forked from shipengai/ECG-Segment-LSTM
-
Notifications
You must be signed in to change notification settings - Fork 0
/
model_lstm.py
287 lines (242 loc) · 9.37 KB
/
model_lstm.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
# -*- coding:utf-8 -*-
"""
本实验基于类似class-oriented 所以效果要好很多
基于record-oriented未做,改一下数据集划分便可以。
"""
import os
import torch
from torch import nn
from torch.utils.data import Dataset, DataLoader
import matplotlib.pyplot as plt
import pickle
import numpy as np
from matplotlib.backends.backend_pdf import PdfPages
TRAIN = False # 训练标志
CONTINUE_TRAIN = False # 接着上次某一次训练结果继续训练
TEST = False # 测试标志 设置成True时候,需要指定加载哪个模型
PAPER_TEST = True # 得到写paper使用的测试指标
SAVE_TEST_FIG = True
EPOCHS = 100
BATCH_SIZE = 32
Seqlength = 300
NUM_SEGS_CLASS = 5
qtdb_pkl = './qtdb_pkl/' # 数据预处理后的路径,便于调试网络
save_path = './ckpt/' # 保存模型的路径
if not os.path.exists(save_path):
os.mkdir(save_path)
class ECGDataset(Dataset):
"""ecg dataset.
返回字典:{'signal': ,'label': }
"""
def __init__(self, qtdb_pkl, data):
"""
:param qtdb_pkl: 数据库存放路径
:param data: 训练集和验证集数据
"""
pkl = os.path.join(qtdb_pkl, data)
with open(pkl, 'rb') as f:
x, y = pickle.load(f)
self.x = x
self.y = y
def __len__(self):
return len(self.x)
def __getitem__(self, idx):
signal = torch.from_numpy(self.x[idx]).float()
label = torch.from_numpy(self.y[idx]).float()
sample = {'signal': signal, 'label': label}
return sample
class SegModel(torch.nn.Module):
def __init__(self, input_size, hidden_size, num_layers, out_size):
super().__init__()
self.features = torch.nn.Sequential(
# torch.nn.Linear(in_features=input_size, out_features=hidden_size),
torch.nn.LSTM(input_size=input_size,
hidden_size=hidden_size,
num_layers=num_layers,
batch_first=True,
bidirectional=True),
)
self.classifier = torch.nn.Sequential(
torch.nn.Linear(2*hidden_size, 2*hidden_size),
torch.nn.ReLU(inplace=True),
torch.nn.Dropout(),
torch.nn.Linear(2 * hidden_size, 2 * hidden_size),
torch.nn.ReLU(inplace=True),
torch.nn.Dropout(),
)
self.output = torch.nn.Linear(2*hidden_size, out_size)
def forward(self, x):
"""
:param x: shape(batch, seq_len, input_size)
:return:
"""
batch, seq_len, nums_fea = x.size()
features, _ = self.features(x)
output = self.classifier(features)
output = self.output(output.view(batch * seq_len, -1))
return output
def train(net, data_loader, epochs):
for step in range(epochs):
net.train()
for i, samples_batch in enumerate(data_loader):
total = 0.0
correct = 0.0
output = net(samples_batch['signal'])
target = samples_batch['label'].contiguous().view(-1).long()
loss = criterion(output, target)
_, predicted = torch.max(output.data, 1)
total += target.size(0)
correct += (predicted == target).sum().item()
if (i+1) % 20 == 0:
print("EPOCHS:{},Iter:{},Loss:{:.4f},Acc:{:.4f}".format(step, i+1, loss.item(), correct/total))
optimizer.zero_grad()
loss.backward()
optimizer.step()
# 每2个epoch,保存一次模型
if (step+1) % 2 == 0:
torch.save(net, save_path+'epoch_{}.ckpt'.format(step))
test(ecg_train_dl, 'train', step)
test(ecg_val_dl, 'val', step)
def test(data_loader, str1, step):
with torch.no_grad():
right = 0.0
total = 0.0
net.eval()
for sample in data_loader:
output = net(sample['signal'])
_, predicted = torch.max(output.data, 1)
label = sample['label'].contiguous().view(-1).long()
total += label.size(0)
right += (predicted == label).sum().item()
print("epoch:{},{} ACC: {:.4f}".format(step, str1, right / total))
def restore_net(ckpt):
# load models
with open(ckpt, 'rb') as f:
net = torch.load(f)
return net
def get_charateristic(y):
Ppos = Qpos = Rpos =Spos = Tpos = 0
for i, val in enumerate(y):
if val == 1 and y[i-1] == 0:
Ppos = i
