-
Notifications
You must be signed in to change notification settings - Fork 6
/
qttention.py
331 lines (282 loc) · 13.3 KB
/
qttention.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
327
328
329
330
331
import os
import pandas as pd
import scipy
import torch
import torch.nn as nn
import torch.nn.functional as F
from Model.QCNN import QCNN
from Model.WDCNN import WDCNN
from utils.DatasetLoader import CustomTensorDataset
from utils.Preprocess import prepro
from train import random_seed
features_in_hook = []
features_out_hook = []
from matplotlib import pyplot as plt
from scipy.io import loadmat
import numpy as np
import os
from sklearn import preprocessing # 0-1编码
from sklearn.model_selection import StratifiedShuffleSplit # 随机划分，保证每一类比例相同
'''
Calculating Qttention for the QCNN network
'''
def hook(module, input, output):
features_in_hook.append(input)
features_out_hook.append(output)
def cal_mm(x, w):
c = []
for i in range(x.shape[0]):
temp_x = x[i, :, :]
temp_w = w[i, :, :]
c.append(torch.mm(temp_x, temp_w.transpose(1, 0)))
return torch.tensor(c)
def attention_compose(map_list, output_shape=2048, stride=1, paddings=0, kernal_size=64):
map = torch.zeros(map_list[0].shape[0], output_shape + 2 * paddings)
for idx, m in enumerate(map_list):
map[:, idx * stride: idx * stride + kernal_size] += m.squeeze()
if idx > 0:
map[:, idx * stride: (idx - 1) * stride + kernal_size] /= 2
return map[:, paddings:output_shape + paddings]
def attention_map(x, wr, wg, wb, br, bg, stride=1, paddings=0, kernal_size=64):
output_shape = x.shape[2]
if paddings != 0:
padding = torch.zeros(1, x.shape[1], paddings)
x = torch.cat((padding, x, padding), 2)
n_sample = int((x.shape[2] - kernal_size) / stride + 1)
map_list = []
for i in range(n_sample):
temp_x = x[:, :, i * stride:i * stride + kernal_size]
temp_x_c = torch.repeat_interleave(temp_x, wb.shape[0], 0)
y1 = temp_x_c * wb
c = cal_mm(temp_x_c, wr)
c = torch.repeat_interleave(c, wg.shape[2]).reshape(wg.shape)
y2 = c * wg
map_list.append(y1 + y2)
map = attention_compose(map_list, output_shape, stride, paddings, kernal_size)
return map
if __name__ == '__main__':
random_seed(42)
model_name = 'qcnn'
model = QCNN()
run_path = 'wandb/qcnnhit6/checkpoint.pth' # Need to be specified as the path to the model file
best_model_dict = torch.load(run_path, map_location=torch.device('cpu'))
model.load_state_dict(best_model_dict)
model.eval()
chosen_data = '0HP'
SNR = 6
length = 2048
path = os.path.join('data', chosen_data)
test_X, test_Y = prepro(d_path=path,
length=length,
number=100,
normal=False,
enc=True,
enc_step=28,
snr=SNR,
property='Test'
)
# the raw dataset without noise
# test_X1, test_Y1 = prepro(d_path=path,
# length=length,
# number=100,
# normal=False,
# enc=True,
# enc_step=28,
# snr=SNR,
# property='Test',
# noise=False
# )
test_X = test_X[:, np.newaxis, :]
test_dataset = CustomTensorDataset(torch.tensor(test_X, dtype=torch.float),
