Upload 2 files
Browse files- utils/data_loader.py +24 -11
- utils/esm_utils.py +1 -0
utils/data_loader.py
CHANGED
@@ -1,6 +1,9 @@
|
|
1 |
import pandas as pd
|
|
|
2 |
from torch.utils.data import Dataset, DataLoader
|
3 |
-
from utils.
|
|
|
|
|
4 |
|
5 |
class ProteinDataset(Dataset):
|
6 |
def __init__(self, csv_file, tokenizer, model):
|
@@ -12,19 +15,29 @@ class ProteinDataset(Dataset):
|
|
12 |
return len(self.data)
|
13 |
|
14 |
def __getitem__(self, idx):
|
15 |
-
sequence = self.data.iloc[idx]['
|
16 |
latents = get_latents(self.model, self.tokenizer, sequence)
|
17 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
|
19 |
def get_dataloaders(config):
|
20 |
-
tokenizer, model = load_esm2_model(config.
|
21 |
|
22 |
-
train_dataset = ProteinDataset(config.
|
23 |
-
val_dataset = ProteinDataset(config.
|
24 |
-
test_dataset = ProteinDataset(config.
|
25 |
|
26 |
-
train_loader = DataLoader(train_dataset, batch_size=config.
|
27 |
-
val_loader = DataLoader(val_dataset, batch_size=config.
|
28 |
-
test_loader = DataLoader(test_dataset, batch_size=config.
|
29 |
|
30 |
-
return train_loader, val_loader, test_loader
|
|
|
1 |
import pandas as pd
|
2 |
+
import torch
|
3 |
from torch.utils.data import Dataset, DataLoader
|
4 |
+
from torch.nn.utils.rnn import pad_sequence
|
5 |
+
from esm_utils import get_latents, load_esm2_model
|
6 |
+
import config
|
7 |
|
8 |
class ProteinDataset(Dataset):
|
9 |
def __init__(self, csv_file, tokenizer, model):
|
|
|
15 |
return len(self.data)
|
16 |
|
17 |
def __getitem__(self, idx):
|
18 |
+
sequence = self.data.iloc[idx]['Sequence']
|
19 |
latents = get_latents(self.model, self.tokenizer, sequence)
|
20 |
+
|
21 |
+
attention_mask = torch.ones_like(latents)
|
22 |
+
attention_mask = torch.mean(attention_mask, dim=-1)
|
23 |
+
|
24 |
+
return latents, attention_mask
|
25 |
+
|
26 |
+
def collate_fn(batch):
|
27 |
+
latents, attention_mask = zip(*batch)
|
28 |
+
latents_padded = pad_sequence([torch.tensor(latent) for latent in latents], batch_first=True, padding_value=0)
|
29 |
+
attention_mask_padded = pad_sequence([torch.tensor(mask) for mask in attention_mask], batch_first=True, padding_value=0)
|
30 |
+
return latents_padded, attention_mask_padded
|
31 |
|
32 |
def get_dataloaders(config):
|
33 |
+
tokenizer, model = load_esm2_model(config.MODEL_NAME)
|
34 |
|
35 |
+
train_dataset = ProteinDataset(config.Loader.DATA_PATH + "/train.csv", tokenizer, model)
|
36 |
+
val_dataset = ProteinDataset(config.Loader.DATA_PATH + "/val.csv", tokenizer, model)
|
37 |
+
test_dataset = ProteinDataset(config.Loader.DATA_PATH + "/test.csv", tokenizer, model)
|
38 |
|
39 |
+
train_loader = DataLoader(train_dataset, batch_size=config.Loader.BATCH_SIZE, num_workers=0, shuffle=True, collate_fn=collate_fn)
|
40 |
+
val_loader = DataLoader(val_dataset, batch_size=config.Loader.BATCH_SIZE, num_workers=0, shuffle=False, collate_fn=collate_fn)
|
41 |
+
test_loader = DataLoader(test_dataset, batch_size=config.Loader.BATCH_SIZE, num_workers=0, shuffle=False, collate_fn=collate_fn)
|
42 |
|
43 |
+
return train_loader, val_loader, test_loader
|
utils/esm_utils.py
CHANGED
@@ -11,3 +11,4 @@ def get_latents(model, tokenizer, sequence):
|
|
11 |
with torch.no_grad():
|
12 |
outputs = model(**inputs)
|
13 |
return outputs.last_hidden_state.squeeze(0)
|
|
|
|
11 |
with torch.no_grad():
|
12 |
outputs = model(**inputs)
|
13 |
return outputs.last_hidden_state.squeeze(0)
|
14 |
+
|