Target#

Store information about the target model and data.

class sacroml.attacks.target.Target(model: SklearnModel | PytorchModel | BaseEstimator | Module | None = None, model_path: str = '', model_module_path: str = '', model_name: str = '', model_params: dict | None = None, train_module_path: str = '', train_params: dict | None = None, batch_size: int | None = None, dataset_name: str = '', dataset_module_path: str = '', features: dict = <factory>, X_train: ndarray | None = None, y_train: ndarray | None = None, X_test: ndarray | None = None, y_test: ndarray | None = None, X_train_orig: ndarray | None = None, y_train_orig: ndarray | None = None, X_test_orig: ndarray | None = None, y_test_orig: ndarray | None = None, proba_train: ndarray | None = None, proba_test: ndarray | None = None, indices_train: Sequence[int] | None = None, indices_test: Sequence[int] | None = None, safemodel: list = <factory>)[source]#

Store information about the target model and data.

Attributes:

modelModelLike: Trained target model.
model_pathstr: Path to a saved model.
model_module_pathstr: Path to module containing model class.
model_namestr: Class name of model.
model_paramsdict or None: Hyperparameters for instantiating the model.
train_module_pathstr: Path to module containing training function.
train_paramsdict or None: Hyperparameters for training the model.
batch_sizeint or None: Batch size used to train the original target model. Threaded to PyTorch model (re)training and inference. Should match the original training batch size since it materially affects attack model fidelity (e.g. shadow models). When None, PyTorch defaults to 32.
dataset_namestr: The name of the dataset.
dataset_module_pathstr: Path to module containing dataset loading function.
featuresdict: Dictionary describing the dataset features.
X_trainnp.ndarray or None: The (processed) training inputs.
y_trainnp.ndarray or None: The (processed) training outputs.
X_testnp.ndarray or None: The (processed) testing inputs.
y_testnp.ndarray or None: The (processed) testing outputs.
X_train_orignp.ndarray or None: The original (unprocessed) training inputs.
y_train_orignp.ndarray or None: The original (unprocessed) training outputs.
X_test_orignp.ndarray or None: The original (unprocessed) testing inputs.
y_test_orignp.ndarray or None: The original (unprocessed) testing outputs.
proba_trainnp.ndarray or None: The model predicted training probabilities.
proba_testnp.ndarray or None: The model predicted testing probabilities.
indices_trainSequence[int] or None: Indices of training set samples.
indices_testSequence[int] or None: Indices of test set samples.
safemodellist: Results of safemodel disclosure checking.

Methods

`add_feature`(name, indices, encoding)	Add a feature description to the data dictionary.
`add_safemodel_results`(data)	Add safemodel disclosure checking results.
`get_generalisation_error`()	Calculate model generalisation error.
`has_data`()	Return whether the target has all processed data.
`has_model`()	Return whether the target has a loaded model.
`has_probas`()	Return whether the target has all probability data.
`has_raw_data`()	Return whether the target has all raw data.
`load`([path])	Load target from persistent storage.
`load_array`(arr_path, attr_name)	Load array from pickle or CSV file.
`load_pytorch_dataset`()	Wrap dataset for Pytorch models given a dataset Python script.
`load_sklearn_dataset`()	Wrap dataset for scikit-learn models given a dataset Python script.
`save`([path, ext])	Save target to persistent storage.

__init__(model: SklearnModel | PytorchModel | BaseEstimator | Module | None = None, model_path: str = '', model_module_path: str = '', model_name: str = '', model_params: dict | None = None, train_module_path: str = '', train_params: dict | None = None, batch_size: int | None = None, dataset_name: str = '', dataset_module_path: str = '', features: dict = <factory>, X_train: ndarray | None = None, y_train: ndarray | None = None, X_test: ndarray | None = None, y_test: ndarray | None = None, X_train_orig: ndarray | None = None, y_train_orig: ndarray | None = None, X_test_orig: ndarray | None = None, y_test_orig: ndarray | None = None, proba_train: ndarray | None = None, proba_test: ndarray | None = None, indices_train: Sequence[int] | None = None, indices_test: Sequence[int] | None = None, safemodel: list = <factory>) → None#

add_feature(name: str, indices: list[int], encoding: str) → None[source]#: Add a feature description to the data dictionary.

add_safemodel_results(data: list) → None[source]#: Add safemodel disclosure checking results.

get_generalisation_error() → float[source]#: Calculate model generalisation error.

has_data() → bool[source]#: Return whether the target has all processed data.

has_model() → bool[source]#: Return whether the target has a loaded model.

has_probas() → bool[source]#: Return whether the target has all probability data.

has_raw_data() → bool[source]#: Return whether the target has all raw data.

load(path: str = 'target') → None[source]#: Load target from persistent storage.

load_array(arr_path: str, attr_name: str) → None[source]#: Load array from pickle or CSV file.

load_pytorch_dataset() → None[source]#: Wrap dataset for Pytorch models given a dataset Python script.

load_sklearn_dataset() → None[source]#: Wrap dataset for scikit-learn models given a dataset Python script.

save(path: str = 'target', ext: str = 'pkl') → None[source]#: Save target to persistent storage.

X_test: ndarray | None = None#

X_test_orig: ndarray | None = None#

X_train: ndarray | None = None#

X_train_orig: ndarray | None = None#

batch_size: int | None = None#

dataset_module_path: str = ''#

dataset_name: str = ''#

features: dict#

indices_test: Sequence[int] | None = None#

indices_train: Sequence[int] | None = None#

model: SklearnModel | PytorchModel | BaseEstimator | Module | None = None#

model_module_path: str = ''#

model_name: str = ''#

model_params: dict | None = None#

model_path: str = ''#

property n_features: int#: Number of features.

proba_test: ndarray | None = None#

proba_train: ndarray | None = None#

safemodel: list#

train_module_path: str = ''#

train_params: dict | None = None#

y_test: ndarray | None = None#

y_test_orig: ndarray | None = None#

y_train: ndarray | None = None#

y_train_orig: ndarray | None = None#