Regularizers

Regularization in PyKEEN.

Classes

`LpRegularizer`(*[, weight, apply_only_once, ...])	A simple L_p norm based regularizer.
`NoRegularizer`([weight, apply_only_once, ...])	A regularizer which does not perform any regularization.
`CombinedRegularizer`(regularizers[, total_weight])	A convex combination of regularizers.
`PowerSumRegularizer`(*[, weight, ...])	A simple x^p based regularizer.
`OrthogonalityRegularizer`(*[, weight, ...])	A regularizer for the soft orthogonality constraints from [wang2014].
`NormLimitRegularizer`(*[, weight, ...])	A regularizer which formulates a soft constraint on a maximum norm.

class Regularizer(weight=1.0, apply_only_once=False, parameters=None)[source]

A base class for all regularizers.

Instantiate the regularizer.

Parameters:

weight (float) – The relative weight of the regularization
apply_only_once (bool) – Should the regularization be applied more than once after reset?
parameters (Optional[Iterable[Parameter]]) – Specific parameters to track. if none given, it’s expected that your model automatically delegates to the update() function.

add_parameter(parameter)[source]

Add a parameter for regularization.

apply_only_once: bool: Should the regularization only be applied once? This was used for ConvKB and defaults to False.

abstract forward(x)[source]

Compute the regularization term for one tensor.

classmethod get_normalized_name()[source]

Get the normalized name of the regularizer class.

hpo_default: ClassVar[Mapping[str, Any]]: The default strategy for optimizing the regularizer’s hyper-parameters

pop_regularization_term()[source]

Return the weighted regularization term, and reset the regularize afterwards.

post_parameter_update()[source]: Reset the regularizer’s term.

Warning

Typically, you want to use the regularization term exactly once to calculate gradients via pop_regularization_term(). In this case, there should be no need to manually call this method.

regularization_term: torch.FloatTensor: The current regularization term (a scalar)

Reset the regularization term to zero.

property term: FloatTensor

Return the weighted regularization term.

update(*tensors)[source]

Update the regularization term based on passed tensors.