-
Notifications
You must be signed in to change notification settings - Fork 1
/
trainer.py
256 lines (226 loc) · 11.7 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
import os
import pdb
import sys
import logging
from tqdm import tqdm, trange
import numpy as np
import torch
from torch.utils.data import DataLoader
from sklearn import metrics
from transformers import Trainer, TrainingArguments
from transformers.trainer_callback import EarlyStoppingCallback
from data_set import MyDataset
class EarlyStopping:
def __init__(self, patience=10, verbose=False, delta=0):
self.patience = patience
self.verbose = verbose
self.counter = 0
self.best_score = None
self.early_stop = False
self.val_loss_min = np.Inf
self.delta = delta
def __call__(self, val_loss):
score = -val_loss
if self.best_score is None:
self.best_score = score
elif score < self.best_score + self.delta:
self.counter += 1
if self.verbose:
print(f'EarlyStopping counter: {self.counter} out of {self.patience}')
if self.counter >= self.patience:
self.early_stop = True
else:
self.best_score = score
self.counter = 0
def set_logger(args, name, output_dir):
if not os.path.exists(output_dir):
os.makedirs(output_dir)
if args.text_name == "text_json_final":
mmsd2_dir = os.path.join(output_dir, "MMSD2")
if not os.path.exists(mmsd2_dir):
os.makedirs(mmsd2_dir)
log_file_path = os.path.join(mmsd2_dir, "training.log")
elif args.text_name == "text_json_clean":
mmsd_dir = os.path.join(output_dir, "MMSD")
if not os.path.exists(mmsd_dir):
os.makedirs(mmsd_dir)
log_file_path = os.path.join(mmsd_dir, "training.log")
logger = logging.getLogger(name)
logger.setLevel(logging.INFO)
if not logger.handlers:
file_handler = logging.FileHandler(log_file_path, mode='w')
file_handler.setLevel(logging.INFO)
file_formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s',
datefmt='%m/%d/%Y %H:%M:%S')
file_handler.setFormatter(file_formatter)
stream_handler = logging.StreamHandler()
stream_handler.setLevel(logging.INFO)
stream_formatter = logging.Formatter('%(message)s')
stream_handler.setFormatter(stream_formatter)
logger.addHandler(file_handler)
logger.addHandler(stream_handler)
return logger
class MyTrainer():
def __init__(self, args, processor):
self.args = args
self.processor = processor
def training(self, args, model, device, train_data, dev_data, processor):
if not os.path.exists(args.output_dir):
os.mkdir(args.output_dir)
train_loader = DataLoader(dataset=train_data,
batch_size=args.train_batch_size,
collate_fn=MyDataset.collate_fn,
shuffle=True)
total_steps = int(len(train_loader) * args.num_train_epochs)
model.to(device)
from transformers.optimization import get_linear_schedule_with_warmup
clip_params = list(map(id, model.model.parameters()))
base_params = filter(lambda p: id(p) not in clip_params, model.parameters())
optimizer = torch.optim.AdamW([
{"params": base_params},
{"params": model.model.parameters(), "lr": args.clip_learning_rate}
], lr=args.learning_rate, weight_decay=args.weight_decay)
scheduler = get_linear_schedule_with_warmup(optimizer,
num_warmup_steps=int(args.warmup_proportion * total_steps),
num_training_steps=total_steps)
logger = logging.getLogger('my_model')
trainable_params = model.calculate_trainable_params()
logger.info(f"Total trainable parameters: {trainable_params}")
logger.info('------------------ Begin Training! ------------------')
validation_step = args.num_validation_steps
early_stopping = EarlyStopping(patience=args.early_stop, verbose=True)
for i_epoch in trange(0, int(args.num_train_epochs), desc="Epoch", disable=False):
sum_loss = 0.
sum_step = 0
steps_since_validation = 0
iter_bar = tqdm(train_loader, desc="Iter (loss=X.XXX)", disable=False)
model.train()
for step, batch in enumerate(iter_bar):
text_list, image_list, label_list, id_list = batch
inputs = processor(text=text_list, images=image_list, padding='max_length', truncation=True,
max_length=args.max_len, return_tensors="pt").to(device)
labels = torch.tensor(label_list).to(device)
loss, score = model(inputs, labels=labels)
sum_loss += loss.item()
sum_step += 1
steps_since_validation += 1
iter_bar.set_description("Iter (loss=%5.3f)" % loss.item())
loss.backward()
optimizer.step()
if args.optimizer_name == 'adam':
scheduler.step()
optimizer.zero_grad()
if steps_since_validation >= validation_step or step == len(train_loader) - 1:
dev_loss, dev_acc, dev_f1, dev_precision, dev_recall = self.evaluate_acc_f1(args, model,
device, dev_data,
processor, mode='dev')
logger.info(
f"epoch is: {i_epoch + 1},\n"
f"dev loss is: {dev_loss:.4f},\n"
f"dev_acc is: {dev_acc:.4f},\n"
f"dev_f1 is: {dev_f1:.4f},\n"
f"dev_precision is: {dev_precision:.4f},\n"
f"dev_recall is: {dev_recall:.4f}."
