-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_arnet_synthetic.py
305 lines (221 loc) · 12.4 KB
/
train_arnet_synthetic.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
# ------------------------------------------------------------------
"""
Script for training and validating on the Synthetic dataset for ARNet model
Contact Person: Mohamad Hakam Shams Eddin <[email protected]>
Computer Vision Group - Institute of Computer Science III - University of Bonn
"""
# ------------------------------------------------------------------
import torch
import numpy as np
from tqdm import tqdm
from Baselines_MIL.models.losses import RankingLoss, SmoothL2Loss, SparsityLoss, DMIL_RankingLoss, CenterLoss
import Baselines_MIL.utils.utils_train as utils
from Baselines_MIL.models.build_arnet import MIL_model
import time
import os
from torch.utils.tensorboard import SummaryWriter
from Baselines_MIL.dataset.Synthetic_dataset import Synthetic_Dataset
import Baselines_MIL.config as config_file
np.set_printoptions(suppress=True)
torch.set_printoptions(sci_mode=False)
np.seterr(divide='ignore', invalid='ignore')
#torch.autograd.set_detect_anomaly(True)
# ------------------------------------------------------------------
def train(config_file):
# read config arguments
config = config_file.read_arguments(train=True)
# get logger
logger = utils.get_logger(config)
# get tensorboard writer
writer = SummaryWriter(os.path.join(config.dir_log, config.name))
# fix random seed
utils.fix_seed(config.seed)
# dataloader
utils.log_string(logger, "loading training dataset ...")
train_dataset = Synthetic_Dataset(
root_datacube=config.root_synthetic,
times=config.times_train,
is_aug=config.is_aug,
is_norm=config.is_norm,
is_clima_scale=config.is_clima_scale,
variables=config.variables,
variables_static=config.variables_static,
x_min=config.x_min,
x_max=config.x_max,
y_min=config.y_min,
y_max=config.y_max,
delta_t=config.delta_t,
window_size=config.window_size
)
utils.log_string(logger, "loading validation dataset ...")
val_dataset = Synthetic_Dataset(
root_datacube=config.root_synthetic,
times=config.times_val,
is_aug=False,
is_norm=config.is_norm,
is_clima_scale=config.is_clima_scale,
variables=config.variables,
variables_static=config.variables_static,
x_min=config.x_min,
x_max=config.x_max,
y_min=config.y_min,
y_max=config.y_max,
delta_t=config.delta_t,
window_size=config.window_size
)
# random_sampler = torch.utils.data.RandomSampler(train_dataset, num_samples=len(val_dataset))
train_dataloader = torch.utils.data.DataLoader(train_dataset,
batch_size=config.batch_size,
# sampler=random_sampler,
shuffle=True,
pin_memory=config.pin_memory,
num_workers=config.n_workers)
# persistent_workers=True)
val_dataloader = torch.utils.data.DataLoader(val_dataset,
batch_size=config.batch_size,
drop_last=False,
shuffle=True,
pin_memory=config.pin_memory,
num_workers=config.n_workers)
# train_dataloader = val_dataloader
utils.log_string(logger, "# training samples: %d" % len(train_dataset))
utils.log_string(logger, "# evaluation samples: %d" % len(val_dataset))
# get models
utils.log_string(logger, "\nloading the model ...")
if config.gpu_id != -1:
# os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
# os.environ["CUDA_VISIBLE_DEVICES"] = str(config.gpu_id)
device = 'cuda'
else:
device = 'cpu'
model = MIL_model(config)
utils.log_string(logger, "model parameters ...")
utils.log_string(logger, "encoder parameters: %d" % utils.count_parameters(model.encoder))
utils.log_string(logger, "classifier parameters: %d" % utils.count_parameters(model.classifier))
utils.log_string(logger, "all parameters: %d\n" % utils.count_parameters(model))
# get losses
utils.log_string(logger, "get criterion ...")
