-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrain.py
192 lines (170 loc) · 7.52 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
import os
import logging
import argparse
import torch
import pickle
# import json
import datetime
from tqdm import tqdm
from models import setup_model
from data.dataset import MultiRMCFDataset,SingleRMCFDataset
# from data.dataloader import GraphVAECollater
from trainer.training_args import TrainingArguments
from trainer.trainer import Trainer
from random import seed
from transformers.trainer_utils import get_last_checkpoint
from callback import RMCFCallback
from segmentor.frag_seg import FragSeg
from data.vocabulary import FragVocab
from processor.frag_processor import FragProcessor
from torch_geometric.loader.dataloader import Collater
logging.basicConfig(format='%(asctime)s - %(pathname)s[line:%(lineno)d] - %(levelname)s: %(message)s',
level=logging.DEBUG)
def fix_seed():
torch.manual_seed(1)
torch.backends.cudnn.enabled = False
torch.backends.cudnn.benchmark = False
torch.backends.cudnn.deterministic = True
seed(43)
def load_dataset(args,vocab,split="train"):
if split == "train":
load_path=args.train_prefix
elif split == "valid":
load_path=args.valid_prefix
elif split == "test":
load_path=args.test_prefix
else:
raise ValueError("Split mush be train, valid or test.")
if split=="train":
train_slice = args.local_rank if not args.debug else 0
dataset = MultiRMCFDataset( num_workers = args.num_workers,
load_path = f'{load_path}{train_slice}.json',
vocab = vocab,
angle_inv=args.angle_intervals
)
else:
dataset = SingleRMCFDataset(load_path = f'{load_path}.json',
vocab = vocab,
angle_inv=args.angle_intervals
)
return dataset
def main(args):
# torch.distributed.init_process_group(backend='nccl', init_method='env://',timeout=datetime.timedelta(seconds=36000))
print('----------------------------------------------------------------')
fix_seed()
# os.environ["CUDA_LAUNCH_BLOCKING"]='1'
args.local_rank = int(os.environ["LOCAL_RANK"]) if not args.debug else -1
print(args.local_rank)
seg = FragSeg(hit_vocab=pickle.load(open(f'{args.seg_vocab_path}','rb')))
vocab = FragVocab(vocab_file=pickle.load(open(f'{args.vocab_path}','rb')))
processor = FragProcessor(segmentor= seg, vocab =vocab )
args.vocab_2d_size = vocab.get_2d_vocab_size()
args.vocab_3d_size = vocab.get_3d_vocab_size()
args.iface_size = vocab.get_iface_size()
train_dataset,valid_dataset = load_dataset(args,vocab,split="train"), load_dataset(args,vocab,split="valid")
model = setup_model(args,processor = processor)
if args.debug:
train_args = TrainingArguments(
output_dir=args.model_dir,
overwrite_output_dir = True,
do_train=True,
do_eval=False,
report_to="none",
ignore_data_skip=True,
# run_name=run_name,
max_grad_norm=1,
max_steps=args.max_steps,
per_device_train_batch_size = args.batch_size,
per_device_eval_batch_size = args.batch_size,
gradient_accumulation_steps=args.update_freq,
learning_rate=args.learning_rate,
num_train_epochs=1,
logging_strategy="steps",
logging_steps=args.logging_steps,
save_strategy="steps",
save_steps=args.save_steps,
evaluation_strategy="no",
save_total_limit=args.save_total_limit,
fp16=args.fp16,
dataloader_num_workers=1,
remove_unused_columns=True
)
else:
train_args =TrainingArguments(
output_dir=args.model_dir,
overwrite_output_dir = True,
do_train=True,
do_eval=False,
report_to="none",
# run_name=run_name,
max_grad_norm=1,
ignore_data_skip=True,
max_steps=args.max_steps,
per_device_train_batch_size = args.batch_size,
per_device_eval_batch_size = args.batch_size,
gradient_accumulation_steps=args.update_freq,
learning_rate=args.learning_rate,
num_train_epochs=1,
logging_strategy="steps",
logging_steps=args.logging_steps,
save_strategy="steps",
save_steps=args.save_steps,
evaluation_strategy="no",
save_total_limit=args.save_total_limit,
fp16=args.fp16,
sharded_ddp= "simple" ,
ddp_find_unused_parameters=True,
local_rank=args.local_rank,
dataloader_num_workers=args.num_workers,
remove_unused_columns=True
)
trainer = Trainer(
model=model,
args=train_args,
train_dataset=train_dataset,
eval_dataset=valid_dataset,
data_collator=Collater(None,None),
)
if args.debug:
args.eval_steps = 60000
eval_callback = RMCFCallback(trainer=trainer,
config=args,
eval_rank=0 if not args.debug else -1
)
trainer.add_callback(eval_callback)
last_checkpoint = get_last_checkpoint(args.model_dir)
trainer.train(resume_from_checkpoint=last_checkpoint)
# trainer.save_model()
if __name__ == '__main__':
# torch.multiprocessing.set_start_method('spawn', force=True)
# TODO: 对args分组！
parser = argparse.ArgumentParser()
parser.add_argument('--train-prefix', type=str)
parser.add_argument('--valid-prefix', type=str)
parser.add_argument('--test-prefix', type=str)
parser.add_argument('--model-dir', type=str)
parser.add_argument('--max-steps', type=int, default=360000)
parser.add_argument('--batch-size', type=int, default=1280)
parser.add_argument('--learning-rate', type=float, default=3e-4)
parser.add_argument('--fp16', action='store_true')
parser.add_argument('--logging-steps', type=int, default=500)
parser.add_argument('--eval-steps', type=int, default=2500,)
parser.add_argument('--save-steps', type=int, default=10000)
parser.add_argument('--save-total-limit', type=int, default=50)
parser.add_argument('--num-workers', type=int, default=4)
parser.add_argument('--update-freq', type=int, default=1, help='update')
parser.add_argument('--local_rank', type=int, default=-1, help='local rank')
parser.add_argument('--model-name', type=str, default='rmcf')
parser.add_argument('--dim-h', type=int, default=512, help='dimension of the hidden')
parser.add_argument('--dim-node', type=int, default=512, help='dimension of the nodes')
parser.add_argument('--dim-edge', type=int, default=512, help='dimension of the edges')
parser.add_argument('--mpnn-steps', type=int, default=3, help='number of mpnn steps')
parser.add_argument('--num-attn-heads', type=int, default=8)
parser.add_argument('--angle-intervals', type=float, default=5.0)
parser.add_argument('--sampling-strategy', type=str, default='random',choices=['random', 'clustering'])
parser.add_argument('--cov-thres', type=float, default=1.25)
parser.add_argument('--vocab-path', type=str,default='geom-drugs/vocab.pkl')
parser.add_argument('--seg-vocab-path', type=str,default='geom-drugs/hit.pkl')
parser.add_argument('--debug', action='store_true')
args = parser.parse_args()
main(args)