-
Notifications
You must be signed in to change notification settings - Fork 16
/
Copy pathTrainer.py
368 lines (300 loc) · 16.5 KB
/
Trainer.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
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
import os
import torch
import json
import numpy as np
import torch.optim as optim
import torch.nn.functional as F
from tqdm import tqdm
from torch import nn
from pathlib import Path
from NILM_Dataloader import *
from metrics import *
class Trainer:
def __init__(self,args,ds_parser,model):
self.args = args
self.device = args.device
self.pretrain = args.pretrain
self.pretrain_num_epochs = args.pretrain_num_epochs
self.num_epochs = args.num_epochs
self.model = model.to(args.device)
self.export_root = Path(args.export_root).joinpath(args.dataset_code).joinpath(args.appliance_names[0])
self.best_model_epoch = None
self.cutoff = torch.tensor(args.cutoff[args.appliance_names[0]] ).to(self.device)
self.threshold = torch.tensor(args.threshold[args.appliance_names[0]] ).to(self.device)
self.min_on = torch.tensor(args.min_on[args.appliance_names[0]] ).to(self.device)
self.min_off = torch.tensor(args.min_off[args.appliance_names[0]] ).to(self.device)
self.C0 = torch.tensor(args.c0[args.appliance_names[0]] ).to(self.device)
self.tau = args.tau
if self.pretrain:
dataloader = NILMDataloader(args, ds_parser, pretrain=True)
self.pretrain_loader, self.pretrain_val_loader = dataloader.get_dataloaders()
dataloader = NILMDataloader(args, ds_parser, pretrain=False)
self.train_loader, self.val_loader = dataloader.get_dataloaders()
self.optimizer = self._create_optimizer()
if args.enable_lr_schedule:
self.lr_scheduler = optim.lr_scheduler.StepLR(self.optimizer,
step_size=args.decay_step,
gamma=args.gamma
)
self.normalize = args.normalize
if self.normalize == 'mean':
self.x_mean, self.x_std = ds_parser.x_mean,ds_parser.x_std
self.x_mean = torch.tensor(self.x_mean).to(self.device)
self.x_std = torch.tensor(self.x_std ).to(self.device)
self.mse = nn.MSELoss()
self.kl = nn.KLDivLoss( reduction = 'batchmean')
self.bceloss = nn.BCEWithLogitsLoss(reduction = 'mean')
self.margin = nn.SoftMarginLoss()
self.l1_on = nn.L1Loss(reduction='sum')
# per epoch
self.train_metrics_dict = {
'mae' : [],
'mre' : [],
'acc' : [],
'precision': [],
'recall' : [],
'f1' : []
}
# per validate() run
self.val_metrics_dict = {
'mae' : [],
'mre' : [],
'acc' : [],
'precision': [],
'recall' : [],
'f1' : []
}
# test set
self.test_metrics_dict = {
'mae' : [],
'mre' : [],
'acc' : [],
'precision': [],
'recall' : [],
'f1' : []
}
self.training_loss = []
self.y_pred_curve, self.y_curve, self.s_pred_curve, self.status_curve = [], [], [], []
def train(self):
_, best_mre, best_acc, _, _, best_f1 = self.validate()
self._save_state_dict()
if self.pretrain:
for epoch in range(self.pretrain_num_epochs):
self.pretrain_one_epoch(epoch+1)
self.model.pretrain = False
for epoch in range(self.num_epochs):
self.train_one_epoch(epoch+1)
mae, mre, acc, precision, recall, f1 = self.validate()
self.update_metrics_dict(mae, mre, acc, precision, recall, f1, mode = 'train')
if f1 + acc - mre > best_f1 + best_acc - best_mre:
best_f1 = f1
best_acc = acc
best_mre = mre
self.best_model_epoch = epoch
self._save_state_dict()
def pretrain_one_epoch(self,epoch):
loss_values = []
self.model.train()
tqdm_dataloader = tqdm(self.pretrain_loader)
for _,batch in enumerate(tqdm_dataloader):
x, y, status = [batch[i].to(self.device) for i in range(3)]
self.optimizer.zero_grad()
mask = (status >= 0)
y_capped = y / self.cutoff
logits, gen_out, logits_y, logits_status = self.get_model_outputs(x,mask)
logits_masked = torch.masked_select(logits , mask).view((-1))
labels_masked = torch.masked_select(y_capped , mask).view((-1))
# status_masked = torch.masked_select(status , mask).view((-1))
# logits_status_masked = torch.masked_select(logits_status, mask).view((-1))
gen_out = gen_out.view(-1)
mask = mask.view(-1).type(torch.DoubleTensor).to(self.device)
total_loss = self.loss_fn_pretrain(logits_masked,labels_masked,gen_out,mask)