#criterion_ranking = RankingLoss(drop_rate=0.).to(device)
criterion_ranking = DMIL_RankingLoss(alpha=config.loss_alpha_arnet,
t=train_dataset.n_lat_window * train_dataset.n_lon_window).to(device)
criterion_center = CenterLoss(lambda_c=config.loss_lambda_c_arnet).to(device)
# get optimizer
utils.log_string(logger, "get optimizer and learning rate scheduler ...")
optimizer = torch.optim.Adam(model.parameters(), lr=config.lr, weight_decay=config.weight_decay,
betas=(config.beta1, config.beta2))
lr_scheduler = utils.get_learning_scheduler(optimizer, config)
# if torch.cuda.device_count() > 1:
# model = torch.nn.DataParallel(model)
model.to(device)
# training loop
utils.log_string(logger, 'training on synthetic dataset ...\n')
# get collector for anomalous events prediction. Used in the evaluator
train_anomaly_collector = utils.anomaly_collector(train_dataset.anomaly, train_dataset.timestep, config)
val_anomaly_collector = utils.anomaly_collector(val_dataset.anomaly, val_dataset.timestep, config)
# get evaluator for anomalous events prediction
eval_train_anomaly = utils.evaluator_anomaly_synthetic(logger, 'Training', config)
eval_val_anomaly = utils.evaluator_anomaly_synthetic(logger, 'Validation', config)
time.sleep(1)
# initialize the best values
best_loss_train = np.inf
best_loss_val = np.inf
for epoch in range(config.n_epochs):
utils.log_string(logger, '################# Epoch (%s/%s) #################' % (epoch + 1, config.n_epochs))
# train
model = model.train()
loss_train = 0
time.sleep(1)
for i, (data_d, _, _, _, mask_extreme_loss, _, timestep) in tqdm(enumerate(train_dataloader),
total=len(train_dataloader),
smoothing=0.9,
postfix=" training"):
optimizer.zero_grad(set_to_none=True)
#_, z_n, z_p = model(data_d.to(device), mask_extreme_loss)
z_n, z_p = model(data_d.to(device), mask_extreme_loss)
loss_ranking, loss_centering = torch.zeros(1, device=device), torch.zeros(1, device=device)
for k in range(len(z_n)):
for v in range(config.in_channels_dynamic):
loss_ranking += criterion_ranking(z_p[k][:, v, ...], z_n[k][:, v, ...], is_training=True) #, k=int(torch.sum(mask_extreme_loss[k] == 1) * (1 - config.de_drop_rate)))
loss_centering += criterion_center(z_n[k][:, v, ...])
loss = (loss_ranking + loss_centering) / len(z_n)
loss.backward()
optimizer.step()
loss_train += loss.item()
anomaly = torch.zeros(data_d.shape, requires_grad=False, device=device)[:, :, 0, :, :, :]
# anomaly N, V, T, H, W
# z_p Np, V, T, C
anomaly = anomaly.permute(1, 2, 0, 3, 4) # anomaly V, T, N, H, W
# segment
for k in range(len(z_n)):
anomaly[:, :, k, mask_extreme_loss[k] == 0] = z_n[k][:, :, :, 0].permute(1, 2, 0)
anomaly[:, :, k, mask_extreme_loss[k] != 0] = z_p[k][:, :, :, 0].permute(1, 2, 0)
anomaly = anomaly.permute(2, 0, 1, 3, 4) # anomaly V, T, N, H, W
anomaly[anomaly > 0.5] = 1
anomaly[anomaly <= 0.5] = 0
#eval_train_anomaly(anomaly.detach().cpu().numpy(), mask_anomaly.cpu().numpy())
train_anomaly_collector(anomaly.detach().cpu().numpy(), timestep.detach().cpu().numpy())
mean_loss_train = loss_train / float(len(train_dataloader))
train_anomaly_collector.majority_vote()
eval_train_anomaly(np.swapaxes(train_anomaly_collector.anomaly, 0, 1), np.swapaxes(train_dataset.anomaly, 0, 1))
