-
Notifications
You must be signed in to change notification settings - Fork 0
/
custom_loss_bi_noise.py
176 lines (154 loc) · 7.23 KB
/
custom_loss_bi_noise.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
import tensorflow as tf
from sklearn.metrics import r2_score
import glob
import pandas as pd
import random
import scipy.io as sio
import numpy as np
from tensorflow import keras
from tensorflow.keras import layers
from tensorflow.keras.layers import Bidirectional
from util.custom_loss import custom_loss
# configure GPU
physical_devices = tf.config.list_physical_devices('GPU')
if len(physical_devices) != 0:
tf.config.experimental.set_memory_growth(physical_devices[0], enable=True)
def get_sz(file):
mat_f = sio.loadmat(file)
dt = pd.DataFrame(mat_f['data'])
return dt
if __name__ == '__main__':
files_path = glob.glob('./simu_data/*.csv')
X_list1 = []
y_list1 = []
X_list = []
y_list = []
for file in files_path:
try:
df = pd.read_csv(file, index_col=0)
x = df.iloc[13:14, :].transpose()
y = df.iloc[:14, :].transpose()
except (IndexError, KeyError, pd.errors.ParserError):
print(file)
continue
else:
X_list1.append(x)
y_list1.append(y)
length = len(x)
random.Random(123).shuffle(X_list1)
random.Random(123).shuffle(y_list1)
features = 1
X_list1 = np.stack(X_list1)
y_list1 = np.stack(y_list1)
y_list2 = np.copy(y_list1)
mean_list = []
std_list = []
i = 0
while i < X_list1.shape[2]:
mean = X_list1[:,:,i].mean()
std = X_list1[:,:,i].std()
X_list1[:,:,i] = (X_list1[:,:,i] - mean) / std
i += 1
i = 0
while i < y_list1.shape[2]:
mean = y_list1[:,:,i].mean()
std = y_list1[:,:,i].std()
y_list2[:,:,i] = (y_list1[:,:,i] - mean) / std
mean_list.append(mean)
std_list.append(std)
i += 1
# train, test, val
X_train_list = X_list1[:int(len(X_list1)*.8)]
X_test_list = X_list1[int(len(X_list1)*.9):]
y_train_list = y_list2[:int(len(y_list2)*.8)]
y_test_list = y_list2[int(len(y_list2)*.9):]
X_val = X_list1[int(len(X_list1)*.8):int(len(X_list1)*.9),:,:]
y_val = y_list2[int(len(y_list2)*.8):int(len(y_list2)*.9),:,:]
# if splitting the recordings is needed, make n_split > 1
n_split = 1
cut1 = [np.split(x, n_split, axis=0) for x in X_train_list]
cut2 = [np.stack(x) for x in cut1]
X_train_list = np.concatenate(cut2)
cut1 = [np.split(x, n_split, axis=0) for x in y_train_list]
cut2 = [np.stack(x) for x in cut1]
y_train_list = np.concatenate(cut2)
cut1 = [np.split(x, n_split, axis=0) for x in X_val]
cut2 = [np.stack(x) for x in cut1]
X_val = np.concatenate(cut2)
cut1 = [np.split(x, n_split, axis=0) for x in y_val]
cut2 = [np.stack(x) for x in cut1]
y_val = np.concatenate(cut2)
T_after_cut = 400
batch_size = 1
targets = 14
nb_cuts = 1
# LSTM model
model = keras.Sequential()
if len(physical_devices) == 0:
model.add(Bidirectional(tf.keras.layers.LSTM(128, return_sequences=True,
stateful=False),
batch_input_shape=(batch_size, T_after_cut, features)))
model.add(Bidirectional(tf.keras.layers.LSTM(32, return_sequences=True,
stateful=False),
batch_input_shape=(batch_size, T_after_cut, 128)))
else:
model.add(Bidirectional(tf.compat.v1.keras.layers.CuDNNLSTM(128, return_sequences=True,
stateful=False),
batch_input_shape=(batch_size, T_after_cut, features)))
model.add(Bidirectional(tf.compat.v1.keras.layers.CuDNNLSTM(32, return_sequences=True,