if val == 2 and y[i-1] == 0:
Qpos = i
if val == 2 and y[i+1] == 3:
Rpos = i
if val == 3 and y[i+1] == 0:
Spos = i
if val == 4 and y[i-1] == 0:
Tpos = i
return Ppos, Qpos, Rpos, Spos, Tpos
def point_equal(label, predict, tolerte):
if predict <= label + tolerte * 250 and predict >= label- tolerte * 250:
return True
else:
return False
def right_point(label_tuple, predict_tuple, tolerte):
n = np.array([0, 0, 0, 0, 0])
for i, (x, x_p) in enumerate(zip(label_tuple, predict_tuple)):
if point_equal(x, x_p, tolerte):
n[i] = 1
return n
def plotlabel(y, bias):
cmap = ['k', 'r', 'g', 'b', 'c', 'y']
start = end = 0
for i in range(len(y) - 1):
if y[i] != y[i + 1]:
end = i
plt.plot(np.arange(start, end), y[start:end] - bias, cmap[int(y[i])])
start = i + 1
if i == len(y) - 2:
end = len(y) - 1
plt.plot(np.arange(start, end), y[start:end] - bias, cmap[int(y[i])])
def caculate_error(label_tuple, predict_tuple):
error = np.zeros((5,))
for i, (x, x_p) in enumerate(zip(label_tuple, predict_tuple)):
error[i] = (x - x_p)/250*100 # (ms)
return error
if __name__ == '__main__':
# loading data
ecg_train_db = ECGDataset(qtdb_pkl, 'train_data.pkl')
ecg_train_dl = DataLoader(ecg_train_db, batch_size=BATCH_SIZE,
shuffle=True, num_workers=1)
ecg_val_db = ECGDataset(qtdb_pkl, 'val_data.pkl')
ecg_val_dl = DataLoader(ecg_val_db, batch_size=BATCH_SIZE,
shuffle=False, num_workers=1)
if TRAIN:
if CONTINUE_TRAIN:
# continue training
net = restore_net(save_path + 'epoch_102.ckpt')
else:
# model
net = SegModel(input_size=2, hidden_size=32, num_layers=2, out_size=NUM_SEGS_CLASS)
optimizer = torch.optim.Adam(net.parameters(), lr=1e-3)
criterion = nn.CrossEntropyLoss()
optimizer.zero_grad()
train(net, ecg_train_dl, EPOCHS)
if TEST:
# vis
net = restore_net(save_path+'epoch_99.ckpt')
net.eval()
# test(ecg_val_dl, 'val', 4)
for i, idx in enumerate([20, 60,
160, 280]):
sample = ecg_val_db[idx]
signal = sample['signal'].numpy()
label = sample['label'].numpy()
# plotecg(signal, label, 0, 1300)
output = net(sample['signal'].unsqueeze(0))
_, predict = torch.max(output, 1)
# 将predict 和 label画出来
predict = predict.numpy()
x = np.arange(len(predict))
plt.subplot(2, 2, i+1)
plt.plot(x, signal[:, 0])
plotlabel(label, 0.2)
plotlabel(predict, 0.4)
plt.show()
if PAPER_TEST:
net = restore_net(save_path + 'epoch_99.ckpt')
net.eval()
print('waiting several minutes')
right_point_num = np.array([0, 0, 0, 0, 0])
error_array = np.zeros(shape=(len(ecg_val_db), 5))
if SAVE_TEST_FIG:
with PdfPages('test.pdf') as pdf:
for i in range(len(ecg_val_db)):
sample = ecg_val_db[i]
signal = sample['signal'].numpy()
label = sample['label'].numpy()
# 得到预测结果
output = net(sample['signal'].unsqueeze(0))
_, predict = torch.max(output, 1)
predict = predict.numpy()
x = np.arange(Seqlength)
plt.plot(x, signal[:, 0])
plotlabel(label, 0.2)
plotlabel(predict, 0.4)
pdf.savefig()
plt.close()
label_points = get_charateristic(label)
predict_points = get_charateristic(predict)
error_array[i] = caculate_error(label_points, predict_points)
# 得到p-end, QRS onset end , T-middle
right_point_num += right_point(label_points,
predict_points, 0.016)
else:
for i in range(len(ecg_val_db)):
sample = ecg_val_db[i]
signal = sample['signal'].numpy()
label = sample['label'].numpy()
# 得到预测结果
output = net(sample['signal'].unsqueeze(0))
_, predict = torch.max(output, 1)
predict = predict.numpy()
# 得到p-end, QRS onset end , T-middle
right_point_num += right_point(get_charateristic(label),
get_charateristic(predict), 0.025)
means = np.mean(error_array, axis=0)
SD = np.std(error_array, axis=0)
print(means)
print(SD)
print(right_point_num/len(ecg_val_db))