.zeros(22)
cand[list(element)] = 1
candidate = candidate + [int(''.join(map(str,map(int,cand))),2)]
else:
for el in element:
candidate.append(el)
candidates.append(candidate)
return candidates
def read_initial_candidates(file_name, config):
""" """
with open(file_name, 'r') as init_file:
all_initial_cands = init_file.readlines()
for line_ind, line in enumerate(all_initial_cands):
all_initial_cands[line_ind] = line.strip().split()
all_initial_cands[line_ind] = map(lambda X: float(X) if X.replace('.','').replace('-','').isdigit() else X, all_initial_cands[line_ind])
# all_initial_cands = np.loadtxt(file_name)
# all_initial_cands = np.ndarray.tolist(all_initial_cands)
###################################3
for init_cand_index in range(len(all_initial_cands)):
all_initial_cands[init_cand_index][0] *= config.configuration['sampling_rate']
all_initial_cands[init_cand_index][1] *= config.configuration['sampling_rate']
if all_initial_cands[init_cand_index][1] < 0:
all_initial_cands[init_cand_index][1] = -1 * all_initial_cands[init_cand_index][1]
######################################
return all_initial_cands
def read_initial_candidates_type5(file_name, config):
""" """
all_initial_cands = np.loadtxt(file_name)
new_initial_cands = np.zeros(shape = (len(all_initial_cands), 3))
all_initial_cands = np.ndarray.tolist(all_initial_cands)
new_initial_cands = np.ndarray.tolist(new_initial_cands)
###################################3
for init_cand_index in range(len(all_initial_cands)):
new_initial_cands[init_cand_index][0] = all_initial_cands[init_cand_index][0] * config.configuration['sampling_rate']
new_initial_cands[init_cand_index][1] = all_initial_cands[init_cand_index][1] * config.configuration['sampling_rate']
if all_initial_cands[init_cand_index][1] < 0:
new_initial_cands[init_cand_index][1] = -1 * all_initial_cands[init_cand_index][1]
new_initial_cands[init_cand_index][2] = int(''.join(map(str, map(int,all_initial_cands[init_cand_index][2:]))), 2)
######################################
return new_initial_cands
def generate_all_candidates(dataset_name, optimization_type):
##########################################################################################################################################################
# the following code block generates potential candidates for Bayesian Optimizer
# different BO types
# type 1 -> only search for time window
# type 2 -> search for time and frequency window
# type 3 -> search for time window and channels
# type 4 -> search for time window and search for frequency window and channels
# type 3-1 -> search for time window and search for frequency window for each channel separately
# type 5 -> search for time window and search for frequency window for each channel separately
BO_type = optimization_type
config = Configuration_BCI.Configuration_BCI('BCI_Framework', dataset_name)
sampling_rate = config.configuration['sampling_rate']
all_subjects_candidates_list = []
for subj_ind, subj in enumerate(config.configuration['subject_names_str']):
mv_size = config.configuration['movement_trial_size_list'][subj_ind]
all_mvs_have_same_lebngth = all(map(lambda x: x == mv_size, config.configuration['movement_trial_size_list']))
n_channels = config.configuration['number_of_channels']
candidates_list = []
if BO_type == 1:
window_start_list = map(int, np.arange(0, mv_size - sampling_rate, sampling_rate/4.0))
window_length_list = map(int, np.arange(0.75 * sampling_rate, mv_size+1, sampling_rate/4.0))
window_list = generate_window_list_time(window_start_list, window_length_list, mv_size)
candidates_list = window_list
# all_initial_cands = read_initial_candidates('BCI_Framework/BO_type1_initial_candidates.txt', config)
elif BO_type == 2:
window_start_list = map(int, np.arange(0, mv_size - sampling_rate, sampling_rate/4.0))