torch.tensor(test_Y))
x = test_dataset.X
y = test_dataset.y
for name, module in model.named_children():
if 'cnn' in name:
module.Conv1D_1.register_forward_hook(hook)
module.Conv1D_2.register_forward_hook(hook)
module.Conv1D_3.register_forward_hook(hook)
module.Conv1D_4.register_forward_hook(hook)
module.Conv1D_5.register_forward_hook(hook)
module.Conv1D_6.register_forward_hook(hook)
wr1 = module.Conv1D_1.weight_r
wg1 = module.Conv1D_1.weight_g
wb1 = module.Conv1D_1.weight_b
br1 = module.Conv1D_1.bias_r
bg1 = module.Conv1D_1.bias_g
wr2 = module.Conv1D_2.weight_r
wg2 = module.Conv1D_2.weight_g
wb2 = module.Conv1D_2.weight_b
br2 = module.Conv1D_2.bias_r
bg2 = module.Conv1D_2.bias_g
wr3 = module.Conv1D_3.weight_r
wg3 = module.Conv1D_3.weight_g
wb3 = module.Conv1D_3.weight_b
br3 = module.Conv1D_3.bias_r
bg3 = module.Conv1D_3.bias_g
wr4 = module.Conv1D_4.weight_r
wg4 = module.Conv1D_4.weight_g
wb4 = module.Conv1D_4.weight_b
br4 = module.Conv1D_4.bias_r
bg4 = module.Conv1D_4.bias_g
wr5 = module.Conv1D_5.weight_r
wg5 = module.Conv1D_5.weight_g
wb5 = module.Conv1D_5.weight_b
br5 = module.Conv1D_5.bias_r
bg5 = module.Conv1D_5.bias_g
wr6 = module.Conv1D_6.weight_r
wg6 = module.Conv1D_6.weight_g
wb6 = module.Conv1D_6.weight_b
br6 = module.Conv1D_6.bias_r
bg6 = module.Conv1D_6.bias_g
outputs = []
y_pre = []
outcnn = []
for i in range(len(y)):
input_tensor = x[i, :, :]
input_tensor = input_tensor[:, np.newaxis, :]
output = attention_map(input_tensor, wr1, wg1, wb1, br1, bg1, 8, 28, 64)
output = output.detach().numpy().squeeze()
yhat = model(input_tensor)
y_predict = yhat.argmax(dim=1)
y_pre.append(y_predict.detach().numpy())
input_tensor2 = features_out_hook[0][0, 0, :].reshape(1, 1, -1)
wrr2 = wr2[:, 0, :].reshape(wr2.shape[0], 1, -1)
wgg2 = wg2[:, 0, :].reshape(wg2.shape[0], 1, -1)
wbb2 = wb2[:, 0, :].reshape(wb2.shape[0], 1, -1)
output2 = attention_map(input_tensor2, wrr2, wgg2, wbb2, br2, bg2, 1, 1, 3)
output2 = F.interpolate(output2.unsqueeze(0), length)
output2 = output2.detach().numpy().squeeze()
input_tensor3 = features_out_hook[1][0, 0, :].reshape(1, 1, -1)
wrr3 = wr3[:, 0, :].reshape(wr3.shape[0], 1, -1)
wgg3 = wg3[:, 0, :].reshape(wg3.shape[0], 1, -1)
wbb3 = wb3[:, 0, :].reshape(wb3.shape[0], 1, -1)
output3 = attention_map(input_tensor3, wrr3, wgg3, wbb3, br3, bg3, 1, 1, 3)
input_tensor3 = F.interpolate(input_tensor3, length)
output3 = F.interpolate(output3.unsqueeze(0), length)
output3 = output3.detach().numpy().squeeze()
input_tensor4 = features_out_hook[2][0, 0, :].reshape(1, 1, -1)
wrr4 = wr4[:, 0, :].reshape(wr4.shape[0], 1, -1)
wgg4 = wg4[:, 0, :].reshape(wg4.shape[0], 1, -1)
wbb4 = wb4[:, 0, :].reshape(wb4.shape[0], 1, -1)
output4 = attention_map(input_tensor4, wrr4, wgg4, wbb4, br4, bg4, 1, 1, 3)
input_tensor4 = F.interpolate(input_tensor4, length)
output4 = F.interpolate(output4.unsqueeze(0), length)
output4 = output4.detach().numpy().squeeze()
