-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathtrain.py
217 lines (172 loc) · 8.38 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
#!/usr/bin/env python
# encoding: utf-8
import argparse
import torch
import os
from brief.model import DecoderGRU, EncoderGRU, Seq2SeqModel, NSeq2SeqModel, EncoderLSTM, DecoderLSTM
from brief.evaluator import Evaluator
from brief.dataset import BriefSourceField, BriefTargetField, BriefDataset, BriefIterator
from tqdm import tqdm
import dill as pickle
parser = argparse.ArgumentParser(description="Nutshell training")
# parser.add_argument("--model", type=str)
parser.add_argument("--train_filename", type=str)
parser.add_argument("--valid_filename", type=str)
parser.add_argument("--checkpoint_dir", type=str)
parser.add_argument("--epoch", type=int, default=1)
parser.add_argument("--batch_size", type=int, default=4)
args = parser.parse_args()
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
def check_args():
print("=============== Command Line Tools Args ===============")
for arg, value in vars(args).items():
print("{:>20} <===> {:<20}".format(arg, value))
print("=======================================================")
status = True
# if not os.path.exists(os.path.join(args.input_data_dir, args.train_filename)):
# status = False
# print("|ERROR| train file doesn't exist")
#
# if not os.path.exists(os.path.join(args.input_data_dir, args.valid_filename)):
# status = False
# print("|ERROR| valid file doesn't exist")
#
# if torch.cuda.is_available() == False:
# status = False
# print("|ERROR| Currently we dont support CPU training")
#
# if torch.cuda.device_count() == 1 and args.multi_gpu == True:
# status = False
# print("|ERROR| We only detected {} GPU".format(torch.cuda.device_count()))
#
# if os.path.isdir(args.checkpoint_dir) and len(os.listdir(args.checkpoint_dir)) != 0:
# status = False
# # exist but not empty
# print("|ERROR| save dir must be empty")
#
# if not os.path.isdir(args.checkpoint_dir):
# print("|NOTE| Doesn't find the save dir, we will create a default one for you")
# os.mkdir(args.checkpoint_dir)
#
# if not os.path.isdir(args.output_data_dir):
# print("|NOTE| Doesn't find the output data dir, we will create a default one for you")
# os.mkdir(args.output_data_dir)
return status
def preprocess():
print("|LOGGING| Processing tokens and datasets...")
SourceField = BriefSourceField()
TargetField = BriefTargetField()
from itertools import islice
columns = []
with open(args.train_filename) as f_input:
for row in islice(f_input, 0, 1):
for _ in row.strip().split("\t"):
columns.append(_)
tv_datafields = [(columns[0], SourceField), (columns[1], TargetField)]
dataset = BriefDataset(train=args.train_filename,
valid=args.valid_filename,
fields=tv_datafields)
train_data, valid_data = dataset.splits()
SourceField.build_vocab(train_data, min_freq=2)
TargetField.build_vocab(train_data, min_freq=2)
print("Source dataset ---")
# print(SourceField.vocab.itos[:])
# print([(k,v) for k,v in enumerate(SourceField.vocab.itos[:])])
print(len(SourceField.vocab.itos))
print(SourceField.vocab.freqs.most_common(20))
print("Target dataset ---")
# print(TargetField.vocab.itos[:])
# print([(k,v) for k,v in enumerate(TargetField.vocab.itos[:])])
print(TargetField.vocab.freqs.most_common(20))
pickle.dump(SourceField, open(os.path.join(args.checkpoint_dir, "SourceField.p"), "wb"))
pickle.dump(TargetField, open(os.path.join(args.checkpoint_dir, "TargetField.p"), "wb"))
############# pre-process done
return train_data, valid_data, SourceField, TargetField
import torch.optim as optim
def train(train_data, valid_data, SourceField, TargetField):
print("| Building batches...")
train_dataloader, valid_dataloader = BriefIterator.splits(train=train_data,
valid=valid_data,
batch_size=args.batch_size)
print("| Building model...")
encoder_model = EncoderLSTM(vocab_size=len(SourceField.vocab))
decoder_model = DecoderLSTM(vocab_size=len(TargetField.vocab))
seq2seq_model = NSeq2SeqModel(encoder=encoder_model,
decoder=decoder_model)
seq2seq_model.to(device)
optimizer = optim.Adam(seq2seq_model.parameters())
criterion = nn.CrossEntropyLoss(ignore_index=1)
valid_loss_history = {}
print("| Training...")
