alibi_detect.ad.model_distillation

Constants

logger

logger: logging.Logger = <Logger alibi_detect.ad.model_distillation (WARNING)>

Instances of the Logger class represent a single logging channel. A "logging channel" indicates an area of an application. Exactly how an "area" is defined is up to the application developer. Since an application can have any number of areas, logging channels are identified by a unique string. Application areas can be nested (e.g. an area of "input processing" might include sub-areas "read CSV files", "read XLS files" and "read Gnumeric files"). To cater for this natural nesting, channel names are organized into a namespace hierarchy where levels are separated by periods, much like the Java or Python package namespace. So in the instance given above, channel names might be "input" for the upper level, and "input.csv", "input.xls" and "input.gnu" for the sub-levels. There is no arbitrary limit to the depth of nesting.

ModelDistillation

Inherits from: BaseDetector, FitMixin, ThresholdMixin, ABC

Constructor

ModelDistillation(self, threshold: float = None, distilled_model: keras.src.models.model.Model = None, model: keras.src.models.model.Model = None, loss_type: str = 'kld', temperature: float = 1.0, data_type: str = None) -> None

Name	Type	Default	Description
threshold	Optional[float]	None	Threshold used for score to determine adversarial instances.
distilled_model	Optional[keras.src.models.model.Model]	None	A tf.keras model to distill.
model	Optional[keras.src.models.model.Model]	None	A trained tf.keras classification model.
loss_type	str	'kld'	Loss for distillation. Supported: 'kld', 'xent'
temperature	float	1.0	Temperature used for model prediction scaling. Temperature <1 sharpens the prediction probability distribution.
data_type	Optional[str]	None	Optionally specifiy the data type (tabular, image or time-series). Added to metadata.

Methods

fit

fit(X: numpy.ndarray, loss_fn: .tensorflow.keras.losses = <function loss_distillation at 0x28ee8c4c0>, optimizer: Union[ForwardRef('tf.keras.optimizers.Optimizer'), ForwardRef('tf.keras.optimizers.legacy.Optimizer'), type[ForwardRef('tf.keras.optimizers.Optimizer')], type[ForwardRef('tf.keras.optimizers.legacy.Optimizer')]] = <class 'keras.src.optimizers.adam.Adam'>, epochs: int = 20, batch_size: int = 128, verbose: bool = True, log_metric: Tuple[str, ForwardRef('tf.keras.metrics')] = None, callbacks: .tensorflow.keras.callbacks = None, preprocess_fn: Callable = None) -> None

Train ModelDistillation detector.

Name	Type	Default	Description
X	numpy.ndarray		Training batch.
loss_fn	.tensorflow.keras.losses	<function loss_distillation at 0x28ee8c4c0>	Loss function used for training.
optimizer	Union[ForwardRef('tf.keras.optimizers.Optimizer'), ForwardRef('tf.keras.optimizers.legacy.Optimizer'), type[ForwardRef('tf.keras.optimizers.Optimizer')], type[ForwardRef('tf.keras.optimizers.legacy.Optimizer')]]	<class 'keras.src.optimizers.adam.Adam'>	Optimizer used for training.
epochs	int	20	Number of training epochs.
batch_size	int	128	Batch size used for training.
verbose	bool	True	Whether to print training progress.
log_metric	Tuple[str, ForwardRef('tf.keras.metrics')]	None	Additional metrics whose progress will be displayed if verbose equals True.
callbacks	.tensorflow.keras.callbacks	None	Callbacks used during training.
preprocess_fn	Callable	None	Preprocessing function applied to each training batch.

Returns

• Type: None

infer_threshold

infer_threshold(X: numpy.ndarray, threshold_perc: float = 99.0, margin: float = 0.0, batch_size: int = 10000000000) -> None

Update threshold by a value inferred from the percentage of instances considered to be

adversarial in a sample of the dataset.

Name	Type	Default	Description
X	numpy.ndarray		Batch of instances.
threshold_perc	float	99.0	Percentage of X considered to be normal based on the adversarial score.
margin	float	0.0	Add margin to threshold. Useful if adversarial instances have significantly higher scores and there is no adversarial instance in X.
batch_size	int	10000000000	Batch size used when computing scores.

Returns

• Type: None

predict

predict(X: numpy.ndarray, batch_size: int = 10000000000, return_instance_score: bool = True) -> Dict[Dict[str, str], Dict[str, numpy.ndarray]]

Predict whether instances are adversarial instances or not.

Name	Type	Default	Description
X	numpy.ndarray		Batch of instances.
batch_size	int	10000000000	Batch size used when computing scores.
return_instance_score	bool	True	Whether to return instance level adversarial scores.

Returns

• Type: Dict[Dict[str, str], Dict[str, numpy.ndarray]]

score

score(X: numpy.ndarray, batch_size: int = 10000000000, return_predictions: bool = False) -> Union[numpy.ndarray, Tuple[numpy.ndarray, numpy.ndarray, numpy.ndarray]]

Compute adversarial scores.

Name	Type	Default	Description
X	numpy.ndarray		Batch of instances to analyze.
batch_size	int	10000000000	Batch size used when computing scores.
return_predictions	bool	False	Whether to return the predictions of the classifier on the original and reconstructed instances.

Returns

• Type: Union[numpy.ndarray, Tuple[numpy.ndarray, numpy.ndarray, numpy.ndarray]]