eval_train_anomaly.get_results()
utils.log_string(logger, '%s mean loss : %.4f' % ('Training', mean_loss_train))
utils.log_string(logger, '%s best mean loss: %.4f\n' % ('Training', best_loss_train))
if mean_loss_train <= best_loss_train:
best_loss_train = mean_loss_train
utils.save_model(model, optimizer, epoch, mean_loss_train, np.nan, logger, config, 'train')
# validation
with torch.no_grad():
model = model.eval()
loss_val = 0
time.sleep(1)
for i, (data_d, _, _, _, mask_extreme_loss, mask_anomaly, timestep) in tqdm(enumerate(val_dataloader),
total=len(val_dataloader),
smoothing=0.9,
postfix=" validation"):
#_, z_n, z_p = model(data_d.to(device), mask_extreme_loss)
z_n, z_p = model(data_d.to(device), mask_extreme_loss)
loss_ranking, loss_centering = torch.zeros(1, device=device), torch.zeros(1, device=device)
for k in range(len(z_n)):
for v in range(config.in_channels_dynamic):
loss_ranking += criterion_ranking(z_p[k][:, v, ...], z_n[k][:, v, ...]) #, k=int(torch.sum(mask_extreme_loss[k] == 1) * (1 - config.de_drop_rate)))
loss_centering += criterion_center(z_n[k][:, v, ...])
loss = (loss_ranking + loss_centering) / len(z_n)
loss_val += loss.item()
anomaly = torch.zeros(mask_anomaly.shape, requires_grad=False, device=device)
# anomaly N, V, T, H, W
# z_p Np, V, T, C
anomaly = anomaly.permute(1, 2, 0, 3, 4) # anomaly V, T, N, H, W
for k in range(len(z_n)):
anomaly[:, :, k, mask_extreme_loss[k] == 0] = z_n[k][:, :, :, 0].permute(1, 2, 0)
anomaly[:, :, k, mask_extreme_loss[k] != 0] = z_p[k][:, :, :, 0].permute(1, 2, 0)
anomaly = anomaly.permute(2, 0, 1, 3, 4) # anomaly V, T, N, H, W
anomaly[anomaly > 0.5] = 1
anomaly[anomaly <= 0.5] = 0
#eval_val_anomaly(anomaly.detach().cpu().numpy(), mask_anomaly.cpu().numpy())
val_anomaly_collector(anomaly.cpu().numpy(), timestep.cpu().numpy())
if i == len(val_dataloader) - 1:
# plot results
im_anomaly = utils.generate_anomaly(anomaly.cpu().numpy())
im_anomaly_gt = utils.generate_anomaly(mask_anomaly.cpu().numpy())
im_anomaly = np.concatenate((im_anomaly, im_anomaly_gt), axis=2)
mean_loss_val = loss_val / float(len(val_dataloader))
val_anomaly_collector.majority_vote()
eval_val_anomaly(np.swapaxes(val_anomaly_collector.anomaly, 0, 1), np.swapaxes(val_dataset.anomaly, 0, 1))
eval_val_anomaly.get_results()
utils.log_string(logger, '%s mean loss : %.4f' % ('Validation', mean_loss_val))
utils.log_string(logger, '%s best mean loss: %.4f\n' % ('Validation', best_loss_val))
if mean_loss_val <= best_loss_val:
best_loss_val = mean_loss_val
utils.save_model(model, optimizer, epoch, mean_loss_train, mean_loss_val, logger, config, 'loss')
for v, var in enumerate(val_dataset.variables_dynamic):
writer.add_images(var, im_anomaly[0, v, ...], epoch + 1, dataformats='HWC')
writer.add_scalars("Loss", {'train': mean_loss_train, 'val': mean_loss_val}, epoch + 1)
eval_train_anomaly.reset()
eval_val_anomaly.reset()
train_anomaly_collector.reset()
val_anomaly_collector.reset()
# lr_scheduler.step()
lr_scheduler.step_update(epoch)
del mask_extreme_loss, mask_anomaly, z_p, z_n, im_anomaly, im_anomaly_gt, anomaly
if __name__ == '__main__':
train(config_file)