input_tensor5 = features_out_hook[3][0, 0, :].reshape(1, 1, -1)
wrr5 = wr5[:, 0, :].reshape(wr5.shape[0], 1, -1)
wgg5 = wg5[:, 0, :].reshape(wg5.shape[0], 1, -1)
wbb5 = wb5[:, 0, :].reshape(wb5.shape[0], 1, -1)
output5 = attention_map(input_tensor5, wrr5, wgg5, wbb5, br5, bg5, 1, 1, 3)
input_tensor5 = F.interpolate(input_tensor5, length)
output5 = F.interpolate(output5.unsqueeze(0), length)
output5 = output5.detach().numpy().squeeze()
input_tensor6 = features_out_hook[4][0, 0, :].reshape(1, 1, -1)
wrr6 = wr6[:, 0, :].reshape(wr6.shape[0], 1, -1)
wgg6 = wg6[:, 0, :].reshape(wg6.shape[0], 1, -1)
wbb6 = wb6[:, 0, :].reshape(wb6.shape[0], 1, -1)
output6 = attention_map(input_tensor6, wrr6, wgg6, wbb6, br6, bg6, 1, 0, 3)
input_tensor6 = F.interpolate(input_tensor6, length)
output6 = F.interpolate(output6.unsqueeze(0), length)
output6 = output6.detach().numpy().squeeze()
output = np.abs(np.gradient(output[0, :]))
output2 = np.abs(np.gradient(output2[0, :]))
output3 = np.abs(np.gradient(output3[0, :]))
output4 = np.abs(np.gradient(output4[0, :]))
output5 = np.abs(np.gradient(output5[0, :]))
output6 = np.abs(np.gradient(output6[0, :]))
map = [output, output2, output3, output4, output5, output6]
map = np.array(map)
outputs.append(map)
print('Process X: %d' % (i))
# Saving qttention maps of the QCNN in *.csv file
qmap = {}
for j in range(10):
idx = np.argwhere(y == j)
temp = np.array(outputs[idx[0,0]:idx[0,-1] +1])
t = temp[1, :, :]
for i, l in enumerate(temp):
if i > 0:
t = np.hstack((t, l))
qmap[j] = t
pd_0 = pd.DataFrame(qmap[0])
pd_1 = pd.DataFrame(qmap[1])
pd_2 = pd.DataFrame(qmap[2])
pd_3 = pd.DataFrame(qmap[3])
pd_4 = pd.DataFrame(qmap[4])
pd_5 = pd.DataFrame(qmap[5])
pd_6 = pd.DataFrame(qmap[6])
pd_7 = pd.DataFrame(qmap[7])
pd_8 = pd.DataFrame(qmap[8])
pd_9 = pd.DataFrame(qmap[9])
pd_10 = pd.DataFrame(y_pre)
pd_11 = pd.DataFrame(test_Y)
if not os.path.exists('data/qmaps'):
os.makedirs('data/qmaps')
pd_0.to_csv('data/qmaps/cqmaps_%s_snr_%d_class_0.csv' % (chosen_data, SNR), header=False)
pd_1.to_csv('data/qmaps/cqmaps_%s_snr_%d_class_1.csv' % (chosen_data, SNR), header=False,)
pd_2.to_csv('data/qmaps/cqmaps_%s_snr_%d_class_2.csv' % (chosen_data, SNR), header=False,)
pd_3.to_csv('data/qmaps/cqmaps_%s_snr_%d_class_3.csv' % (chosen_data, SNR), header=False,)
pd_4.to_csv('data/qmaps/cqmaps_%s_snr_%d_class_4.csv' % (chosen_data, SNR), header=False,)
pd_5.to_csv('data/qmaps/cqmaps_%s_snr_%d_class_5.csv' % (chosen_data, SNR), header=False,)
pd_6.to_csv('data/qmaps/cqmaps_%s_snr_%d_class_6.csv' % (chosen_data, SNR), header=False,)
pd_7.to_csv('data/qmaps/cqmaps_%s_snr_%d_class_7.csv' % (chosen_data, SNR), header=False,)
pd_8.to_csv('data/qmaps/cqmaps_%s_snr_%d_class_8.csv' % (chosen_data, SNR), header=False,)
pd_9.to_csv('data/qmaps/cqmaps_%s_snr_%d_class_9.csv' % (chosen_data, SNR), header=False,)
pd_10.to_csv('data/qmaps/predict.csv', header=False,)
