-
Notifications
You must be signed in to change notification settings - Fork 1
/
trainer.py
308 lines (250 loc) · 13.3 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
import os
import math
import multiprocessing
from tqdm import tqdm
from argparse import Namespace
from typing import Iterable, Optional
mp = multiprocessing.get_context("spawn")
from utils import _create_model_training_folder
import torch
import torch.nn.functional as F
import torchvision
from torch.nn.parameter import Parameter
from torch.utils.tensorboard import SummaryWriter
from tspipe import TSPipe
from tspipe.profiler import profile_semantic
from tspipe.dataloader import FastDataLoader, DummyInputGenerator
class BYOLTrainer:
def __init__(self, online_network, target_network, predictor, optimizer, device, scheduler, **params):
self.online_network = online_network
self.target_network = target_network
self.optimizer = optimizer
self.device = device
self.predictor = predictor
self.max_epochs = params['max_epochs']
self.writer = SummaryWriter()
self.m = params['m']
self.batch_size = params['batch_size']
self.num_workers = params['num_workers']
self.checkpoint_interval = params['checkpoint_interval']
self.image_x = eval(params['input_shape'])[0]
self.scheduler = scheduler
_create_model_training_folder(self.writer, files_to_same=["./config/config.yaml", "main.py", 'trainer.py'])
self.dummy_input = True if params['dummy_input'] == True else False
if self.dummy_input:
print("Warning: Dummy Input Enabled.")
@torch.no_grad()
def _update_target_network_parameters(self):
"""
Momentum update of the key encoder
"""
for param_q, param_k in zip(self.online_network.parameters(), self.target_network.parameters()):
param_k.data = param_k.data * self.m + param_q.data * (1. - self.m)
@staticmethod
def regression_loss(x, y):
x = F.normalize(x, dim=1)
y = F.normalize(y, dim=1)
return 2 - 2 * (x * y).sum(dim=-1)
def initializes_target_network(self):
# init momentum network as encoder net
for param_q, param_k in zip(self.online_network.parameters(), self.target_network.parameters()):
param_k.data.copy_(param_q.data) # initialize
param_k.requires_grad = False # not update by gradient
def train(self, train_dataset):
train_loader = FastDataLoader(train_dataset, batch_size=self.batch_size,
num_workers=self.num_workers, drop_last=False, shuffle=True, pin_memory=True)
niter = 0
model_checkpoints_folder = os.path.join(self.writer.log_dir, 'checkpoints')
self.initializes_target_network()
batch_id = 0
for epoch_counter in range(self.max_epochs):
if self.dummy_input:
dummy_input_gen = DummyInputGenerator(self.batch_size, input_shape=self.image_x)
pbar = tqdm(dummy_input_gen)
else:
pbar = tqdm(train_loader)
for (batch_view_1, batch_view_2), _ in pbar:
batch_id += 1
profile_semantic(niter, 0, 0, False, None, 0, 'copy')
batch_view_1 = batch_view_1.to(self.device)
profile_semantic(niter, 0, 0, False, None, 0, 'copy_finish')
profile_semantic(niter, 1, 0, False, None, 0, 'copy')
batch_view_2 = batch_view_2.to(self.device)
profile_semantic(niter, 1, 0, False, None, 0, 'copy_finish')
if niter == 0:
grid = torchvision.utils.make_grid(batch_view_1[:32])
self.writer.add_image('views_1', grid, global_step=niter)
grid = torchvision.utils.make_grid(batch_view_2[:32])
self.writer.add_image('views_2', grid, global_step=niter)
loss = self.update(batch_view_1, batch_view_2, niter)
