niteshade.defence

Data poisoning defence strategy classes following a logical hierarchy.

Classes

Defender()	Abstractclass that the defenders use.
DefenderGroup(defender_list[, ...])	Class allowing the grouping of defenders through a input list containing defender objects.
Distance_metric([type])	A Distance_metric class for the feasibleset defender.
FeasibleSetDefender(initial_dataset_x, ...)	A FeasibleSetDefender class, inheriting from the OutlierDefender.
KNN_Defender(init_x, init_y, ...[, one_hot])	A KNN class, inheriting from the PointModifierDefender, that flips the labels of input points if the proportion of the most frequent label of nearest neighbours exceeds a threshold.
ModelDefender()	Abstractclass for defenders that use a strategy that needs access to the model.
OutlierDefender(initial_dataset_x, ...)	Abstractclass for defenders that use a outlier filtering strategy.
PointModifierDefender()	Abstractclass for defenders that use a strategy that modifies the input points.
SoftmaxDefender([threshold, delay, one_hot])	A SoftmaxDefender class, inheriting from the ModelDefender.