-
Notifications
You must be signed in to change notification settings - Fork 1
/
train.py
276 lines (210 loc) · 10.3 KB
/
train.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
import argparse
import logging
import time
import datetime
import random
import pickle
import json
from pathlib import Path
from collections import namedtuple
from typing import List, Tuple
from utils import read_text, add_start_end_tokens, batch_iter
from vocab import Vocab, Vocabularies
from models import Seq2Seq
from tqdm import tqdm
import numpy as np
import torch
import torch.optim as optim
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__)
EXPERIMENTS_DIR = Path("experiments")
def print_random_samples(data: List[Tuple[str, str]], n: int=5):
"""
Print the randomly selected samples from the given dataset.
"""
indices = random.sample(range(0, len(data)), k=n)
for idx in indices:
tr_sent, en_sent = data[idx]
print("TR: ", tr_sent)
print("EN: ", en_sent)
print("="*50)
def evaluate_ppl(model, valid_data, batch_size=32):
"""
Evaluate the perplexity on valid sentences
model: Seq2Seq Model
valid_data: list of tuples containing source and target sentence
batch_size: batch size
"""
was_training = model.training
model.eval()
cum_loss = 0.
cum_tgt_words = 0.
# no_grad() signals backend to throw away all gradients
with torch.no_grad():
for src_sents, tgt_sents in batch_iter(valid_data, batch_size):
loss = -model(src_sents, tgt_sents).sum()
cum_loss += float(loss)
tgt_word_num_to_predict = sum(len(s[1:]) for s in tgt_sents)
cum_tgt_words += tgt_word_num_to_predict
ppl = np.exp(cum_loss / cum_tgt_words)
if was_training:
model.train()
return ppl
def train(model, optimizer, train_data, valid_data, args):
batch_size = args["batch_size"]
n_epochs = args["n_epochs"]
clip_grad = args["clip_grad"]
model_save_path = args["model_save_path"]
lr_decay = args["lr_decay"]
epoch_patience = args["patience"]
max_trial = args["max_trial"]
device = args["device"]
hist_valid_scores = []
epoch_counter = 0
num_trial = 0
while epoch_counter < n_epochs:
model.train()
total_train_loss = 0.0
epoch_start_time = time.time()
total_train_iteration = len(train_data) // batch_size
for src_sents, tgt_sents in tqdm(batch_iter(train_data, batch_size=batch_size, shuffle=True), total=total_train_iteration):
optimizer.zero_grad()
loss = (-model(src_sents, tgt_sents).sum()) / batch_size
loss.backward()
total_train_loss += float(loss)
torch.nn.utils.clip_grad_norm_(model.parameters(), clip_grad)
optimizer.step()
epoch_time = str(datetime.timedelta(seconds=round(time.time() - epoch_start_time)))
logger.info("Epoch {} done in: {}".format(epoch_counter+1, epoch_time))
valid_start_time = time.time()
model.eval()
with torch.no_grad():
# compute validation perplexity
valid_ppl = evaluate_ppl(model, valid_data, batch_size=128)
valid_metric = -valid_ppl
is_better = len(hist_valid_scores) == 0 or valid_metric > max(hist_valid_scores)
hist_valid_scores.append(valid_metric)
if is_better:
patience = 0
logger.info("Saving the model...")
model.save(model_save_path)
torch.save(optimizer.state_dict(), model_save_path + '.optim')
elif patience < epoch_patience:
patience += 1
if patience == epoch_patience:
num_trial += 1
if num_trial == max_trial:
logger.info("Early Stopping hit.")
break
# Decaying the learning rate.
lr = optimizer.param_groups[0]['lr'] * lr_decay
logger.info("Loading the previous best model. Decayed the learning rate to: {}".format(lr))