total_loss.backward()
self.optimizer.step()
loss_values.append(total_loss.item())
average_loss = np.mean(np.array(loss_values))
self.training_loss.append(average_loss)
tqdm_dataloader.set_description('Epoch {}, loss {:.2f}'.format(epoch, average_loss))
if self.args.enable_lr_schedule:
self.lr_scheduler.step()
def train_one_epoch(self,epoch):
loss_values = []
self.model.train()
tqdm_dataloader = tqdm(self.train_loader)
for _,batch in enumerate(tqdm_dataloader):
x, y, status = [batch[i].to(self.device) for i in range(3)]
self.optimizer.zero_grad()
y_capped = y / self.cutoff
logits,_, logits_y, logits_status = self.get_model_outputs(x)
total_loss = self.loss_fn_train(logits,y_capped,logits_status,status)
total_loss.backward()
self.optimizer.step()
loss_values.append(total_loss.item())
average_loss = np.mean(np.array(loss_values))
self.training_loss.append(average_loss)
tqdm_dataloader.set_description('Epoch {}, loss {:.2f}'.format(epoch, average_loss))
if self.args.enable_lr_schedule:
self.lr_scheduler.step()
def validate(self):
self.model.eval()
self.val_metrics_dict = {
'mae' : [],
'mre' : [],
'acc' : [],
'precision': [],
'recall' : [],
'f1' : []
}
with torch.no_grad():
tqdm_dataloader = tqdm(self.val_loader)
for _,batch in enumerate(tqdm_dataloader):
x, y, status = [batch[i].to(self.device) for i in range(3)]
y_capped = y / self.cutoff
_,_, logits_y, logits_status = self.get_model_outputs(x)
logits_y = logits_y * logits_status
acc,precision,recall,f1 = acc_precision_recall_f1_score(logits_status,status)
mae, mre = regression_errors(logits_y, y_capped)
self.update_metrics_dict(mae,mre,acc,precision,recall,f1,mode = 'val')
acc_mean = np.mean(np.concatenate(self.val_metrics_dict['acc']).reshape(-1))
f1_mean = np.mean(np.concatenate(self.val_metrics_dict['f1'] ).reshape(-1))
mre_mean = np.mean(np.concatenate(self.val_metrics_dict['mre']).reshape(-1))
tqdm_dataloader.set_description('Validation, rel_err {:.2f}, acc {:.2f}, f1 {:.2f}'.format(mre_mean, acc_mean, f1_mean))
return [np.array(np.concatenate(v)).mean(axis=0) for v in self.val_metrics_dict.values()]
def test(self,test_loader):
self._load_best_model()
self.model.eval()
y_pred_curve, y_curve,s_pred_curve,status_curve = [], [], [], []
with torch.no_grad():
tqdm_dataloader = tqdm(test_loader)
for _, batch in enumerate(tqdm_dataloader):
x, y, status = [batch[i].to(self.device) for i in range(3)]
y_capped = y / self.cutoff
_,_, logits_y, logits_status = self.get_model_outputs(x)
logits_y = logits_y * logits_status
acc,precision,recall,f1 = acc_precision_recall_f1_score(logits_status,status)
mae, mre = regression_errors(logits_y, y_capped)
self.update_metrics_dict(mae,mre,acc,precision,recall,f1, mode = 'test')
acc_mean = np.mean(np.concatenate(self.test_metrics_dict['acc']).reshape(-1))
f1_mean = np.mean(np.concatenate(self.test_metrics_dict['f1'] ).reshape(-1))
mre_mean = np.mean(np.concatenate(self.test_metrics_dict['mre']).reshape(-1))
tqdm_dataloader.set_description('Test, rel_err {:.2f}, acc {:.2f}, f1 {:.2f}'.format(mre_mean, acc_mean, f1_mean))
y_pred_curve.append(logits_y.detach().cpu().numpy().squeeze())
y_curve.append( y.detach( ).cpu().numpy().squeeze())
s_pred_curve.append(logits_status.detach().cpu().numpy().squeeze())
status_curve.append(status.detach().cpu().numpy().squeeze())
self.y_pred_curve = np.concatenate(y_pred_curve).reshape(1,-1)
self.y_curve = np.concatenate(y_curve ).reshape(1,-1)
self.s_pred_curve = np.concatenate(s_pred_curve).reshape(1,-1)
self.status_curve = np.concatenate(status_curve).reshape(1,-1)
self._save_result({'gt': self.y_curve.tolist(),'pred': self.y_pred_curve.tolist()}, 'test_result.json')
mre, mae = regression_errors(self.y_pred_curve, self.y_curve)
acc, precision,recall, f1 = acc_precision_recall_f1_score(self.s_pred_curve, self.status_curve)
return mre,mae,acc, precision,recall, f1
def _save_state_dict(self):
if not os.path.exists(self.export_root):
os.makedirs(self.export_root)
print('Saving best model...')
