src.common.WSDataLoader

class src.common.WSDataLoader(config)

Bases: ABC

Abstract base class for weak supervision data loaders

Methods

create_dataloader(x_data, labels[, ...])	Create PyTorch DataLoader
create_soft_dataloader(x_data, soft_labels)	Create DataLoader for soft labels
create_weak_dataloader(x_data, weak_labels)	Create DataLoader for weak labels
load_split_data(dataset, seed)	Load train/val/test splits

create_dataloader(x_data, labels, batch_size=200, shuffle=True)

Create PyTorch DataLoader

Args:

x_data: Input data labels: Labels batch_size: Batch size shuffle: Whether to shuffle data

Returns:

PyTorch DataLoader

Parameters:

• x_data (numpy.ndarray)

• labels (numpy.ndarray)

• batch_size (int)

• shuffle (bool)

Return type:

torch.utils.data.DataLoader

create_soft_dataloader(x_data, soft_labels, batch_size=200, shuffle=True)

Create DataLoader for soft labels

Args:

x_data: Input data soft_labels: Soft labels/probabilities batch_size: Batch size shuffle: Whether to shuffle data

Returns:

PyTorch DataLoader

Parameters:

• x_data (numpy.ndarray)

• soft_labels (numpy.ndarray)

• batch_size (int)

• shuffle (bool)

Return type:

torch.utils.data.DataLoader

create_weak_dataloader(x_data, weak_labels, batch_size=200, shuffle=True)

Create DataLoader for weak labels

Args:

x_data: Input data weak_labels: Weak supervision labels batch_size: Batch size shuffle: Whether to shuffle data

Returns:

PyTorch DataLoader

Parameters:

• x_data (numpy.ndarray)

• weak_labels (numpy.ndarray)

• batch_size (int)

• shuffle (bool)

Return type:

torch.utils.data.DataLoader

abstractmethod load_split_data(dataset, seed)

Load train/val/test splits

Args:

dataset: Dataset name seed: Random seed

Returns:

Tuple of (train_data, val_data, test_data, metadata)

Parameters:

• dataset (str)

• seed (int)

Return type:

Tuple[Any, Any, Any, Any | None]