# Loading the previous best model.
params = torch.load(model_save_path, map_location=lambda storage, loc: storage)
model.load_state_dict(params["state_dict"])
model.to(device)
optimizer.load_state_dict(torch.load(model_save_path + '.optim'))
for param_group in optimizer.param_groups:
param_group["lr"] = lr
patience = 0
train_loss = total_train_loss / total_train_iteration
valid_ppl = hist_valid_scores[-1]
logger.info("Epoch: {:02d}/{} Loss: {:.4f} Valid_Ppl: {:.4f}".format(epoch_counter+1,
n_epochs,
train_loss,
valid_ppl))
valid_end_time = str(datetime.timedelta(seconds=round(time.time() - valid_start_time)))
logger.info("Validation done in: {}\n".format(valid_end_time))
epoch_counter += 1
def main():
arg_parser = argparse.ArgumentParser(description="Neural Machine Translation Training")
arg_parser.add_argument("--train_data", required=True, nargs="+",
help="Parallel training data")
arg_parser.add_argument("--valid_data", required=True, nargs="+",
help="Parallel validation data")
arg_parser.add_argument("--n_epochs", type=int, default=30, help="")
arg_parser.add_argument("--batch_size", type=int, default=32, help="")
arg_parser.add_argument("--embedding_dim", type=int, default=64,
help="Embedding dimension for the word embeddings")
arg_parser.add_argument("--hidden_size", type=int, default=64,
help="RNN hidden dimension")
arg_parser.add_argument("--num_layers", type=int, default=1,
help="Number of RNN Layers")
arg_parser.add_argument("--bidirectional", action="store_true",
help="Whether or not bidirectional RNNs")
arg_parser.add_argument("--dropout_p", type=float, default=0.1,
help="Dropout probability for word embeddings and Decoder networks output")
arg_parser.add_argument("--initial_lr", type=float, default=0.001,
help="Initial learning rate for the optimizer")
arg_parser.add_argument("--uniform_init", type=float, default=0.0,
help="Uniformly initialization of the model's parameter")
arg_parser.add_argument("--clip_grad", type=float, default=5.0,
help="Gradient clipping value to be applied to the model")
arg_parser.add_argument("--lr_decay", type=float, default=0.5,
help="Learning rate decay if the validation metric doesn't improve")
arg_parser.add_argument("--patience", type=int, default=5,
help="Learning rate decay patience")
arg_parser.add_argument("--max_trial", type=int, default=5,
help="Maximum number of trials for early stopping")
arg_parser.add_argument("--device", type=str, default="cpu",
help="Device to run the model")
arg_parser.add_argument("--model_name", type=str, default="model.bin",
help="Model name")
args = arg_parser.parse_args()
args = vars(args)
print(args)
device = "cuda" if args["device"] == "cuda" else "cpu"
if not torch.cuda.is_available() and args["device"] == "cuda":
logger.info("Device is specified as cuda. But there is no cuda device available in your system.")
exit(0)
if not EXPERIMENTS_DIR.exists():
EXPERIMENTS_DIR.mkdir()
current_experiment_name = datetime.datetime.now().strftime("%m_%d_%Y_%H_%M_%S")
current_experiment_dir = Path(EXPERIMENTS_DIR / current_experiment_name)
# Create a new experiment directory based on timestamp.
current_experiment_dir.mkdir()
# Save the command line arguments to the file.
with open(current_experiment_dir / "params.json", "w") as f:
json.dump(args, f, indent=4)
tr_train_dataset_fn, en_train_dataset_fn = args["train_data"]
tr_valid_dataset_fn, en_valid_dataset_fn = args["valid_data"]
tr_train_data = read_text(tr_train_dataset_fn)
en_train_data = read_text(en_train_dataset_fn)
tr_valid_data = read_text(tr_valid_dataset_fn)
en_valid_data = read_text(en_valid_dataset_fn)
logger.info("Total train sentences: {}".format(len(tr_train_data)))
logger.info("Total valid sentences: {}".format(len(tr_valid_data)))
train_data = list(zip(tr_train_data, en_train_data))
valid_data = list(zip(tr_valid_data, en_valid_data))
logger.info("Random samples from training data")
print_random_samples(train_data, n=3)
logger.info("Random samples from validation data")
print_random_samples(valid_data, n=3)
src_train, tgt_train = add_start_end_tokens(train_data)
src_valid, tgt_valid = add_start_end_tokens(valid_data)
train_data = list(zip(src_train, tgt_train))
valid_data = list(zip(src_valid, tgt_valid))
src_vocab = Vocab(src_train)
tgt_vocab = Vocab(tgt_train)
vocabs = Vocabularies(src_vocab, tgt_vocab)
with open(current_experiment_dir / "vocabs.pkl", "wb") as f:
pickle.dump(vocabs, f)
logger.info("Total words in the source language: {}".format(len(src_vocab)))
logger.info("Total words in the target language: {}".format(len(tgt_vocab)))
model = Seq2Seq(vocabs=vocabs, embedding_dim=args["embedding_dim"], hidden_size=args["hidden_size"],
num_layers=args["num_layers"], bidirectional=args["bidirectional"],
dropout_p=args["dropout_p"], device=device)
model.to(device)
print(model)
if args["uniform_init"] > 0:
for p in model.parameters():
p.data.uniform_(-args["uniform_init"], args["uniform_init"])
optimizer = optim.Adam(model.parameters(), lr=args["initial_lr"])
args["model_save_path"] = str(current_experiment_dir / args["model_name"])
train(model, optimizer, train_data, valid_data, args)
if __name__ == "__main__":
main()