-
Notifications
You must be signed in to change notification settings - Fork 4
/
train.py
169 lines (140 loc) · 6.76 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
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Mon Aug 24 21:23:32 2020
@author: sarroutim2
"""
import functools
import itertools
import time
from tools import get_dataset
from tools import create_task
from tools import create_mixture
import logging
import argparse
import torch
import transformers
from models import T5Classifier
from models import BERTClassifier
from tools import MisinfoDataset
from tools import create_data_loader
import numpy as np
import pandas as pd
from transformers import BertModel, BertTokenizer, AdamW, get_linear_schedule_with_warmup
import os
from sklearn.utils import shuffle
def train(args):
if torch.cuda.is_available():
device = torch.device("cuda")
else:
device = torch.device("cpu")
if args.model_type=='t5':
if os.path.exists(args.checkpoint+args.model_spec_b)==False:
os.makedirs(args.checkpoint+args.model_spec_b)
checkpoint=args.checkpoint+args.model_spec+"/"
logging.info('Creating T5 model...')
model = T5Classifier(args.model_spec, checkpoint, device)
##create tasks:
create_task(args.data, 'HealthVer')
#create mixture of tasks:
#mixture=['MEDIQA','RQE']
#create_mixture(mixture,'MEDNLP')
#train
##### total steps
tokenizer = BertTokenizer.from_pretrained(args.model_spec_b)
df = pd.read_csv(args.data+"healthver_train.csv")
train_data_loader = create_data_loader(df, tokenizer, args.sequence_length_inputs, args.batch_size)
total_steps = len(train_data_loader) * args.epochs
print(total_steps)
#####
model.train(
mixture_or_task_name="HealthVer",
steps=total_steps,
save_steps=args.save_steps,
sequence_length={"inputs": args.sequence_length_inputs, "targets": args.sequence_length_targets_t5},
split="train",
batch_size=args.batch_size,
optimizer=functools.partial(transformers.AdamW, lr=1e-4),)
elif args.model_type=='bert':
if os.path.exists(args.checkpoint+args.model_spec_b)==False:
os.makedirs(args.checkpoint+args.model_spec_b)
checkpoint=args.checkpoint+args.model_spec_b+"/"
#WebVer_train_emnlp
df = pd.read_csv(args.data+"healthver_train.csv")
#df = df [:4000]
#df = shuffle(df)
#df.to_csv(args.data+"train_shuffle.csv")
df_dev = pd.read_csv(args.data+"healthver_dev.csv")
logging.info('Creating BERT model...')
model = BERTClassifier(args.model_spec_b, checkpoint, device)
tokenizer = BertTokenizer.from_pretrained(args.model_spec_b)
'''
encoding = tokenizer.encode_plus(
'text text',
'text text',
add_special_tokens=True,
max_length=10,
return_token_type_ids=False,
pad_to_max_length=True,
return_attention_mask=True,
#return_tensors='pt',
)
print("Multi segment token (str): {}".format(tokenizer.convert_ids_to_tokens(encoding['input_ids'])))
print("Multi segment token (str): {}".format(encoding['input_ids']))'''
train_data_loader = create_data_loader(df, tokenizer, args.sequence_length_inputs, args.batch_size)
val_data_loader = create_data_loader(df_dev, tokenizer, args.sequence_length_inputs, args.batch_size)
total_steps = len(train_data_loader) * args.epochs
optimizer = AdamW(model.parameters(), lr=2e-5, correct_bias=True)
model.train(
train_data_loader,
val_data_loader,
optimizer = optimizer,
scheduler = get_linear_schedule_with_warmup(
optimizer,
num_warmup_steps=0,
num_training_steps=total_steps
),
epochs=args.epochs,
checkpoint=checkpoint,
df_train = len(df),
df_eval =len (df_dev)
)
else:
raise ValueError("model_type should be either T5 or BERT")
if __name__== '__main__':
parser = argparse.ArgumentParser()
# Session parameters.
parser.add_argument('--model-type', type=str, default='t5',
help='model type: bert or T5')
parser.add_argument('--data', type=str, default='./data/',
help='data for each task')
parser.add_argument('--checkpoint', type=str, default='./checkpoints-healthver/',
help='Path for saving trained models')
parser.add_argument('--model-spec', type=str, default='t5-base',
help='--model-spec: A str to pass into the pretrained_model_name_or_path'
'argument of `transformers.T5ForConditionalGeneration.from_pretrained'
'(e.g. `"t5-base"` "t5_3b_covid" or a path to a previously trained model) or an'
'instance of the `transformers.configuration_t5.T5Config` class to use'
'to directly construct the `transformers.T5ForConditionalGeneration object.')
parser.add_argument('--model-spec-b', type=str, default='bert-base-uncased',
help='--model-spec: A str to pass into the pretrained_model_name_or_path'
'e.g. lordtt13/COVID-SciBERT, allenai/scibert_scivocab_uncased'
'mrm8488/scibert_scivocab-finetuned-CORD19'
'monologg/biobert_v1.0_pubmed_pmc'
'monologg/biobert_v1.1_pubmed'
'lordtt13/COVID-SciBERT'
'bert-base-uncased, bert-large-uncased, distilbert-base-cased'
'(e.g. `"bert-base-cased, scibert_scivocab_uncased"` ')
parser.add_argument('--epochs', type=int, default=20,
help='Step size for saving trained models')
parser.add_argument('--steps', type=int, default=100,
help='Step size for saving trained models')
parser.add_argument('--save-steps', type=int, default=2000,
help='Step size for saving trained models')
parser.add_argument('--batch-size', type=int, default=8,
help='16 for bert, 8 for t5')
parser.add_argument('--sequence-length-inputs', type=int, default=300)#400
parser.add_argument('--sequence-length-targets', type=int, default=3)
parser.add_argument('--sequence-length-targets-t5', type=int, default=5)
args = parser.parse_args()
train(args)