-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathutils.py
207 lines (172 loc) · 10 KB
/
utils.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
import numpy as np
from typing import Sequence, Tuple
from scipy.signal import resample
from tensorflow.random import set_seed
def load_GIGA_MI_ME(db,
sbj: int,
eeg_ch_names: Sequence[str],
fs: float,
f_bank: np.ndarray,
vwt: np.ndarray,
new_fs: float,
tf_repr) -> Tuple[np.ndarray, np.ndarray]:
index_eeg_chs = db.format_channels_selectors(channels = eeg_ch_names) - 1
# tf_repr = TimeFrequencyRpr(sfreq = fs, f_bank = f_bank, vwt = vwt)
db.load_subject(sbj)
X, y = db.get_data(classes = ['left hand mi', 'right hand mi']) #Load MI classes, all channels {EEG}, reject bad trials, uV
X = X[:, index_eeg_chs, :] #spatial rearrangement
X = np.squeeze(tf_repr.transform(X))
#Resampling
if new_fs == fs:
print('No resampling, since new sampling rate same.')
else:
print("Resampling from {:f} to {:f} Hz.".format(fs, new_fs))
X = resample(X, int((X.shape[-1]/fs)*new_fs), axis = -1)
return X, y
def load_BCICIV2a(db,
sbj: int,
mode: str,
fs: float,
f_bank: np.ndarray,
vwt: np.ndarray,
new_fs: float) -> np.ndarray:
tf_repr = TimeFrequencyRpr(sfreq = fs, f_bank = f_bank, vwt = vwt)
db.load_subject(sbj, mode = mode)
X, y = db.get_data() #Load all classes, all channels {EEG, EOG}, reject bad trials
X = X[:,:-3,:] # pick EEG channels
X = X*1e6 #uV
X = np.squeeze(tf_repr.transform(X))
#Resampling
if new_fs == fs:
print('No resampling, since new sampling rate same.')
else:
print("Resampling from {:f} to {:f} Hz.".format(fs, new_fs))
X = resample(X, int((X.shape[-1]/fs)*new_fs), axis = -1)
return X, y
def load_DB(db_name, **load_args):
if db_name == 'BCICIV2a':
X_train, y_train = load_BCICIV2a(**load_args, mode = 'training')
X_test, y_test = load_BCICIV2a(**load_args, mode = 'evaluation')
X_train = np.concatenate([X_train, X_test], axis = 0)
y_train = np.concatenate([y_train, y_test], axis = 0)
elif db_name == 'GIGA_MI_ME':
X_train, y_train = load_GIGA_MI_ME(**load_args)
else:
raise ValueError('No valid database name')
return X_train, y_train
#######################
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.backend import clear_session
from sklearn.metrics import (
accuracy_score,
cohen_kappa_score,
roc_auc_score,
f1_score,
recall_score,
precision_score
)
def train(model, db_name, tf_repr, load_args, cv_args, model_args, compile_args, fit_args, seed):
X_train, y_train = load_GIGA_MI_ME(**load_args)
X_train = X_train[..., np.newaxis]
# print(X_train.shape, y_train.shape)
cv_results = {'params': [],
'mean_acc': np.zeros(cv_args['cv'].get_n_splits()),
'mean_kappa': np.zeros(cv_args['cv'].get_n_splits()),
'mean_auc': np.zeros(cv_args['cv'].get_n_splits()),
'mean_f1_left': np.zeros(cv_args['cv'].get_n_splits()),
'mean_f1_right': np.zeros(cv_args['cv'].get_n_splits()),
'mean_recall_left': np.zeros(cv_args['cv'].get_n_splits()),
'mean_recall_right': np.zeros(cv_args['cv'].get_n_splits()),
'mean_precision_left': np.zeros(cv_args['cv'].get_n_splits()),
'mean_precision_right': np.zeros(cv_args['cv'].get_n_splits()),}
if model_args['nb_classes'] == 4:
cv_results['mean_f1_legs'] = np.zeros(cv_args['cv'].get_n_splits())
cv_results['mean_f1_tongue'] = np.zeros(cv_args['cv'].get_n_splits())
cv_results['mean_recall_legs'] = np.zeros(cv_args['cv'].get_n_splits())
cv_results['mean_recall_tongue'] = np.zeros(cv_args['cv'].get_n_splits())
cv_results['mean_precision_legs'] = np.zeros(cv_args['cv'].get_n_splits())
cv_results['mean_precision_tongue'] = np.zeros(cv_args['cv'].get_n_splits())
k = 0
max_acc = -np.inf
for train_index, val_index in cv_args['cv'].split(X_train, y_train):
X, X_val = X_train[train_index], X_train[val_index]
y, y_val = y_train[train_index], y_train[val_index]
# print(val_index)
if model_args['autoencoder']:
y = [X, y]
batch_size, C, T = X.shape[:-1]
clear_session()
set_seed(seed)
# model_cll, model_params = get_model(model_args['model_name'], model_args['nb_classes'])
# model = model_cll(**model_params, Chans = C, Samples = T)
model.compile(loss = compile_args['loss'],
optimizer = Adam(compile_args['init_lr']))
history = model.fit(X, y,
batch_size = batch_size,
**fit_args)
if model_args['autoencoder']:
y_prob = model.predict(X_val)[-1]
y_pred = np.argmax(y_prob, axis = 1)
else:
y_prob = model.predict(X_val)
y_pred = np.argmax(y_prob, axis = 1)
cv_results['mean_acc'][k] = accuracy_score(y_val, y_pred)
cv_results['mean_kappa'][k] = cohen_kappa_score(y_val, y_pred)
