|
import pandas as pd |
|
import torch |
|
import config |
|
import random |
|
from torch.utils.data import Dataset, DataLoader |
|
from torch.nn.utils.rnn import pad_sequence |
|
from pretrained_models import load_esm2_model |
|
|
|
class ProteinDataset(Dataset): |
|
def __init__(self, csv_file, tokenizer): |
|
self.tokenizer = tokenizer |
|
self.data = pd.read_csv(csv_file) |
|
self.max_len = max([len(seq) for seq in self.data['Sequence'].tolist()]) |
|
|
|
def __len__(self): |
|
return len(self.data) |
|
|
|
def __getitem__(self, idx): |
|
sequence = self.data.iloc[idx]['Sequence'].upper() |
|
|
|
|
|
num_masks = int(len(sequence) * 0.15) |
|
mask_indices = random.sample(range(len(sequence)), num_masks) |
|
masked_sequence = ''.join(["<mask>" if i in mask_indices else sequence[i] for i in range(len(sequence))]) |
|
|
|
inputs = self.tokenizer(masked_sequence, padding="max_length", truncation=True, max_length=self.max_len, return_tensors='pt') |
|
input_ids = inputs['input_ids'].squeeze() |
|
attention_mask = inputs['attention_mask'].squeeze() |
|
|
|
labels = self.tokenizer(masked_sequence, return_tensors='pt', padding='max_length', max_length=self.max_len, truncation=True)['input_ids'].squeeze() |
|
labels = torch.where(input_ids == self.tokenizer.mask_token_id, labels, -100) |
|
|
|
return {"input_ids": input_ids, "attention_mask": attention_mask, "labels": labels} |
|
|
|
|
|
|
|
def get_dataloaders(config): |
|
tokenizer, model = load_esm2_model(config.ESM_MODEL_PATH) |
|
|
|
train_dataset = ProteinDataset(config.TRAIN_DATA, tokenizer) |
|
val_dataset = ProteinDataset(config.VAL_DATA, tokenizer) |
|
test_dataset = ProteinDataset(config.TEST_DATA, tokenizer) |
|
|
|
train_loader = DataLoader(train_dataset, batch_size=config.BATCH_SIZE, shuffle=True) |
|
val_loader = DataLoader(val_dataset, batch_size=config.BATCH_SIZE, shuffle=False) |
|
test_loader = DataLoader(test_dataset, batch_size=config.BATCH_SIZE, shuffle=False) |
|
|
|
return train_loader, val_loader, test_loader |