window_length_list = map(int, np.arange(0.75 * sampling_rate, mv_size+1, sampling_rate/4.0))
window_list_mv = generate_window_list_time(window_start_list, window_length_list, mv_size)
candidates_list.append(window_list_mv)
# window_start_list_freq = np.arange(2, 6, 0.5)
# window_length_list_freq = np.arange(26, 30, 0.75)
# window_list_freq = generate_window_list_freq(window_start_list_freq, window_length_list_freq, 32)
# window_list_freq = zip(window_list_freq, window_list_freq)
# window_list_freq = revise_candidates(window_list_freq, 0)
window_list_freq = []
alpha_beta_candidates = []
window_start_list_freq = np.arange(7, 9.5, 0.5)
window_length_list_freq = np.arange(4, 6, 0.75)
window_list_freq_alpha = generate_window_list_freq(window_start_list_freq, window_length_list_freq, 14)
alpha_beta_candidates.append(window_list_freq_alpha)
window_start_list_freq = np.arange(15, 19, 0.5)
window_length_list_freq = np.arange(5, 10, 0.75)
window_list_freq_beta = generate_window_list_freq(window_start_list_freq, window_length_list_freq, 26)
alpha_beta_candidates.append(window_list_freq_beta)
raw_candidates_alpha_beta = list(itertools.product(*alpha_beta_candidates))
candidates_list_alpha_beta = revise_candidates(raw_candidates_alpha_beta, BO_type)
window_list_freq = window_list_freq + candidates_list_alpha_beta
candidates_list.append(window_list_freq)
# candidates_list.append(candidates_list_alpha_beta)
# all_initial_cands = read_initial_candidates('BCI_Framework/BO_type2_initial_candidates.txt', config)
elif BO_type == 3:
window_start_list = map(int, np.arange(0, mv_size - sampling_rate, sampling_rate/4.0))
window_length_list = map(int, np.arange(0.75 * sampling_rate, mv_size+1, sampling_rate/4.0))
window_list_mv = generate_window_list_time(window_start_list, window_length_list, mv_size)
candidates_list.append(window_list_mv)
# channels_list = [['ALL'], ['CSP2'], ['CSP4'], ['CSP6'], ['CS']]
channels_list = [[0], [1], [2], [3], [4]]#???????????????????????????????????????????????????????????????????????????????????????????
candidates_list.append(channels_list)
# all_initial_cands = read_initial_candidates('BCI_Framework/BO_type3_initial_candidates.txt', config)
elif BO_type == 4:
window_start_list = map(int, np.arange(100, mv_size - sampling_rate, sampling_rate/4.0))##
window_length_list = map(int, np.arange(0.75 * sampling_rate, mv_size+1, sampling_rate/4.0))
window_list_mv = generate_window_list_time(window_start_list, window_length_list, mv_size)
candidates_list.append(window_list_mv)
window_start_list_freq = np.arange(2, 6, 0.5)
window_length_list_freq = np.arange(26, 30, 0.75)
window_list_freq = generate_window_list_freq(window_start_list_freq, window_length_list_freq, 32)
window_list_freq = zip(window_list_freq, window_list_freq)
window_list_freq = revise_candidates(window_list_freq, 0)
alpha_beta_candidates = []
window_start_list_freq = np.arange(7, 9.5, 0.5)
window_length_list_freq = np.arange(4, 6, 0.75)
window_list_freq_alpha = generate_window_list_freq(window_start_list_freq, window_length_list_freq, 14)
alpha_beta_candidates.append(window_list_freq_alpha)
window_start_list_freq = np.arange(15, 19, 0.5)
window_length_list_freq = np.arange(5, 10, 0.75)
window_list_freq_beta = generate_window_list_freq(window_start_list_freq, window_length_list_freq, 26)
alpha_beta_candidates.append(window_list_freq_beta)
raw_candidates_alpha_beta = list(itertools.product(*alpha_beta_candidates))
candidates_list_alpha_beta = revise_candidates(raw_candidates_alpha_beta, BO_type)
window_list_freq = window_list_freq + candidates_list_alpha_beta
candidates_list.append(window_list_freq)
channels_list = [[0], [1], [2], [3], [4]]#???????????????????????????????????????????????????????????????????????????????????????????