self.writer.add_scalar('loss', loss, global_step=niter)
# torch.cuda.nvtx.range_push("BackwardCompute")
profile_semantic(niter, 0, 0, False, None, 0, 'backward')
self.optimizer.zero_grad()
loss.backward()
profile_semantic(niter, 0, 0, False, None, 0, 'backward_finish')
profile_semantic(niter, 0, 0, False, None, 0, 'optimize')
self.optimizer.step()
# torch.cuda.nvtx.range_pop()
self._update_target_network_parameters() # update the key encoder
profile_semantic(niter, 0, 0, False, None, 0, 'optimize_finish')
pbar.set_postfix({'loss': loss, 'batch_id': batch_id})
niter += 1
if batch_id % 100 == 0:
self.save_model(os.path.join(model_checkpoints_folder, f'model_batch{batch_id}_part0.pt'))
if batch_id > 1:
loss_fn = torch.nn.MSELoss(reduction='sum')
print("End of epoch {}".format(epoch_counter))
if self.scheduler is not None:
self.scheduler.step()
# save checkpoints
self.save_model(os.path.join(model_checkpoints_folder, 'model.pth'))
def update(self, batch_view_1, batch_view_2, niter = 0):
# compute query feature
profile_semantic(niter, 0, 0, False, None, 0, 'compute')
predictions_from_view_1 = self.predictor(self.online_network(batch_view_1))
profile_semantic(niter, 0, 0, False, None, 0, 'compute_finish')
profile_semantic(niter, 1, 0, False, None, 0, 'compute')
predictions_from_view_2 = self.predictor(self.online_network(batch_view_2))
profile_semantic(niter, 1, 0, False, None, 0, 'compute_finish')
# compute key features
with torch.no_grad():
profile_semantic(niter, 0, 0, True, None, 0, 'compute')
targets_to_view_2 = self.target_network(batch_view_1)
profile_semantic(niter, 0, 0, True, None, 0, 'compute_finish')
profile_semantic(niter, 1, 0, True, None, 0, 'compute')
targets_to_view_1 = self.target_network(batch_view_2)
profile_semantic(niter, 1, 0, True, None, 0, 'compute_finish')
profile_semantic(niter, 0, 0, False, None, 0, 'loss')
loss = self.regression_loss(predictions_from_view_1, targets_to_view_1)
loss += self.regression_loss(predictions_from_view_2, targets_to_view_2)
profile_semantic(niter, 0, 0, False, None, 0, 'loss')
return loss.mean()
def save_model(self, PATH):
torch.save({
'online_network_state_dict': self.online_network.state_dict(),
'target_network_state_dict': self.target_network.state_dict(),
'optimizer_state_dict': self.optimizer.state_dict(),
}, PATH)
class DummyBYOLTrainer(BYOLTrainer):
def train(self, train_dataset):
train_loader = FastDataLoader(train_dataset, batch_size=self.batch_size,
num_workers=self.num_workers, drop_last=False, shuffle=True)
self.initializes_target_network()
for epoch_counter in range(self.max_epochs):
pbar = tqdm(train_loader)
for (batch_view_1, batch_view_2), _ in pbar:
# do nothing
pass
print("End of epoch {}".format(epoch_counter))
class TSPipeBYOLTrainer(BYOLTrainer):
def __init__(self, online_network, target_network, predictor, optimizer: torch.optim.Optimizer, device, scheduler, **params):
super().__init__(online_network, target_network, predictor, optimizer, device, scheduler, **params)
self.optimizer = optimizer
self.online_network = online_network
self.target_network = target_network
self.predictor_network = predictor
self.dummy_input = True if params['dummy_input'] == True else False
self.image_x = eval(params['input_shape'])[0]
self.scheduler = scheduler
self.params = params
if self.dummy_input:
print("Warning: Dummy Input Enabled.")
