-
Notifications
You must be signed in to change notification settings - Fork 2
/
syn_gen_train.py
239 lines (214 loc) · 8.86 KB
/
syn_gen_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
import os
import torch
import torch.utils.data
import torch.optim as optim
import torch.nn as nn
import Core.Constants as Constants
import argparse
from Core.Utils import set_seed_everywhere
from Core.Optim import ScheduledOptim
from SynGen.Dataset import Dataset, collate_fn
from SynGen.Model import MultiEncVAETransformer
from SynGen.TrainFunc import train
def parse_args():
"""
Wrapper function of argument parsing process.
"""
parser = argparse.ArgumentParser()
parser.add_argument(
'--ori_dir', type=str, default=os.path.join(Constants.TRAIN_PATH, 'para_train_ori.pt'),
help='source training data location.'
)
parser.add_argument(
'--ref_dir', type=str, default=os.path.join(Constants.TRAIN_PATH, 'para_train_ref.pt'),
help='reference training data location.'
)
parser.add_argument(
'--dict_dir', type=str, default=os.path.join(Constants.TRAIN_PATH, 'para_train_dict.pt'),
help='token to index dictionary save location.'
)
parser.add_argument(
'--epoch', type=int, default=20, help='number of training epochs'
)
parser.add_argument(
'--lr', type=float, default=2.0, help='learning rate scale factor'
)
parser.add_argument(
'--batch_size', type=int, default=128, help='batch size'
)
parser.add_argument(
'--d_model', type=int, default=64, help='model dimension'
)
parser.add_argument(
'--d_inner', type=int, default=512, help='inner dimension'
)
parser.add_argument(
'--d_k', type=int, default=64, help='dimension of key and query'
)
parser.add_argument(
'--d_v', type=int, default=64, help='dimension of value'
)
parser.add_argument(
'--n_trf_enc_layer', type=int, default=4, help='number of Transformer encoder layers'
)
parser.add_argument(
'--n_trf_dec_layer', type=int, default=6, help='number of Transformer decoder layers'
)
parser.add_argument(
'--n_src_attn_head', type=int, default=4, help='number of source syntax encoder attention heads'
)
parser.add_argument(
'--n_tmpl_attn_head', type=int, default=4, help='number of template encoder attention heads'
)
parser.add_argument(
'--dropout', type=float, default=0.1, help='dropout ratio'
)
parser.add_argument(
'--tgt_emb_prj_weight_sharing', action='store_true',
help='whether share weights between embedding and projection layers'
)
parser.add_argument(
'--log', type=str, default=os.path.join('logs', 'LogFile'), help='log filepath to save'
)
parser.add_argument(
'--model_save', type=str, default=os.path.join('models', 'model'),
help='the path of the model to save'
)
parser.add_argument(
'--no_cuda', action='store_true', help='disable cuda'
)
parser.add_argument(
'--label_smoothing', action='store_true', help='whether use label smoothing'
)
parser.add_argument(
'--n_warmup_steps', type=int, default=12800, help='number of warm-up steps'
)
parser.add_argument(
'--pin_memory', action='store_true', help='whether pin your cuda memory during training'
)
parser.add_argument(
'--random_seed', type=int, default=42
)
args = parser.parse_args()
args.cuda = not args.no_cuda and torch.cuda.is_available()
return args
def main():
""" Main Function"""
args = parse_args()
device = torch.device("cuda" if args.cuda else "cpu")
set_seed_everywhere(args.random_seed, args.cuda)
# ========= Loading Dataset ========= #
data_ori = torch.load(args.ori_dir)
data_ref = torch.load(args.ref_dir)
w2i_dict = torch.load(args.dict_dir)
training_data, validation_data = prepare_dataloaders(data_ori, data_ref, w2i_dict, args)
args.n_syn_token = training_data.dataset.n_syn_token
args.n_lvl_token = training_data.dataset.n_lvl_token
args.max_token_src_len = max(
[len(s) for s in data_ori['train']['src_syntax'] + data_ori['valid']['src_syntax'] +
data_ref['train']['src_syntax'] + data_ref['valid']['src_syntax']]
)
args.max_token_tmpl_len = max(
[len(s) for s in data_ori['train']['tmpl_syntax'] + data_ori['valid']['tmpl_syntax'] +
