Repository navigation
Expand file tree
/
Copy pathtrain.py
More file actions
300 lines (261 loc) Β· 10.9 KB
/
Copy pathtrain.py
File metadata and controls
300 lines (261 loc) Β· 10.9 KB
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
# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import argparse
import os
import random
import time
from functools import partial
import numpy as np
import paddle
from criterion import ParserCriterion
from data import build_vocab, convert_example, create_dataloader
from metric import ParserEvaluator
from model.dep import BiAffineParser
from utils import decode, flat_words
from paddlenlp.datasets import load_dataset
from paddlenlp.transformers import AutoModel, AutoTokenizer
from paddlenlp.transformers.optimization import LinearDecayWithWarmup
# yapf: disable
parser = argparse.ArgumentParser()
# Train
parser.add_argument("--device", choices=["cpu", "gpu"], default="gpu", help="Select which device to train model, defaults to gpu.")
parser.add_argument("--task_name", choices=["nlpcc13_evsam05_thu", "nlpcc13_evsam05_hit"], type=str, default="nlpcc13_evsam05_thu", help="Select the task.")
parser.add_argument("--encoding_model", choices=["lstm", "lstm-pe", "ernie-3.0-medium-zh", "ernie-1.0", "ernie-tiny", "ernie-gram-zh"], type=str, default="ernie-3.0-medium-zh", help="Select the encoding model.")
parser.add_argument("--epochs", type=int, default=100, help="Number of epoches for training.")
parser.add_argument("--save_dir", type=str, default='model_file/', help="Directory to save model parameters.")
parser.add_argument("--batch_size", type=int, default=1000, help="Numbers of examples a batch for training.")
parser.add_argument("--init_from_params", type=str, default=None, help="The path of model parameters to be loaded.")
parser.add_argument("--clip", type=float, default=1.0, help="The threshold of gradient clip.")
parser.add_argument("--lstm_lr", type=float, default=0.002, help="The Learning rate of lstm encoding model.")
parser.add_argument("--ernie_lr", type=float, default=5e-05, help="The Learning rate of ernie encoding model.")
parser.add_argument("--seed", type=int, default=1000, help="Random seed for initialization.")
# Preprocess
parser.add_argument("--n_buckets", type=int, default=15, help="Number of buckets to devide the dataset.")
# Postprocess
parser.add_argument("--tree", type=bool, default=True, help="Ensure the output conforms to the tree structure.")
# Lstm
parser.add_argument("--feat", choices=["char", "pos"], type=str, default=None, help="The feature representation to use.")
# Ernie
parser.add_argument("--warmup_proportion", type=float, default=0.0, help="Linear warmup proportion over total steps.")
parser.add_argument("--weight_decay", type=float, default=0.0, help="Weight decay if we apply some.")
args = parser.parse_args()
# yapf: enable
def set_seed(seed):
random.seed(seed)
np.random.seed(seed)
paddle.seed(seed)
@paddle.no_grad()
def batch_evaluate(
model,
metric,
criterion,
data_loader,
word_pad_index,
word_bos_index,
word_eos_index,
):
model.eval()
metric.reset()
losses = []
for batch in data_loader():
if args.encoding_model.startswith("ernie") or args.encoding_model == "lstm-pe":
words, arcs, rels = batch
words, feats = flat_words(words)
s_arc, s_rel, words = model(words, feats)
else:
words, feats, arcs, rels = batch
s_arc, s_rel, words = model(words, feats)
mask = paddle.logical_and(
paddle.logical_and(words != word_pad_index, words != word_bos_index),
words != word_eos_index,
)
loss = criterion(s_arc, s_rel, arcs, rels, mask)
losses.append(loss.numpy().item())
arc_preds, rel_preds = decode(s_arc, s_rel, mask)
metric.update(arc_preds, rel_preds, arcs, rels, mask)
uas, las = metric.accumulate()
total_loss = np.mean(losses)
model.train()
metric.reset()
return total_loss, uas, las
def do_train(args):
set_seed(args.seed)
paddle.set_device(args.device)
rank = paddle.distributed.get_rank()
if paddle.distributed.get_world_size() > 1:
paddle.distributed.init_parallel_env()
if args.encoding_model == "ernie-gram-zh":
tokenizer = AutoTokenizer.from_pretrained(args.encoding_model)
elif args.encoding_model.startswith("ernie"):
tokenizer = AutoTokenizer.from_pretrained(args.encoding_model)
elif args.encoding_model == "lstm-pe":
tokenizer = AutoTokenizer.from_pretrained("ernie-3.0-medium-zh")
else:
tokenizer = None
train_ds, dev_ds = load_dataset(args.task_name, splits=["train", "dev"])
# Build the vocabs based on train corpus
word_examples = [seq["FORM"] for seq in train_ds]
if args.feat == "pos":
feat_examples = [seq["CPOS"] for seq in train_ds]
