-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_scaff.py
210 lines (166 loc) · 6.23 KB
/
train_scaff.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
# %%
import torch
import numpy as np
import json
import datetime
import deepchem as dc
from dataloader import MolDataset
from model import BertClassifier
from sklearn.model_selection import train_test_split
from transformers import AutoTokenizer, AutoModel, TrainingArguments, Trainer, IntervalStrategy
from splitter import random_scaffold_split
# %%
from config import *
cfg = update_cfg(cfg)
def compute_metrics(p):
from sklearn.metrics import accuracy_score, f1_score, roc_auc_score
labels = p.label_ids
pred = p.predictions.argmax(-1)
auc = roc_auc_score(y_true=labels, y_score=pred)
return {'auc': auc}
import logging
def get_logger(log_dir='./trace.log'):
log = logging.getLogger()
console = logging.StreamHandler()
log.addHandler(console)
formatter = logging.Formatter(
fmt='[%(levelname)s] %(asctime)s [%(filename)12s:%(lineno)5d]:\t%(message)s'
)
console.setFormatter(formatter)
log.setLevel(logging.DEBUG)
log.propagate = False
fh = logging.FileHandler(filename=log_dir, encoding='utf-8', mode='a')
fh.setLevel(logging.INFO)
fh.setFormatter(formatter)
log.addHandler(fh)
return log
def set_seed(seed):
# random.seed(seed)
# os.environ['PYTHONHASHSEED'] = str(seed)
np.random.seed(seed)
torch.manual_seed(seed)
# torch.cuda.manual_seed(seed)
# torch.backends.cudnn.deterministic = True
seed = 0
set_seed(seed)
# %%
# -------------- params -----------------
# dataset_name = 'bace'
dataset_name = cfg.dataset
lm_model_name = cfg.lm.model.name
# lm_model_name = 'microsoft/deberta-base'
# lm_model_name = 'mossaic-candle/adaptive-lm-molecules'
# lm_model_name = 'roberta-base'
# lm_model_name = 'bert-base-cased'
# lm_model_name = 'bert-base-uncased'
# lm_model_name = 'prajjwal1/bert-tiny'
lm_model_feat_shrink = ''
use_gpt = cfg.lm.use_gpt
dataset_path = 'dataset/gpt/{}_gpt3_result.json'.format(dataset_name)
# model_name = cfg.lm.model.name
feat_shrink = cfg.lm.model.feat_shrink
weight_decay = cfg.lm.train.weight_decay
dropout = cfg.lm.train.dropout
att_dropout = cfg.lm.train.att_dropout
cla_dropout = cfg.lm.train.cla_dropout
batch_size = cfg.lm.train.batch_size
epochs = cfg.lm.train.epochs
warmup_epochs = cfg.lm.train.warmup_epochs
eval_patience = cfg.lm.train.eval_patience
grad_acc_steps = cfg.lm.train.grad_acc_steps
lr = cfg.lm.train.lr
semi_ratio = cfg.semi_ratio
split_traintest_seed = 0
split_trainval_seed = 0
# dataset
texts, labels = [], []
all_smiles = []
with open(dataset_path, 'r', encoding='UTF-8') as f:
all_infos = json.load(f)
for line in all_infos:
# texts.append(line['data']) # smiles
query = line['question']
# location_tag = 'briefly:'
location_tag = 'Here is the SMILES string: '
smiles_location = query.find(location_tag) + len(location_tag)
smiles = query[smiles_location:]
all_smiles.append(smiles)
if not use_gpt:
texts.append(smiles)
else:
texts.append(line['answer']) # ChatGPT answer
labels.append(int(line['label']))
# print(len(labels), labels)
tokenizer = AutoTokenizer.from_pretrained(lm_model_name)
X = tokenizer(texts, padding=True, truncation=True, max_length=512)
dataset = MolDataset(dataset_path=dataset_path, encodings=X)
# ---------------- logger ------------------
time_prefix = datetime.datetime.strftime(datetime.datetime.now(), '%m-%d-%H-%M')
exp_tag = '{}_epochs={}_semiratio={}'.format(time_prefix, epochs, semi_ratio)
log_dir = 'log_scaff/{}/{}/'.format(dataset_name, lm_model_name.replace('/', '-'))
os.makedirs(log_dir, exist_ok=True)
logger = get_logger(log_dir=log_dir+exp_tag+'.log')
# ---------------- 5-fold scaffold split ------------------
for scaff_seed in [0, 1, 2, 3, 4]:
dc_dataset = dc.data.DiskDataset.from_numpy(X=all_smiles, y=labels, ids=all_smiles)
idx_subtrain, idx_val, idx_test = random_scaffold_split(dc_dataset, smiles_list=all_smiles, seed=scaff_seed)
y_train = [labels[idx] for idx in idx_subtrain]
n_labels = torch.unique(torch.tensor(y_train)).shape[0]
n_dataset = len(labels)
train_dataset = torch.utils.data.Subset(dataset, idx_subtrain)
val_dataset = torch.utils.data.Subset(dataset, idx_val)
test_dataset = torch.utils.data.Subset(dataset, idx_test)
# print(len(train_dataset))
# Define pretrained tokenizer and model
bert_model = AutoModel.from_pretrained(lm_model_name)
cls_model = BertClassifier(bert_model, n_labels=n_labels, feat_shrink=lm_model_feat_shrink)
cls_model.config.dropout = dropout
cls_model.config.attention_dropout = att_dropout
trainable_params = sum(p.numel() for p in cls_model.parameters() if p.requires_grad)
print('Number of parameters: {}M'.format(trainable_params/1e6))
# ------------- train -------------------
# Define training parameters
eq_batch_size = batch_size
train_steps = n_dataset // eq_batch_size + 1
# eval_steps = n_dataset // eq_batch_size
eval_steps = 100000
warmup_steps = int(warmup_epochs * train_steps)
print('eval_steps: {}, warmup_steps: {}', eval_steps, warmup_steps)
# Define Trainer
args = TrainingArguments(
output_dir='/data00/qc/smiles',
# do_train=True,
# do_eval=True,
eval_steps=eval_steps,
evaluation_strategy=IntervalStrategy.STEPS,
# evaluation_strategy="epoch",
save_steps=eval_steps,
learning_rate=lr,
weight_decay=weight_decay,
# save_total_limit=1,
load_best_model_at_end=True,
gradient_accumulation_steps=grad_acc_steps,
per_device_train_batch_size=batch_size,
per_device_eval_batch_size=batch_size*8,
warmup_steps=warmup_steps,
num_train_epochs=epochs,
dataloader_num_workers=1,
fp16=True,
# dataloader_drop_last=True,
metric_for_best_model='auc',
# label_names=["labels"],
)
trainer = Trainer(
model=cls_model,
args=args,
train_dataset=train_dataset,
eval_dataset=val_dataset,
compute_metrics=compute_metrics,
# callbacks=[EarlyStoppingCallback(early_stopping_patience=3)],
)
# Train pre-trained model
trainer.train()
trainer.evaluate()
logger.info('-'*80)
predictions = trainer.predict(test_dataset)
logger.info('predictions: %s', predictions)