Datasets:
File size: 4,672 Bytes
84a242a | 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 | import random
import pandas as pd
import pytorch_lightning as pl
from torch_geometric.data import Data, Dataset
from torch_geometric.loader import DataLoader, DataListLoader
from .dataset import CombinedDataset, prepare_data_composite, protein_graph_path
from sklearn.model_selection import train_test_split, KFold
from torch_geometric.utils import degree
import torch
from tqdm import tqdm
class DInterface(pl.LightningDataModule):
def __init__(self, num_workers=8, dataset='', **kwargs):
super().__init__()
pl.seed_everything(kwargs.get('seed'), workers=True)
self.num_workers = num_workers
self.kwargs = kwargs
self.batch_size = kwargs.get('batch_size', 8)
self.incl_columns = kwargs.get('incl_columns', [])
self.test_size = kwargs.get('test_size', 0.5)
self.seed = kwargs.get('seed')
self.k_folds = kwargs.get('k_folds', None) # None means no K-Fold, just train/val/test split
self.current_fold = kwargs.get('fold_num', 0) # Default to first fold
self.benchmark = kwargs.get('benchmark', 'posex_self_docking')
# Prepare dataset differently based on AS criteria
if isinstance(self.benchmark, list):
for b in self.benchmark:
ds, num_classes = prepare_data_composite(self.incl_columns, b)
if hasattr(self, 'dataset'):
self.dataset = pd.concat([self.dataset, ds], ignore_index=True)
else:
self.dataset = ds
self.num_classes = num_classes
elif isinstance(self.benchmark, str):
self.dataset, self.num_classes = prepare_data_composite(self.incl_columns, self.benchmark)
else:
raise ValueError("benchmark must be a string or a list of strings")
self.load_data_module()
def setup(self, stage=None):
if self.k_folds is not None:
print("K-Fold enabled")
self.kf = KFold(n_splits=self.k_folds, shuffle=True, random_state=self.seed)
self.folds = list(self.kf.split(self.dataset))
train_idx, test_idx = self.folds[self.current_fold]
self.trainset, self.valset = train_test_split(self.dataset.iloc[train_idx], test_size=0.1, random_state=self.seed)
self.testset = self.dataset.iloc[test_idx]
else:
print("K-Fold disabled")
if self.test_size == 1 or self.test_size == 0:
self.trainset, self.valset = train_test_split(self.dataset, test_size=0.1, random_state=self.seed)
self.testset = self.dataset
else:
train_val, self.testset = train_test_split(self.dataset, test_size=self.test_size, random_state=self.seed)
self.trainset, self.valset = train_test_split(train_val, test_size=self.test_size, random_state=self.seed)
def train_dataloader(self):
# Combining the protein and ligand graphs into the dataset
trainset_combined = CombinedDataset(self.trainset)
return DataLoader(trainset_combined, batch_size=self.batch_size, shuffle=True, num_workers=self.num_workers, persistent_workers=False, drop_last=False)
def val_dataloader(self):
valset_combined = CombinedDataset(self.valset)
return DataLoader(valset_combined, batch_size=self.batch_size, shuffle=False, num_workers=self.num_workers, persistent_workers=False, drop_last=False)
def test_dataloader(self):
testset_combined = CombinedDataset(self.testset)
return DataLoader(testset_combined, batch_size=self.batch_size, shuffle=False, num_workers=self.num_workers, persistent_workers=False)
def get_train_deg(self):
deg_hist = torch.zeros(256, dtype=torch.long)
return deg_hist
graph_ds = self.trainset # or self.graph_dataset if that’s what you saved
for idx in range(len(graph_ds)):
# Load just the graph part directly
protein_name = graph_ds.iloc[idx]['protein']
ligand_name = graph_ds.iloc[idx]['ligand']
graph_path = f'{protein_graph_path}/pyg_graph_{protein_name}_{ligand_name}_protein.pt'
graph = torch.load(graph_path, weights_only=False)
d = degree(graph.edge_index[1], num_nodes=graph.x.size(0), dtype=torch.long)
deg_hist.index_add_(0, d, torch.ones_like(d))
return deg_hist
# These are from the template, but not used in our implementation.
def load_data_module(self):
pass
def instancialize(self, **other_args):
pass |