)
early_stopping(dev_loss)
if early_stopping.early_stop:
print("Early stopping")
break
steps_since_validation = 0
if early_stopping.early_stop:
logger.info("Early stopping")
break
logger.info('------------------ Train done! ------------------')
if args.text_name == "text_json_final":
model_path_to_save = os.path.join(args.output_dir, "MMSD2")
elif args.text_name == "text_json_clean":
model_path_to_save = os.path.join(args.output_dir, "MMSD")
model_to_save = (model.module if hasattr(model, "module") else model)
torch.save(model_to_save.state_dict(), os.path.join(model_path_to_save, 'model.pt'))
torch.cuda.empty_cache()
def testing(self, args, model, device, test_data, processor):
if args.text_name == "text_json_final":
load_path = os.path.join(args.output_dir, "MMSD2")
elif args.text_name == "text_json_clean":
load_path = os.path.join(args.output_dir, "MMSD")
model_path = os.path.join(load_path, 'model.pt')
if not os.path.exists(model_path):
print(f"Model file not found: {model_path}")
sys.exit(1)
model.load_state_dict(torch.load(model_path))
logger = logging.getLogger('my_model')
logger.info('------------------ Begin Testing! ------------------')
model.eval()
test_loss, test_acc, test_f1, test_precision, test_recall = self.evaluate_acc_f1(args, model,
device, test_data,
processor, mode='test',
macro=True)
_, _, test_f1_, test_precision_, test_recall_ = self.evaluate_acc_f1(args, model, device,
test_data, processor, mode='test')
logger.info(
f"Test done! \n "
f"test_acc: {test_acc:.4f},\n"
f"marco_test_f1: {test_f1:.4f},\n"
f"marco_test_precision: {test_precision:.4f},\n"
f"macro_test_recall: {test_recall:.4f},\n"
f"micro_test_f1: {test_f1_:.4f},\n"
f"micro_test_precision: {test_precision_:.4f},\n"
f"micro_test_recall: {test_recall_:.4f}"
)
def evaluate_acc_f1(self, args, model, device, data, processor, mode, macro=False, pre=None):
data_loader = DataLoader(data, batch_size=args.dev_batch_size, collate_fn=MyDataset.collate_fn, shuffle=False)
n_correct, n_total = 0, 0
t_targets_all, t_outputs_all = None, None
model.eval()
sum_loss = 0.
sum_step = 0
with torch.no_grad():
for i_batch, t_batch in enumerate(tqdm(data_loader, desc="Evaluating")):
text_list, image_list, label_list, id_list = t_batch
inputs = processor(text=text_list, images=image_list, padding='max_length',
truncation=True, max_length=args.max_len, return_tensors="pt").to(device)
labels = torch.tensor(label_list).to(device)
t_targets = labels
loss, t_outputs = model(inputs, labels=labels)
sum_loss += loss.item()
sum_step += 1
outputs = torch.argmax(t_outputs, -1)
n_correct += (outputs == t_targets).sum().item()
n_total += len(outputs)
if t_targets_all is None:
t_targets_all = t_targets
t_outputs_all = outputs
else:
t_targets_all = torch.cat((t_targets_all, t_targets), dim=0)
t_outputs_all = torch.cat((t_outputs_all, outputs), dim=0)
logger = logging.getLogger(__name__)
if mode == 'test':
logger.info("test loss: {:.4f}".format(sum_loss / sum_step))
else:
logger.info("dev loss: {:.4f}".format(sum_loss / sum_step))
final_loss = sum_loss / sum_step
if pre != None:
with open(pre, 'w', encoding='utf-8') as fout:
predict = t_outputs_all.cpu().numpy().tolist()
label = t_targets_all.cpu().numpy().tolist()
for x, y, z in zip(predict, label):
fout.write(str(x) + str(y) + z + '\n')
if not macro:
acc = n_correct / n_total
f1 = metrics.f1_score(t_targets_all.cpu(), t_outputs_all.cpu())
precision = metrics.precision_score(t_targets_all.cpu(), t_outputs_all.cpu())
recall = metrics.recall_score(t_targets_all.cpu(), t_outputs_all.cpu())
else:
acc = n_correct / n_total
f1 = metrics.f1_score(t_targets_all.cpu(), t_outputs_all.cpu(), labels=[0, 1], average='macro')
precision = metrics.precision_score(t_targets_all.cpu(), t_outputs_all.cpu(), labels=[0, 1],
average='macro')
recall = metrics.recall_score(t_targets_all.cpu(), t_outputs_all.cpu(), labels=[0, 1], average='macro')
return final_loss, acc, f1, precision, recall
def train(self, args, model, processor, device, train_data, dev_data, test_data):
if args.train:
self.training(args, model, device, train_data, dev_data, processor)
if args.test:
self.testing(args, model, device, test_data, processor)
if not args.train and not args.test:
print("No action specified. Please use --train and/or --test.")