torch.save(self.model.state_dict(), self.export_root.joinpath('best_acc_model.pth'))
def update_metrics_dict(self,mae,mre,acc,precision,recall,f1, mode = 'val'):
if mode=='train':
self.train_metrics_dict['mae' ].append(mae)
self.train_metrics_dict['mre' ].append(mre)
self.train_metrics_dict['acc' ].append(acc)
self.train_metrics_dict['precision'].append(precision)
self.train_metrics_dict['recall' ].append(recall)
self.train_metrics_dict['f1' ].append(f1)
elif mode=='val':
self.val_metrics_dict['mae' ].append(mae)
self.val_metrics_dict['mre' ].append(mre)
self.val_metrics_dict['acc' ].append(acc)
self.val_metrics_dict['precision'].append(precision)
self.val_metrics_dict['recall' ].append(recall)
self.val_metrics_dict['f1' ].append(f1)
else:
self.test_metrics_dict['mae' ].append(mae)
self.test_metrics_dict['mre' ].append(mre)
self.test_metrics_dict['acc' ].append(acc)
self.test_metrics_dict['precision'].append(precision)
self.test_metrics_dict['recall' ].append(recall)
self.test_metrics_dict['f1' ].append(f1)
def get_model_outputs(self, x,mask=None):
logits, gen_out = self.model(x,mask)
logits_y = self.cutoff_energy(logits*self.cutoff)
logits_status = self.compute_status(logits_y)
return logits, gen_out, logits_y, logits_status
def cutoff_energy(self,data):
data[data<5] = 0
data = torch.min(data,self.cutoff.double())
return data
def compute_status(self,data):
status = (data>=self.threshold)*1
return status
def _create_optimizer(self):
param_optimizer = list(self.model.named_parameters())
no_decay = ['bias', 'layer_norm']
optimizer_grouped_parameters = [
{
'params': [p for n, p in param_optimizer if not any(nd in n for nd in no_decay)],
'weight_decay': self.args.weight_decay,
},
{'params': [p for n, p in param_optimizer if any(nd in n for nd in no_decay)], 'weight_decay': 0.0},
]
if self.args.optimizer.lower() == 'adamw':
return optim.AdamW(optimizer_grouped_parameters, lr=self.args.lr)
elif self.args.optimizer.lower() == 'adam':
return optim.Adam(optimizer_grouped_parameters, lr=self.args.lr)
elif self.args.optimizer.lower() == 'sgd':
return optim.SGD(optimizer_grouped_parameters, lr=self.args.lr, momentum=self.args.momentum)
else:
raise ValueError
def _save_state_dict(self):
if not os.path.exists(self.export_root):
os.makedirs(self.export_root)
print('Saving best model...')
torch.save(self.model.state_dict(), self.export_root.joinpath('best_acc_model.pth'))
def _load_best_model(self):
try:
self.model.load_state_dict(torch.load(self.export_root.joinpath('best_acc_model.pth')))
self.model.to(self.device)
except:
print('Failed to load best model, continue testing with current model...')
def _save_result(self,data,filename):
if not os.path.exists(self.export_root):
os.makedirs(self.export_root)
filepath = Path(self.export_root).joinpath(filename)
with filepath.open('w') as f:
json.dump(data, f, indent=2)
def loss_fn_gen(self,logits_masked,labels_masked):
mse_arg_1 = logits_masked.contiguous().view(-1).double()
mse_arg_2 = labels_masked.contiguous().view(-1).double()
kl_arg_1 = torch.log(F.softmax(logits_masked.squeeze()/self.tau, dim=-1) + 1e-9)
kl_arg_2 = F.softmax(labels_masked.squeeze()/self.tau, dim=-1)
mse_loss = self.mse(mse_arg_1,mse_arg_2)
kl_loss = self.kl(kl_arg_1,kl_arg_2)
loss = mse_loss + kl_loss
return loss
def loss_fn_disc(self,gen_out,mask):
return self.bceloss(gen_out,mask)
def loss_fn_pretrain(self,logits_masked,labels_masked,gen_out,mask):
gen_loss = self.loss_fn_gen(logits_masked,labels_masked)
disc_loss = self.loss_fn_disc(gen_out,mask)
return gen_loss + disc_loss
def loss_fn_train(self,logits,labels,logits_status,status):
kl_arg_1 = torch.log(F.softmax(logits.squeeze() / 0.1, dim=-1) + 1e-9)
kl_arg_2 = F.softmax(labels.squeeze() / 0.1, dim=-1)
mse_arg_1 = logits.contiguous().view(-1).double()
mse_arg_2 = labels.contiguous().view(-1).double()
margin_arg_1 = (logits_status * 2 - 1).contiguous().view(-1).double()
margin_arg_2 = (status * 2 - 1).contiguous().view(-1).double()
kl_loss = self.kl( kl_arg_1, kl_arg_2)
mse_loss = self.mse(mse_arg_1, mse_arg_2)
margin_loss = self.margin(margin_arg_1, margin_arg_2)
total_loss = kl_loss + mse_loss + margin_loss
on_mask = ((status == 1) + (status != logits_status.reshape(status.shape))) >= 1
if on_mask.sum() > 0:
total_size = torch.tensor(on_mask.shape).prod()
logits_on = torch.masked_select(logits.reshape(on_mask.shape), on_mask)
labels_on = torch.masked_select(labels.reshape(on_mask.shape), on_mask)
loss_l1_on = self.l1_on(logits_on.contiguous().view(-1), labels_on.contiguous().view(-1))
total_loss += self.C0 * loss_l1_on / total_size
return total_loss