candidates_list.append(channels_list)
# all_initial_cands = read_initial_candidates('BCI_Framework/BO_type4_initial_candidates.txt', config)
# elif BO_type == 5:
#
# window_start_list = map(int, np.arange(0, mv_size - sampling_rate, sampling_rate/4.0))
# window_length_list = map(int, np.arange(0.75 * sampling_rate, mv_size+1, sampling_rate/4.0))
# window_list_mv = generate_window_list_time(window_start_list, window_length_list, mv_size)
# candidates_list.append(window_list_mv)
#
# indices_list = generate_channel_indices(n_channels)
# candidates_list.append(indices_list)
# all_initial_cands = read_initial_candidates('BCI_Framework/BO_type4_initial_candidates.txt', config)
if BO_type != 1:
raw_candidates = list(itertools.product(*candidates_list))
candidates_list = revise_candidates(raw_candidates, BO_type)
# candidates_list = all_initial_cands + candidates_list ################################################
# Job_Params.n_initial_candidates = len(all_initial_cands) ############################################
all_subjects_candidates_list.append(candidates_list)
if all_mvs_have_same_lebngth:
break
##########################################################################################################################################################
return all_subjects_candidates_list, 0
if __name__ == '__main__':
print "Bayesian Optimization for BCI"
# start_time = time.time()
datasets = ['BCICIII3b']#['BCICIII3b','BCICIV2b', 'BCICIV2a']
classifier = 'LogisticRegression'
feature = 'BP' #'morlet'] #morlet for type 2 and 4 does not work!!!!!
optimization_types_dict = {('BCICIII3b','BP'):[2], ('BCICIII3b','morlet'):[1], ('BCICIV2b','BP') : [2], ('BCICIV2b','morlet') : [1], ('BCICIV2a','BP') : [4], ('BCICIV2a','morlet') : [3]}
# BO_selection_types = ["GPEIOptChooser1"]
BO_selection_types = ["GPEIOptChooser1", "RandomForestEIChooser1", "RandomChooser1",
"GPEIOptChooser2", "RandomForestEIChooser2", "RandomChooser2",
"GPEIOptChooser3", "RandomForestEIChooser3", "RandomChooser3",
"GPEIOptChooser4", "RandomForestEIChooser4", "RandomChooser4",
"GPEIOptChooser5", "RandomForestEIChooser5", "RandomChooser5",
"GPEIOptChooser6", "RandomForestEIChooser6", "RandomChooser6",
"GPEIOptChooser7", "RandomForestEIChooser7", "RandomChooser7",
"GPEIOptChooser8", "RandomForestEIChooser8", "RandomChooser8",
"GPEIOptChooser9", "RandomForestEIChooser9", "RandomChooser9",
"GPEIOptChooser10", "RandomForestEIChooser10", "RandomChooser10"]
all_subjects_candidates_dict = {}
for dataset_ind, dataset in enumerate(datasets):
# for bo_type in BO_selection_types:
for optimization_type in optimization_types_dict[(dataset, feature)]:
print dataset, feature, optimization_type
all_subjects_candidates_dict[(dataset_ind, optimization_type)], _ = generate_all_candidates(dataset, optimization_type)
first_iteration = True
finished = {}
while True:
if first_iteration == False and all(finished.values()):
break
for dataset_ind, dataset in enumerate(datasets):
for bo_type_ind, bo_type in enumerate(BO_selection_types):
for optimization_type in optimization_types_dict[(dataset, feature)]:
print dataset, feature, bo_type, optimization_type
class Job_Params:
job_dir = '../Candidates'
num_all_jobs = 60
dataset = dataset
from random import randrange
seed = randrange(50)
classifier_name = classifier
feature_extraction = feature
n_concurrent_jobs = 1
chooser_module = bo_type
Job_Params.n_initial_candidates = 0
config = Configuration_BCI.Configuration_BCI('BCI_Framework', dataset)
complete_jobs = np.zeros(config.configuration['number_of_subjects'])
all_mvs_have_same_lebngth = all(map(lambda x: x == config.configuration['movement_trial_size_list'][0], config.configuration['movement_trial_size_list']))
for subj_ind, subj in enumerate(config.configuration['subject_names_str']):
if all_mvs_have_same_lebngth:
all_subjects_candidates_list = all_subjects_candidates_dict[(dataset_ind, optimization_type)][0]
else:
all_subjects_candidates_list = all_subjects_candidates_dict[(dataset_ind, optimization_type)][subj_ind]
sp = spearmint_lite.spearmint_lite(Job_Params, all_subjects_candidates_list, config, optimization_type)
finished[(subj+str(dataset_ind), optimization_type)] = sp.main(Job_Params, complete_jobs, subj)
# sleep(1)
first_iteration = False
# execution_time = time.time() - start_time
# print execution_time