Dataset Handlers

Researcher supplied Python modules that contain a dataset class (to handle processing, splitting, etc.) that are passed to the sacroml.attacks.target.Target must implement one of these abstract classes.

Scikit-learn models that use numpy arrays should implement SklearnDataHandler.

PyTorch models that use DataLoaders should implement PyTorchDataHandler.

API Reference

Abstract data handler supporting both PyTorch and scikit-learn.

class sacroml.attacks.data.BaseDataHandler[source]

Base data handling interface.

abstractmethod __init__() → None[source]: Instantiate a data handler.

class sacroml.attacks.data.PyTorchDataHandler[source]

PyTorch dataset handling interface.

Methods

`get_dataloader`(dataset, indices[, ...])	Return a data loader with a requested subset of samples.
`get_dataset`()	Return a processed dataset.
`get_raw_dataset`()	Return a raw unprocessed dataset.

abstractmethod __init__() → None: Instantiate a data handler.

abstractmethod get_dataloader(dataset: Dataset, indices: Sequence[int], batch_size: int = 32, shuffle: bool = False) → DataLoader[source]

Return a data loader with a requested subset of samples.

Parameters:

datasetDataset: A (processed) PyTorch dataset.
indicesSequence[int]: The indices to load from the dataset.
batch_sizeint: The batch_size to sample the dataset.
shufflebool: Whether to shuffle the data.

Returns:

DataLoader: A PyTorch DataLoader.

abstractmethod get_dataset() → Dataset[source]

Return a processed dataset.

Returns:

Dataset: A (processed) PyTorch dataset.

abstractmethod get_raw_dataset() → Dataset | None[source]

Return a raw unprocessed dataset.

Returns:

Dataset | None: An unprocessed PyTorch dataset.

class sacroml.attacks.data.SklearnDataHandler[source]

Scikit-learn data handling interface.

Methods

`get_data`()	Return the processed data arrays.
`get_raw_data`()	Return the original unprocessed data arrays.
`get_subset`(X, y, indices)	Return a subset of the data.

abstractmethod __init__() → None: Instantiate a data handler.

abstractmethod get_data() → tuple[ndarray, ndarray][source]

Return the processed data arrays.

Returns:

tuple[np.ndarray, np.ndarray]: Features (X) and targets (y) as numpy arrays.

abstractmethod get_raw_data() → tuple[ndarray, ndarray] | None[source]

Return the original unprocessed data arrays.

Returns:

tuple[np.ndarray, np.ndarray] | None: Features (X) and targets (y) as numpy arrays.

abstractmethod get_subset(X: ndarray, y: ndarray, indices: Sequence[int]) → tuple[ndarray, ndarray][source]

Return a subset of the data.

Parameters:

Xnp.ndarray: Feature array.
ynp.ndarray: Target array.
indicesSequence[int]: The indices to extract.

Returns:

tuple[np.ndarray, np.ndarray]: Subset of features and targets.