@staticmethod
def contrastive_loss(online_view_1, online_view_2, target_view_1, target_view_2, args: Namespace, extra_args: Namespace):
loss = TSPipeBYOLTrainer.regression_loss(online_view_1, target_view_2)
loss += TSPipeBYOLTrainer.regression_loss(online_view_2, target_view_1)
return loss.mean()
@staticmethod
def calculate_target_network_parameters(m, online_new_param, target_param:Optional[Iterable[Parameter]] = None):
"""
Momentum update of the key encoder
"""
@torch.no_grad()
def calc():
result = []
for param_q, param_k in zip(online_new_param, target_param):
detached = param_k.clone().detach()
detached = detached * m + param_q.data * (1. - m)
result.append(detached)
return result
return calc()
def train(self, train_dataset):
model_checkpoints_folder = os.path.join(self.writer.log_dir, 'checkpoints')
self.initializes_target_network()
initial_lr = self.optimizer.param_groups[0]['lr']
print(f"initial_lr : {initial_lr}")
initial_momentum = self.params['m']
print(f"initial_momentum : {initial_momentum}")
warmup_epochs = 10
lr = self.adjust_learning_rate(1, warmup_epochs = warmup_epochs, initial_lr = initial_lr)
m = self.adjust_moco_momentum(0, initial_momentum)
self.tspipe = TSPipe(self.online_network,
self.target_network,
self.predictor_network,
self.optimizer,
TSPipeBYOLTrainer.contrastive_loss,
TSPipeBYOLTrainer.calculate_target_network_parameters,
self.m,
model_checkpoints_folder
)
if self.tspipe.is_primary:
# prepare dataloader
if self.dummy_input:
train_loader = DummyInputGenerator(self.batch_size, input_shape=self.image_x)
else:
train_loader = FastDataLoader(train_dataset, batch_size=self.batch_size,
num_workers=self.num_workers, drop_last=False, shuffle=True, pin_memory=False)
iters_per_epoch = len(train_loader)
print(f"iters_per_epoch : {iters_per_epoch}")
niter = 0
for epoch_counter in range(self.max_epochs):
pbar = tqdm(train_loader)
for (batch_view_1, batch_view_2), _ in pbar:
if niter == 0:
grid = torchvision.utils.make_grid(batch_view_1[:32])
self.writer.add_image('views_1', grid, global_step=niter)
grid = torchvision.utils.make_grid(batch_view_2[:32])
self.writer.add_image('views_2', grid, global_step=niter)
loss = self.tspipe.feed(batch_view_1.share_memory_(), batch_view_2.share_memory_())
if loss is not None:
self.writer.add_scalar('loss', loss, global_step=niter)
pbar.set_postfix({'loss': loss, 'batch_id': niter})
niter += 1
print("End of epoch {}".format(epoch_counter))
self.tspipe.feed_epoch()
lr = self.adjust_learning_rate(epoch_counter+1, warmup_epochs = warmup_epochs, initial_lr = initial_lr)
m = self.adjust_moco_momentum(epoch_counter, initial_momentum)
self.tspipe.update_lr(lr)
self.tspipe.update_momentum(m)
self.writer.add_scalar('learning_rate', lr, global_step=niter)
self.writer.add_scalar('momentum', m, global_step=niter)
self.tspipe.stop()
# save checkpoints
print("Saving checkpoints...")
self.save_model(os.path.join(model_checkpoints_folder, 'model.pth'))
print("Saving checkpoints OK")
def adjust_learning_rate(self, epoch, warmup_epochs, initial_lr):
"""Decays the learning rate with half-cycle cosine after warmup"""
if epoch < warmup_epochs:
lr = initial_lr * epoch / warmup_epochs
else:
lr = initial_lr * 0.5 * (1. + math.cos(math.pi * (epoch - warmup_epochs) / (self.params['max_epochs'] - warmup_epochs)))
for param_group in self.optimizer.param_groups:
param_group['lr'] = lr
return lr
def adjust_moco_momentum(self, epoch, initial_momentum):
"""Adjust moco momentum based on current epoch"""
m = 1. - 0.5 * (1. + math.cos(math.pi * epoch / self.params['max_epochs'])) * (1. - initial_momentum)
return m
def update(self, batch_view_1, batch_view_2):
# compute query feature
predictions_from_view_1 = self.predictor(self.online_network(batch_view_1))
predictions_from_view_2 = self.predictor(self.online_network(batch_view_2))
# compute key features
with torch.no_grad():
targets_to_view_2 = self.target_network(batch_view_1)
targets_to_view_1 = self.target_network(batch_view_2)
loss = self.regression_loss(predictions_from_view_1, targets_to_view_1)
loss += self.regression_loss(predictions_from_view_2, targets_to_view_2)
return loss.mean()