elif args.feat == "char":
feat_examples = [token for seq in train_ds for token in seq["FORM"]]
else:
feat_examples = None
rel_examples = [seq["DEPREL"] for seq in train_ds]
train_corpus = [word_examples, feat_examples, rel_examples]
vocabs = build_vocab(
train_corpus,
tokenizer,
encoding_model=args.encoding_model,
feat=args.feat,
)
word_vocab, feat_vocab, rel_vocab = vocabs
if not os.path.exists(args.save_dir):
os.makedirs(args.save_dir)
# Save vocabs into json file
word_vocab.to_json(path=os.path.join(args.save_dir, "word_vocab.json"))
rel_vocab.to_json(path=os.path.join(args.save_dir, "rel_vocab.json"))
if feat_vocab:
n_feats = len(feat_vocab)
feat_vocab.to_json(path=os.path.join(args.save_dir, "feat_vocab.json"))
word_pad_index = word_vocab.to_indices("[PAD]")
word_bos_index = word_vocab.to_indices("[BOS]")
word_eos_index = word_vocab.to_indices("[EOS]")
else:
n_feats = None
word_pad_index = word_vocab.to_indices("[PAD]")
word_bos_index = word_vocab.to_indices("[CLS]")
word_eos_index = word_vocab.to_indices("[SEP]")
n_rels, n_words = len(rel_vocab), len(word_vocab)
trans_fn = partial(
convert_example,
vocabs=vocabs,
encoding_model=args.encoding_model,
feat=args.feat,
)
train_data_loader, _ = create_dataloader(
train_ds,
batch_size=args.batch_size,
mode="train",
n_buckets=args.n_buckets,
trans_fn=trans_fn,
)
dev_data_loader, _ = create_dataloader(
dev_ds,
batch_size=args.batch_size,
mode="dev",
n_buckets=args.n_buckets,
trans_fn=trans_fn,
)
# Load pretrained model if encoding model is ernie-3.0-medium-zh, ernie-1.0, ernie-tiny or ernie-gram-zh
if args.encoding_model in ["ernie-3.0-medium-zh", "ernie-1.0", "ernie-tiny", "ernie-gram-zh"]:
pretrained_model = AutoModel.from_pretrained(args.encoding_model)
else:
pretrained_model = None
# Load ddparser model
model = BiAffineParser(
encoding_model=args.encoding_model,
feat=args.feat,
n_rels=n_rels,
n_feats=n_feats,
n_words=n_words,
pad_index=word_pad_index,
eos_index=word_eos_index,
pretrained_model=pretrained_model,
)
# Define learning rate
if args.encoding_model.startswith("ernie"):
lr = args.ernie_lr
else:
lr = args.lstm_lr
# Continue training from a pretrained model if the checkpoint is specified
if args.init_from_params and os.path.isfile(args.init_from_params):
state_dict = paddle.load(args.init_from_params)
model.set_dict(state_dict)
# Data parallel for distributed training
model = paddle.DataParallel(model)
num_training_steps = len(list(train_data_loader)) * args.epochs
# Define the training strategy
lr_scheduler = LinearDecayWithWarmup(lr, num_training_steps, args.warmup_proportion)
grad_clip = paddle.nn.ClipGradByGlobalNorm(clip_norm=args.clip)
if args.encoding_model.startswith("ernie"):
optimizer = paddle.optimizer.AdamW(
learning_rate=lr_scheduler,
parameters=model.parameters(),
weight_decay=args.weight_decay,
grad_clip=grad_clip,
)
else:
optimizer = paddle.optimizer.Adam(
learning_rate=lr,
beta1=0.9,
beta2=0.9,
epsilon=1e-12,
parameters=model.parameters(),
grad_clip=grad_clip,
)
# Load metric and criterion
best_las = 0
metric = ParserEvaluator()
criterion = ParserCriterion()
# Epoch train
global_step = 0
tic_train = time.time()
for epoch in range(1, args.epochs + 1):
for inputs in train_data_loader():
if args.encoding_model.startswith("ernie") or args.encoding_model == "lstm-pe":
words, arcs, rels = inputs
words, feats = flat_words(words)
s_arc, s_rel, words = model(words, feats)
else:
words, feats, arcs, rels = inputs
s_arc, s_rel, words = model(words, feats)
mask = paddle.logical_and(
paddle.logical_and(words != word_pad_index, words != word_bos_index),
words != word_eos_index,
)
loss = criterion(s_arc, s_rel, arcs, rels, mask)
loss.backward()
optimizer.step()
lr_scheduler.step()
optimizer.clear_grad()
global_step += 1
if global_step % 10 == 0 and rank == 0:
print(
"global step %d, epoch: %d, loss: %.5f, speed: %.2f step/s"
% (global_step, epoch, loss.numpy().item(), 10 / (time.time() - tic_train))
)
tic_train = time.time()
if rank == 0:
# Evaluate on dev dataset
loss, uas, las = batch_evaluate(
model,
metric,
criterion,
dev_data_loader,
word_pad_index,
word_bos_index,
word_eos_index,
)
print("eval loss: %.5f, UAS: %.2f%%, LAS: %.2f%%" % (loss, uas * 100, las * 100))
# Save model parameter of last epoch
save_param_path = os.path.join(args.save_dir, "last_epoch.pdparams")
paddle.save(model.state_dict(), save_param_path)
# Save the model if it get a higher score of las
if las > best_las:
save_param_path = os.path.join(args.save_dir, "best.pdparams")
paddle.save(model.state_dict(), save_param_path)
best_las = las
if __name__ == "__main__":
do_train(args)