pd_11.to_csv('data/qmaps/truelabel.csv', header=False,)
# Saving input signals in *.csv file
input = {}
for j in range(10):
idx = np.argwhere(test_Y == j)
temp = test_X[idx[0, 0]: idx[-1, 0] + 1].ravel()
input[j] = temp
pd_0 = pd.DataFrame(input[0])
pd_1 = pd.DataFrame(input[1])
pd_2 = pd.DataFrame(input[2])
pd_3 = pd.DataFrame(input[3])
pd_4 = pd.DataFrame(input[4])
pd_5 = pd.DataFrame(input[5])
pd_6 = pd.DataFrame(input[6])
pd_7 = pd.DataFrame(input[7])
pd_8 = pd.DataFrame(input[8])
pd_9 = pd.DataFrame(input[9])
if not os.path.exists('data/input'):
os.makedirs('data/input')
pd_0.to_csv('data/input/input_%s_snr_%d_class_0.csv' % (chosen_data, SNR), header=False)
pd_1.to_csv('data/input/input_%s_snr_%d_class_1.csv' % (chosen_data, SNR), header=False, )
pd_2.to_csv('data/input/input_%s_snr_%d_class_2.csv' % (chosen_data, SNR), header=False, )
pd_3.to_csv('data/input/input_%s_snr_%d_class_3.csv' % (chosen_data, SNR), header=False, )
pd_4.to_csv('data/input/input_%s_snr_%d_class_4.csv' % (chosen_data, SNR), header=False, )
pd_5.to_csv('data/input/input_%s_snr_%d_class_5.csv' % (chosen_data, SNR), header=False, )
pd_6.to_csv('data/input/input_%s_snr_%d_class_6.csv' % (chosen_data, SNR), header=False, )
pd_7.to_csv('data/input/input_%s_snr_%d_class_7.csv' % (chosen_data, SNR), header=False, )
pd_8.to_csv('data/input/input_%s_snr_%d_class_8.csv' % (chosen_data, SNR), header=False, )
pd_9.to_csv('data/input/input_%s_snr_%d_class_9.csv' % (chosen_data, SNR), header=False, )
# input2 = {}
# for j in range(10):
# idx = np.argwhere(test_Y == j)
# temp = test_X1[idx[0, 0]: idx[-1, 0] + 1].ravel()
# input2[j] = temp
#
# pd_0 = pd.DataFrame(input2[0])
# pd_1 = pd.DataFrame(input2[1])
# pd_2 = pd.DataFrame(input2[2])
# pd_3 = pd.DataFrame(input2[3])
# pd_4 = pd.DataFrame(input2[4])
# pd_5 = pd.DataFrame(input2[5])
# pd_6 = pd.DataFrame(input2[6])
# pd_7 = pd.DataFrame(input2[7])
# pd_8 = pd.DataFrame(input2[8])
# pd_9 = pd.DataFrame(input2[9])
#
# pd_0.to_csv('data/input/rawinput_%s_snr_%d_class_0.csv' % (chosen_data, SNR), header=False)
# pd_1.to_csv('data/input/rawinput_%s_snr_%d_class_1.csv' % (chosen_data, SNR), header=False, )
# pd_2.to_csv('data/input/rawinput_%s_snr_%d_class_2.csv' % (chosen_data, SNR), header=False, )
# pd_3.to_csv('data/input/rawinput_%s_snr_%d_class_3.csv' % (chosen_data, SNR), header=False, )
# pd_4.to_csv('data/input/rawinput_%s_snr_%d_class_4.csv' % (chosen_data, SNR), header=False, )
# pd_5.to_csv('data/input/rawinput_%s_snr_%d_class_5.csv' % (chosen_data, SNR), header=False, )
# pd_6.to_csv('data/input/rawinput_%s_snr_%d_class_6.csv' % (chosen_data, SNR), header=False, )
# pd_7.to_csv('data/input/rawinput_%s_snr_%d_class_7.csv' % (chosen_data, SNR), header=False, )
# pd_8.to_csv('data/input/rawinput_%s_snr_%d_class_8.csv' % (chosen_data, SNR), header=False, )
# pd_9.to_csv('data/input/rawinput_%s_snr_%d_class_9.csv' % (chosen_data, SNR), header=False, )