data_ref['train']['tmpl_syntax'] + data_ref['valid']['tmpl_syntax']]
)
args.max_token_tgt_len = args.max_token_src_len
print(args)
# ========= Preparing Model ========= #
multi_enc_vae_transformer = MultiEncVAETransformer(
n_syn_token=args.n_syn_token,
n_lvl_token=args.n_lvl_token,
max_tmpl_len=args.max_token_tmpl_len,
max_src_len=args.max_token_src_len,
max_tgt_len=args.max_token_tgt_len,
d_model=args.d_model,
d_inner=args.d_inner,
n_trf_enc_layer=args.n_trf_enc_layer,
n_trf_dec_layer=args.n_trf_dec_layer,
n_src_attn_head=args.n_src_attn_head,
n_tmpl_attn_head=args.n_tmpl_attn_head,
d_k=args.d_k,
d_v=args.d_v,
dropout=args.dropout,
tgt_emb_prj_weight_sharing=args.tgt_emb_prj_weight_sharing,
device=device
)
if args.cuda and torch.cuda.device_count() > 1:
multi_enc_vae_transformer = nn.DataParallel(multi_enc_vae_transformer).to(device)
else:
multi_enc_vae_transformer = multi_enc_vae_transformer.to(device)
optimizer = ScheduledOptim(
optim.Adam(
filter(lambda x: x.requires_grad, multi_enc_vae_transformer.parameters()),
betas=(0.9, 0.98), eps=1e-09),
args.lr, args.d_model, args.n_warmup_steps)
train(multi_enc_vae_transformer, training_data, validation_data, optimizer, device, args)
def prepare_dataloaders(data_ori, data_ref, w2i_dict, args):
# ========= Preparing DataLoader ========= #
print("preparing dataloader")
train_src_syn_ori = data_ori['train']['src_syntax']
valid_src_syn_ori = data_ori['valid']['src_syntax']
train_src_lvl_ori = data_ori['train']['src_level']
valid_src_lvl_ori = data_ori['valid']['src_level']
train_src_path_ori = data_ori['train']['src_path']
valid_src_path_ori = data_ori['valid']['src_path']
train_tmpl_syn_ori = data_ori['train']['tmpl_syntax']
valid_tmpl_syn_ori = data_ori['valid']['tmpl_syntax']
train_tmpl_lvl_ori = data_ori['train']['tmpl_level']
valid_tmpl_lvl_ori = data_ori['valid']['tmpl_level']
train_tmpl_path_ori = data_ori['train']['tmpl_path']
valid_tmpl_path_ori = data_ori['valid']['tmpl_path']
train_src_syn_ref = data_ref['train']['src_syntax']
valid_src_syn_ref = data_ref['valid']['src_syntax']
train_src_lvl_ref = data_ref['train']['src_level']
valid_src_lvl_ref = data_ref['valid']['src_level']
train_src_path_ref = data_ref['train']['src_path']
valid_src_path_ref = data_ref['valid']['src_path']
train_tmpl_syn_ref = data_ref['train']['tmpl_syntax']
valid_tmpl_syn_ref = data_ref['valid']['tmpl_syntax']
train_tmpl_lvl_ref = data_ref['train']['tmpl_level']
valid_tmpl_lvl_ref = data_ref['valid']['tmpl_level']
train_tmpl_path_ref = data_ref['train']['tmpl_path']
valid_tmpl_path_ref = data_ref['valid']['tmpl_path']
train_loader = torch.utils.data.DataLoader(
Dataset(
syn_token2idx=w2i_dict['syntax'],
lvl_token2idx=w2i_dict['level'],
src_syn_insts=train_src_syn_ori + train_src_syn_ref,
src_lvl_insts=train_src_lvl_ori + train_src_lvl_ref,
src_path_insts=train_src_path_ori + train_src_path_ref,
tmpl_syn_insts=train_tmpl_syn_ref + train_tmpl_syn_ori,
tmpl_lvl_insts=train_tmpl_lvl_ref + train_tmpl_lvl_ori,
tmpl_path_insts=train_tmpl_path_ref + train_tmpl_path_ori,
tgt_syn_insts=train_src_syn_ref + train_src_syn_ori,
tgt_lvl_insts=train_src_lvl_ref + train_src_lvl_ori
),
num_workers=4,
batch_size=args.batch_size,
collate_fn=collate_fn,
shuffle=True,
pin_memory=args.pin_memory,
drop_last=False
)
valid_loader = torch.utils.data.DataLoader(
Dataset(
syn_token2idx=w2i_dict['syntax'],
lvl_token2idx=w2i_dict['level'],
src_syn_insts=valid_src_syn_ori + valid_src_syn_ref,
src_lvl_insts=valid_src_lvl_ori + valid_src_lvl_ref,
src_path_insts=valid_src_path_ori + valid_src_path_ref,
tmpl_syn_insts=valid_tmpl_syn_ref + valid_tmpl_syn_ori,
tmpl_lvl_insts=valid_tmpl_lvl_ref + valid_tmpl_lvl_ori,
tmpl_path_insts=valid_tmpl_path_ref + valid_tmpl_path_ori,
tgt_syn_insts=valid_src_syn_ref + valid_src_syn_ori,
tgt_lvl_insts=valid_src_lvl_ref + valid_src_lvl_ori
),
num_workers=4,
batch_size=args.batch_size,
collate_fn=collate_fn,
pin_memory=args.pin_memory,
drop_last=False
)
return train_loader, valid_loader
if __name__ == '__main__':
main()