-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrainer.py
199 lines (175 loc) · 9.29 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
import logging
import os
import sys
import torch
from tqdm import tqdm
from torch.utils.data import DataLoader
logging.basicConfig(
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
datefmt="%m/%d/%Y %H:%M:%S",
level=logging.INFO,
)
logger = logging.getLogger(__name__)
class BaseTrainer(object):
def __init__(self, dataset, params, model, device):
self.model = model.to(device)
self.dataset = dataset
self.params = params
self.checkpoint = params.checkpoint
self.device = device
self.optimizer = torch.optim.Adam(filter(lambda p: p.requires_grad, self.model.parameters()), lr=self.params.lr,
weight_decay=self.params.weight_decay)
if self.params.resume_train:
logger.info("===============================")
logger.info("Resuming from a trained {}".format(self.model.name))
self.load()
else:
logger.info("===============================")
logger.info("Starting a new training for {} ...".format(self.model.name))
if not os.path.exists(self.params.cache_dir):
os.makedirs(self.params.cache_dir)
def save(self, epoch):
suffix = "{}_{}".format(epoch, self.params.dataset)
if not os.path.exists(self.params.cache_dir):
os.makedirs(self.params.cache_dir)
name = self.model.save(self.params.cache_dir, suffix)
logger.info("Saving the trained model and optimizer as {} in {} ...".format(name, self.params.cache_dir))
torch.save(self.optimizer.state_dict(), os.path.join(self.params.cache_dir, "optimizer_{}".format(name)))
def load(self):
logger.info("Loading the pretrained model from the checkpoint {} in {}....".format(self.checkpoint, self.params.cache_dir))
self.model.load(self.params.cache_dir, self.checkpoint)
try:
if self.device == torch.device("cpu"):
checkpoint = torch.load(os.path.join(self.params.cache_dir,
"optimizer_{}".format(self.checkpoint)),
map_location=self.device)
else:
checkpoint = torch.load(os.path.join(self.params.cache_dir,
"optimizer_{}".format(self.checkpoint)))
self.optimizer.load_state_dict(checkpoint)
except:
logger.warning("Failing to load pretrained optimizer")
def train(self, tb_writer=None, aimed_types=['sub', 'obj'], tester=None):
train_dataloader = DataLoader(torch.arange(self.dataset.len['train']),
batch_size=self.params.batch_size, shuffle=self.params.train_shuffle)
ttl_cnt = 0
for epoch in tqdm(range(self.params.max_epochs), desc="Epoch"):
self.model.train()
sys.stdout.flush()
ttl_loss = 0
batch_cnt = 0
skipped_data = 0
for batch_idx in tqdm(train_dataloader, desc="Batch", miniters=100, mininterval=60):
for missing in aimed_types:
self.optimizer.zero_grad()
xs, y_true = self.dataset.nextBatch(
batch_idx, type=missing)
if y_true.shape[0] < 1:
skipped_data += batch_idx.shape[0]
logger.warning("No history data. Skip!\nAccumulated Number of Missing Data: {} for types: {}".format(skipped_data, aimed_types))
continue
batch_loss = self.model.loss(xs, y_true)
batch_loss.backward()
loss = batch_loss.detach().cpu().item()
ttl_loss += loss
batch_cnt += 1
self.optimizer.step()
if tb_writer is not None:
tb_writer.add_scalar("train/loss", loss, ttl_cnt + batch_cnt)
if self.params.verbose:
logger.info("Batch {} loss: {}".format(batch_cnt, loss))
logger.info("loss: {}".format(ttl_loss / batch_cnt))
ttl_cnt += batch_cnt
if (epoch + 1) % self.params.save_steps == 0 or (epoch + 1) == self.params.max_epochs:
self.save(epoch + 1)
if self.params.eval_steps < 0:
continue
if tester is None:
logger.warning("Non-given Tester: Trying to evaluate the validation dataset during training but no tester is given!")
