/
RO_diagnoser_train_script.py
199 lines (180 loc) · 7.2 KB
/
RO_diagnoser_train_script.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
'''
The identifier of RO systems. Compared with RO_identifier.py, the outputs are different.
'''
import os
import torch
import argparse
import torch.optim as optim
import numpy as np
import matplotlib.pyplot as plt
from torch.nn import MSELoss
from torch.nn import CrossEntropyLoss
from cnn_gru_diagnoser import cnn_gru_mode_detector
from cnn_gru_diagnoser import cnn_gru_pf_isolator
from cnn_gru_diagnoser import cnn_gru_pf_identifier
from cnn_gru_obs_model import cnn_gru_obs_model
from data_manager import data_manager
from utilities import np2tensor
def mse(input, target, use_cuda=True):
'''
input: tensor, should be cuda is avaliable
target: np.array
'''
target = torch.tensor(target).float().cuda() if torch.cuda.is_available() and use_cuda \
else torch.tensor(target).float()
loss = MSELoss()
return loss(input, target)
def cross_entropy(input, target, use_cuda=True):
'''
input: tensor, should be cuda is avaliable
target: np.array
'''
target = torch.tensor(target).long().cuda() if torch.cuda.is_available() and use_cuda \
else torch.tensor(target).long()
target = target.view(-1)
_, _, C = input.size()
input = input.view(-1, C)
loss = CrossEntropyLoss()
return loss(input, target)
def new_data_manager(cfg, si):
data_mana = data_manager(cfg, si)
return data_mana
def sample_data(data_mana, batch, normal_proportion, snr_or_pro, mask, res):
r = data_mana.sample_all(size=batch, \
normal_proportion=normal_proportion, \
snr_or_pro=snr_or_pro,\
norm_o=np.array([1,1,1,10,10e8]), \
norm_s=np.array([1,1,1,10,10e8,10e8]),\
mask=mask,\
res=res)
return r
def show_loss(i, loss, running_loss):
running_loss += loss.item()
if i%10==9:
ave_loss = running_loss / 10
msg = '# %d loss:%.3f' %(i + 1, ave_loss)
print(msg)
running_loss = 0
else:
print('#', end='', flush=True)
return running_loss
def train(save_path, model_name, model_type, epoch, batch, normal_proportion, \
data_mana, diagnoser, optimizer, obs_snr, mask, use_cuda, res):
train_loss = []
running_loss = 0
for i in range(epoch):
optimizer.zero_grad()
x, m, state, fp_mode, fp_value = sample_data(data_mana, batch, normal_proportion=normal_proportion, snr_or_pro=obs_snr, mask=mask, res=res)
m = m%3
y_head = diagnoser(np2tensor(x, use_cuda))
if model_type=='detector':
loss = cross_entropy(y_head, m, use_cuda)
elif model_type=='isolator':
loss = cross_entropy(y_head, fp_mode, use_cuda)
elif model_type=='f_f' or model_type=='f_r' or model_type=='f_m' :
fp_vec = (np.sum(fp_value, (0, 1))!=0)
index = np.sum(fp_vec*np.array([0, 1, 2]))
y = fp_value[:, :, [index]]
loss = 400*mse(y_head, y, use_cuda)
elif model_type=='obs':
loss = 10*mse(y_head, state, use_cuda)
else:
raise RuntimeError('Unknown Type.')
train_loss.append(loss.item())
running_loss = show_loss(i, loss, running_loss)
loss.backward()
optimizer.step()
save_model(diagnoser, save_path, model_name)
plot(train_loss, save_path, model_name)
return train_loss
def get_model(t, use_cuda=True):
print('CNN-GRU Model')
if t=='detector':
model = cnn_gru_mode_detector( x_size=5,\
cnn_feature_map=[32, 64, 128, 64], cnn_kernel_size=[64, 32, 16, 8],\
num_layers=2, hidden_size=64, dropout=0.5, \
fc_size=[64, 32], mode_size=3)
elif t=='isolator':
model = cnn_gru_pf_isolator(x_size=5,\
cnn_feature_map=[32, 64, 128, 64], cnn_kernel_size=[64, 32, 16, 8],\
num_layers=2, hidden_size=64, dropout=0.5, \
fc_size=[64, 32], pf_size=3)
elif t=='identifier':
model = cnn_gru_pf_identifier(x_size=5,\
cnn_feature_map=[32, 64, 128, 64], cnn_kernel_size=[64, 32, 16, 8],\
num_layers=2, hidden_size=64, dropout=0.5, \
fc_size=[64, 32])
elif t=='obs':
model = cnn_gru_obs_model(x_size=5, \
cnn_feature_map=[32, 64, 128, 64], cnn_kernel_size=[64, 32, 16, 8],\
num_layers=2, hidden_size=64, dropout=0.5, \
fc_size=[64, 32], o_size=6)
else:
raise RuntimeError('Unknown Type.')
if torch.cuda.is_available() and use_cuda:
model.cuda()
return model
def save_model(model, path, name):
torch.save(model, os.path.join(path, name))
def plot(train_loss, path, name):
plt.cla()
plt.plot(np.array(train_loss))
plt.title("Training Loss")
plt.xlabel("Epoch")
plt.ylabel("Loss")
plt.savefig(os.path.join(path, name+'.svg'), format='svg')
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('-d', '--data', type=str, help='choose the key values.')
parser.add_argument('-t', '--type', type=str, help='model type.')
parser.add_argument('-o', '--output', type=str, help='choose output names.')
parser.add_argument('-r', '--res', type=str, help='residual.')
args = parser.parse_args()
use_cuda, res = False, (args.res is None)
data_set, model_name = args.data, args.output
# mask
if args.type=='detector':# mode detector
model = get_model('detector', use_cuda)
mask = []
normal_proportion = 0.1
res = False
elif args.type=='isolator': # fault parameter isolator
model = get_model('isolator', use_cuda)
mask = ['s_mode1', 's_mode2', 's_mode3']
normal_proportion = 0.625
elif args.type=='f_f':
model = get_model('identifier', use_cuda)
mask = ['normal', 's_mode1', 's_mode2', 's_mode3', 'f_r', 'f_m']
normal_proportion = 0
elif args.type=='f_r':
model = get_model('identifier', use_cuda)
mask = ['normal', 's_mode1', 's_mode2', 's_mode3', 'f_f', 'f_m']
normal_proportion = 0
elif args.type=='f_m':
model = get_model('identifier', use_cuda)
mask = ['normal', 's_mode1', 's_mode2', 's_mode3', 'f_f', 'f_r']
normal_proportion = 0
elif args.type=='obs':
model = get_model('obs', use_cuda)
mask = []
normal_proportion = 0.1
res = False
else:
raise RuntimeError('Unknown Type.')
save_path = 'model'
if not os.path.isdir(save_path):
os.makedirs(save_path)
epoch = 2000
batch = 7 # 7 is used to debug. When train it on cloud, set it as 20 or 40.
# data manager
si = 0.01
obs_snr = 20
data_cfg = '{}/RO.cfg'.format(data_set)
if not os.path.exists(data_cfg):
raise RuntimeError('Data set does not exist.')
data_mana = new_data_manager(data_cfg, si)
# optimizer
optimizer = optim.Adam(model.parameters(), lr=0.001, weight_decay=1e-3)
# train
train(save_path, model_name, args.type, epoch, batch, normal_proportion, \
data_mana, model, optimizer, obs_snr, mask, use_cuda, res)