if model_args['nb_classes'] == 2:
cv_results['mean_auc'][k] = roc_auc_score(y_val, y_prob[:, 1], average = 'macro')
cv_results['mean_f1_left'][k] = f1_score(y_val, y_pred, pos_label = 0, average = 'binary')
cv_results['mean_f1_right'][k] = f1_score(y_val, y_pred, pos_label = 1, average = 'binary')
cv_results['mean_recall_left'][k] = recall_score(y_val, y_pred, pos_label = 0, average = 'binary')
cv_results['mean_recall_right'][k] = recall_score(y_val, y_pred, pos_label = 1, average = 'binary')
cv_results['mean_precision_left'][k] = precision_score(y_val, y_pred, pos_label = 0, average = 'binary')
cv_results['mean_precision_right'][k] = precision_score(y_val, y_pred, pos_label = 1, average = 'binary')
else:
cv_results['mean_auc'][k] = roc_auc_score(y_val, y_prob, average = 'macro', multi_class = 'ovo')
cv_results['mean_f1_left'][k] = f1_score(y_val, y_pred, pos_label = 0, average = 'micro')
cv_results['mean_f1_right'][k] = f1_score(y_val, y_pred, pos_label = 1, average = 'micro')
cv_results['mean_f1_legs'][k] = f1_score(y_val, y_pred, pos_label = 2, average = 'micro')
cv_results['mean_f1_tongue'][k] = f1_score(y_val, y_pred, pos_label = 3, average = 'micro')
cv_results['mean_recall_left'][k] = recall_score(y_val, y_pred, pos_label = 0, average = 'micro')
cv_results['mean_recall_right'][k] = recall_score(y_val, y_pred, pos_label = 1, average = 'micro')
cv_results['mean_recall_legs'][k] = recall_score(y_val, y_pred, pos_label = 2, average = 'micro')
cv_results['mean_recall_tongue'][k] = recall_score(y_val, y_pred, pos_label = 3, average = 'micro')
cv_results['mean_precision_left'][k] = precision_score(y_val, y_pred, pos_label = 0, average = 'micro')
cv_results['mean_precision_right'][k] = precision_score(y_val, y_pred, pos_label = 1, average = 'micro')
cv_results['mean_precision_legs'][k] = precision_score(y_val, y_pred, pos_label = 2, average = 'micro')
cv_results['mean_precision_tongue'][k] = precision_score(y_val, y_pred, pos_label = 3, average = 'micro')
if cv_results['mean_acc'][k] > max_acc:
max_acc = cv_results['mean_acc'][k]
# model.save_weights('sbj' + str(load_args['sbj']) +'.h5')
k += 1
cv_results['std_acc'] = round(cv_results['mean_acc'].std(), 3)
cv_results['mean_acc'] = round(cv_results['mean_acc'].mean(), 3)
cv_results['std_kappa'] = round(cv_results['mean_kappa'].std(), 3)
cv_results['mean_kappa'] = round(cv_results['mean_kappa'].mean(), 3)
cv_results['std_auc'] = round(cv_results['mean_auc'].std(), 3)
cv_results['mean_auc'] = round(cv_results['mean_auc'].mean(), 3)
cv_results['mean_f1_left'] = round(cv_results['mean_f1_left'].mean(), 3)
cv_results['std_f1_left'] = round(cv_results['mean_f1_left'].std(), 3)
cv_results['mean_f1_right'] = round(cv_results['mean_f1_right'].mean(), 3)
cv_results['std_f1_right'] = round(cv_results['mean_f1_right'].std(), 3)
cv_results['mean_recall_left'] = round(cv_results['mean_recall_left'].mean(), 3)
cv_results['std_recall_left'] = round(cv_results['mean_recall_left'].std(), 3)
cv_results['mean_recall_right'] = round(cv_results['mean_recall_right'].mean(), 3)
cv_results['std_recall_right'] = round(cv_results['mean_recall_right'].std(), 3)
cv_results['mean_precision_left'] = round(cv_results['mean_precision_left'].mean(), 3)
cv_results['std_precision_left'] = round(cv_results['mean_precision_left'].std(), 3)
cv_results['mean_precision_right'] = round(cv_results['mean_precision_right'].mean(), 3)
cv_results['std_precision_right'] = round(cv_results['mean_precision_right'].std(), 3)
if model_args['nb_classes'] == 4:
cv_results['mean_f1_legs'] = round(cv_results['mean_f1_legs'].mean(), 3)
cv_results['std_f1_legs'] = round(cv_results['mean_f1_legs'].std(), 3)
cv_results['mean_f1_tongue'] = round(cv_results['mean_f1_tongue'].mean(), 3)
cv_results['std_f1_tongue'] = round(cv_results['mean_f1_tongue'].std(), 3)
cv_results['mean_recall_legs'] = round(cv_results['mean_recall_legs'].mean(), 3)
cv_results['std_recall_legs'] = round(cv_results['mean_recall_legs'].std(), 3)
cv_results['mean_recall_tongue'] = round(cv_results['mean_recall_tongue'].mean(), 3)
cv_results['std_recall_tongue'] = round(cv_results['mean_recall_tongue'].std(), 3)
cv_results['mean_precision_legs'] = round(cv_results['mean_precision_legs'].mean(), 3)
cv_results['std_precision_legs'] = round(cv_results['mean_precision_legs'].std(), 3)
cv_results['mean_precision_tongue'] = round(cv_results['mean_precision_tongue'].mean(), 3)
cv_results['std_precision_tongue'] = round(cv_results['mean_precision_tongue'].std(), 3)
return cv_results