-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathesm_stapler.py
130 lines (106 loc) · 4.77 KB
/
esm_stapler.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
"""Retrain ESM-2 on Olga's data.
"""
import sys
from datasets import Dataset, DatasetDict
import evaluate
import numpy as np
import pandas as pd
import torch
from transformers import AutoTokenizer, EsmForSequenceClassification, logging, \
Trainer, TrainingArguments
import wandb
from sklearn.model_selection import train_test_split
def configure_trainer(tokenizer, tokenized_dataset, model_name, sep):
"""
Configure the Trainer.
parameters
----------
tokenizer: torch.Tokenizer
Tokenizer.
tokenized_dataset : torch.Dataset
Dataset containing the tokenized training and test datasets.
model_name : str
Name of the ESM-2 model.
seq: str
Separator sequence.
Returns
-------
trainer : torch.Trainer
Trainer used to retrain ESM-2.
"""
# Configure model
model = EsmForSequenceClassification.from_pretrained(f'facebook/{model_name}', num_labels=2)
model.resize_token_embeddings(len(tokenizer))
# Initialize wandb
wandb.init(project='stapler_esm', name=f'{model_name}_{sep}_ep_pc_v2_aabb')
# Configure training arguments
training_args = TrainingArguments(output_dir=f'tmp/stapler_{model_name}_{sep}_epv2_aabb',
evaluation_strategy='epoch',
per_device_train_batch_size=64,
per_device_eval_batch_size=64,
num_train_epochs=100,
logging_strategy='epoch',
learning_rate=0.000001,
save_total_limit=1,
report_to='wandb',
load_best_model_at_end=True,
metric_for_best_model="accuracy",
save_strategy='epoch',)
# Configure metrics
metric = evaluate.load('accuracy')
def compute_metrics(eval_pred):
logits, labels = eval_pred
predictions = np.argmax(logits, axis=-1)
return metric.compute(predictions=predictions, references=labels)
# Instantiate Trainer
trainer = Trainer(model=model,
args=training_args,
train_dataset=tokenized_dataset['train'],
eval_dataset=tokenized_dataset['test'],
compute_metrics=compute_metrics)
return trainer
def main(model_name, sep):
"""
Entry point of the program.
"""
# Load data
print('No. of cuda devices:', torch.cuda.device_count())
df = pd.read_csv('train-set_full-seq.csv')
# test_df = pd.read_csv('train-set_full-seq.csv')[0.7*23544:]
df['label_true_pair']=df['label_true_pair'].astype('int')
def insert_1_after_characters(s):
return '1'.join(s) + '1'
# train_df['seq_2'] = train_df['seq_2'].apply(insert_1_after_characters)
# test_df['seq_2'] = test_df['seq_2'].apply(insert_1_after_characters)
### comment for 1 vocab
df['epitope_aa'] = df['epitope_aa'].apply(insert_1_after_characters)
train_df, test_df = train_test_split(df, test_size=0.3)
# print((train_df).head())
# Format data
train_df = pd.DataFrame({'seq': train_df['cdr3_alpha_aa'] + sep + train_df['epitope_aa']+ sep +train_df['cdr3_beta_aa'],
'label': train_df['label_true_pair']})
test_df = pd.DataFrame({'seq': test_df['cdr3_alpha_aa'] + sep + test_df['epitope_aa']+ sep +test_df['cdr3_beta_aa'],
'label': test_df['label_true_pair']})
dataset = DatasetDict({
'train': Dataset.from_pandas(train_df),
'test': Dataset.from_pandas(test_df)
})
print('loading model')
# Load tokenizer and add custom tokens
tokenizer = AutoTokenizer.from_pretrained(f'facebook/{model_name}')
tokenizer.add_tokens([sep])
epitope_vocab = ["A1", "C1", "D1", "E1", "F1", "G1", "H1", "I1", "K1", "L1", "M1", "N1", "P1", "Q1", "R1", "S1", "T1", "V1", "W1", "Y1"]
########### comment for 1 vocab
tokenizer.add_tokens(epitope_vocab)
# Tokenize sequences
def tokenize_function(dataset):
return tokenizer(dataset['seq'], return_tensors='pt', max_length=len(tokenizer), padding='max_length', truncation=True)
tokenized_dataset = dataset.map(tokenize_function, batched=True, batch_size=16)
print('loading trainer')
# Configure Trainer
trainer = configure_trainer(tokenizer, tokenized_dataset, model_name, sep)
print('training')
# Run model
trainer.train()
if __name__ == '__main__':
main('esm2_t6_8M_UR50D', '0')