stateful=False),
batch_input_shape=(batch_size, T_after_cut, 128)))
model.add(layers.TimeDistributed(layers.Dense(targets, activation='linear')))
optimizer = keras.optimizers.RMSprop(lr=0.001)
model.compile(loss=custom_loss, optimizer=optimizer, run_eagerly=True)
earlyStopping = tf.keras.callbacks.EarlyStopping(monitor='val_loss', min_delta=0.0005, patience=10, verbose=0,
mode='auto', restore_best_weights=True)
n_epochs = 10000
n_batch = 1
i = 0
model.fit(X_train_list, y_train_list,
epochs=n_epochs,
batch_size=n_batch,
verbose=0,
shuffle=False,
validation_data=(X_val, y_val),
callbacks=earlyStopping)
model.save_weights('saved_weights/bi_128_32_stateless')
y_val_list = []
for i in range(X_val.shape[0]):
y_val_pred = model.predict(X_val[i:i+1,:])
y_val_list.append(y_val_pred)
y_val_preds = np.concatenate(y_val_list,axis=0)
y_pred_list = []
for i in range(X_test_list.shape[0]):
y_pred = model.predict(X_test_list[i:i+1,:])
y_pred_list.append(y_pred)
y_preds = np.concatenate(y_pred_list,axis=0)
print('validation')
print('I-P Potential R squared',r2_score(y_val[:,:,0],y_val_preds[:,:,0]))
print('I-P Potential Derivative R squared',r2_score(y_val[:,:,1],y_val_preds[:,:,1]))
print('P-I Potential R squared',r2_score(y_val[:,:,2],y_val_preds[:,:,2]))
print('P-I Potential Derivative R squared',r2_score(y_val[:,:,3],y_val_preds[:,:,3]))
print('P-E Potential R squared',r2_score(y_val[:,:,4],y_val_preds[:,:,4]))
print('P-E Potential Derivative R squared',r2_score(y_val[:,:,5],y_val_preds[:,:,5]))
print('E-P Potential R squared',r2_score(y_val[:,:,6],y_val_preds[:,:,6]))
print('E-P Potential Derivative R squared',r2_score(y_val[:,:,7],y_val_preds[:,:,7]))
print('Input R squared',r2_score(y_val[:,:,8],y_val_preds[:,:,8]))
print('I-P Connectivity Strength R squared',r2_score(y_val[:,:,9],y_val_preds[:,:,9]))
print('P-I Connectivity Strength R squared',r2_score(y_val[:,:,10],y_val_preds[:,:,10]))
print('P-E Connectivity Strength R squared',r2_score(y_val[:,:,11],y_val_preds[:,:,11]))
print('E-P Connectivity Strength R squared',r2_score(y_val[:,:,12],y_val_preds[:,:,12]))
print('test')
print('I-P Potential R squared',r2_score(y_test_list[:,:,0],y_preds[:,:,0]))
print('I-P Potential Derivative R squared',r2_score(y_test_list[:,:,1],y_preds[:,:,1]))
print('P-I Potential R squared',r2_score(y_test_list[:,:,2],y_preds[:,:,2]))
print('P-I Potential Derivative R squared',r2_score(y_test_list[:,:,3],y_preds[:,:,3]))
print('P-E Potential R squared',r2_score(y_test_list[:,:,4],y_preds[:,:,4]))
print('P-E Potential Derivative R squared',r2_score(y_test_list[:,:,5],y_preds[:,:,5]))
print('E-P Potential R squared',r2_score(y_test_list[:,:,6],y_preds[:,:,6]))
print('E-P Potential Derivative R squared',r2_score(y_test_list[:,:,7],y_preds[:,:,7]))
print('Input R squared',r2_score(y_test_list[:,:,8],y_preds[:,:,8]))
print('I-P Connectivity Strength R squared',r2_score(y_test_list[:,:,9],y_preds[:,:,9]))
print('P-I Connectivity Strength R squared',r2_score(y_test_list[:,:,10],y_preds[:,:,10]))
print('P-E Connectivity Strength R squared',r2_score(y_test_list[:,:,11],y_preds[:,:,11]))
print('E-P Connectivity Strength R squared',r2_score(y_test_list[:,:,12],y_preds[:,:,12]))