if (epoch + 1) % self.params.eval_steps == 0 or (epoch + 1) == self.params.max_epochs:
tester.test(tb_writer, epoch + 1, valid_or_test="valid", aimed_types=aimed_types)
logger.info("Training done")
logger.info("===============================")
class MixTrainer(object):
def __init__(self, dataset, params, model, device):
self.model = model.to(device)
self.dataset = dataset
self.params = params
self.checkpoint = params.checkpoint
self.device = device
self.optimizer = torch.optim.Adam(filter(lambda p: p.requires_grad, self.model.parameters()), lr=self.params.lr,
weight_decay=self.params.weight_decay)
if self.params.resume_train:
logger.info("===============================")
logger.info("Resuming from a trained {}".format(self.model.name))
self.load()
else:
logger.info("===============================")
logger.info("Starting a new training for {} ...".format(self.model.name))
if not os.path.exists(self.params.cache_dir):
os.makedirs(self.params.cache_dir)
def save(self, epoch):
suffix = "{}_{}".format(epoch, self.params.dataset)
if not os.path.exists(self.params.cache_dir):
os.makedirs(self.params.cache_dir)
name = self.model.save(self.params.cache_dir, suffix)
logger.info("Saving the trained model and optimizer as {} in {} ...".format(name, self.params.cache_dir))
torch.save(self.optimizer.state_dict(), os.path.join(self.params.cache_dir, "optimizer_{}".format(name)))
def load(self):
logger.info("Loading the pretrained model from the checkpoint {} in {}....".format(self.checkpoint, self.params.cache_dir))
self.model.load(self.params.cache_dir, self.checkpoint)
try:
if self.device == torch.device("cpu"):
checkpoint = torch.load(os.path.join(self.params.cache_dir,
"optimizer_{}".format(self.checkpoint)),
map_location=self.device)
else:
checkpoint = torch.load(os.path.join(self.params.cache_dir,
"optimizer_{}".format(self.checkpoint)))
self.optimizer.load_state_dict(checkpoint)
except:
logger.warning("Failing to load pretrained optimizer")
def train(self, tb_writer=None, tester=None, early_stop=False):
train_dataloader = DataLoader(torch.arange(self.dataset.len['train']),
batch_size=self.params.batch_size, shuffle=self.params.train_shuffle)
ttl_cnt = 0
for epoch in tqdm(range(self.params.max_epochs), desc="Epoch"):
self.model.train()
sys.stdout.flush()
ttl_loss = 0
batch_cnt = 0
skipped_data = 0
i = 0
for batch_idx in tqdm(train_dataloader, desc="Batch", miniters=100, mininterval=60):
i += 1
if early_stop:
if i>3:
break
self.optimizer.zero_grad()
xs, y_true = self.dataset.nextBatch(batch_idx)
if y_true.shape[0] < 1:
skipped_data += batch_idx.shape[0]
logger.warning("No history data. Skip!\nAccumulated Number of Missing Data: {} for types: {}".format(skipped_data, aimed_types))
continue
batch_loss = self.model.loss(xs, y_true)
batch_loss.backward()
loss = batch_loss.detach().cpu().item()
ttl_loss += loss
batch_cnt += 1
self.optimizer.step()
if tb_writer is not None:
tb_writer.add_scalar("train/loss", loss, ttl_cnt + batch_cnt)
if self.params.verbose:
logger.info("Batch {} loss: {}".format(batch_cnt, loss))
logger.info("loss: {}".format(ttl_loss / batch_cnt))
ttl_cnt += batch_cnt
if (epoch + 1) % self.params.save_steps == 0 or (epoch + 1) == self.params.max_epochs:
self.save(epoch + 1)
if self.params.eval_steps < 0:
continue
if tester is None:
logger.warning("Non-given Tester: Trying to evaluate the validation dataset during training but no tester is given!")
if (epoch + 1) % self.params.eval_steps == 0 or (epoch + 1) == self.params.max_epochs:
tester.test(tb_writer, epoch + 1, valid_or_test="valid")
logger.info("Training done")
logger.info("===============================")