for epoch in range(1, args.epoch+1):
train_step(seq2seq_model, train_dataloader, optimizer, criterion, epoch)
eval_step(seq2seq_model, valid_dataloader, optimizer, criterion, valid_loss_history, epoch)
test_step(SourceField, TargetField, seq2seq_model)
import torch.nn as nn
def train_step(seq2seq_model, train_dataloader, optimizer, criterion, epoch):
seq2seq_model.train()
evaluator = Evaluator(criterion)
tqdm_progress = tqdm(train_dataloader, desc="| Training epoch {}/{}".format(epoch, args.epoch))
for source_seq, target_seq in tqdm_progress:
# print(source_seq)
# print(target_seq)
optimizer.zero_grad()
output = seq2seq_model(source_seq, target_seq)
# print(output.shape)
# print(target_seq.shape)
loss, ppl = evaluator.evaluate(output, target_seq)
optimizer.step()
tqdm_progress.set_postfix({"Loss": "{:.4f}".format(loss),
"PPL": "{:.4f}".format(ppl)})
def eval_step(seq2seq_model, valid_dataloader, optimizer, criterion, eval_loss_history, epoch):
seq2seq_model.eval()
evaluator = Evaluator(criterion)
tqdm_progress = tqdm(valid_dataloader, desc="| Validating epoch {}/{}".format(epoch, args.epoch))
with torch.no_grad():
for source_seq, target_seq in tqdm_progress:
output = seq2seq_model(source_seq, target_seq, 0)
loss, ppl, rougef1, rougef2 = evaluator.evaluate(output, target_seq, mode="eval")
tqdm_progress.set_postfix({"Loss": "{:.4f}".format(loss),
"PPL": "{:.4f}".format(ppl),
"Rouge-1": "{:.4f}".format(rougef1),
"Rouge-2": "{:.4f}".format(rougef2)})
torch.save({"model_args": seq2seq_model.get_args(),
"model_state_dict": seq2seq_model.state_dict()},
os.path.join(args.checkpoint_dir, "checkpoint_{}.pt".format(epoch)))
TEST_CASE_tgt = ["<sos>"]
aa = """为 期 三 个 月 的 全 国 公 路 执 法 专 项 整 改 工 作 刚 结 束 ，
整 治 重 点 包 括 对 非 法 超 限 运 输 车 辆 只 收 费 不 卸 载 、
伙 同 社 会 闲 散 人 员 擅 自 放 行 等 。 然 而 ， 在 重 要 省 道 滨 唐 公 路 津 冀 交 界 处 ，
执 法 治 超 沦 为 摆 设 ， 大 肆 收 费 后 擅 自 放 行 ， 超 载 问 题 严 重 失 控 。 <eos>"""
### 津 冀 交 界 公 路 治 超 载 乱 象 官 卡 执 法 沦 为 摆 设
a = """i am lazy. <eos>"""
b = """i am cold. <eos>"""
TEST_CASE_src = [aa]
def test_step(SourceField, TargetField, seq2seq_model):
source_vocab = SourceField.vocab.stoi
source_batch_indexed = [[source_vocab[token] for token in sent.split()] for sent in TEST_CASE_src]
source_batch_indexed = torch.LongTensor(source_batch_indexed).to(device)
# print(source_batch_indexed)
target_voab = TargetField.vocab.stoi
target_batch_indexed = [[target_voab[token] for token in sent.split()] for sent in TEST_CASE_tgt]
target_batch_indexed = torch.LongTensor(target_batch_indexed).to(device)
source_seq = source_batch_indexed
target_seq = target_batch_indexed
seq2seq_model.eval()
with torch.no_grad():
output = seq2seq_model.infer(source_seq, target_seq)
# target = target_seq[:, 1:]
# target = target.contiguous().view(-1)
# output = output[:, 1:, :]
# output = output.contiguous().view(-1, output.shape[2])
# topv, topi = output.topk(k=1)
# topi = topi.squeeze(-1)
# print(topi)
# print(topi.shape)
# print("out ---")
# print(output)
words = [TargetField.vocab.itos[word_idx] for word_idx in output]
print("Predicted -> {}".format(" ".join(words)))
if __name__ == "__main__":
tr, vl, src, tgt = preprocess()
train(tr, vl, src, tgt)