PowerSeriesGradientPerturbation

class PowerSeriesGradientPerturbation(func, grad, func_args=None, grad_args=None, n_red=1, step_size=0.5, step_size_modifier=1.0, n_restart=-1, disable_gradient_scaling=False, iterative_scaling=False, bypass_objective_decrease=False)

Bases: GradientPerturbation

Objective function perturbation using gradient descent with step size reduction according to a power series. Has the option to restart the power series after a certain number of steps.

Parameters:

• func (Callable) – The function to be reduced.

• grad (Callable) – The gradient of the function to be reduced.

• func_args (Optional[List]) – Additional arguments to be passed to the function, by default [].

• grad_args (Optional[List]) – Additional arguments to be passed to the gradient function, by default [].

• n_red (int) – The number of reductions, by default 1.

• step_size (float) – The step size for the gradient descent, by default 0.5.

• step_size_modifier (float) – Scaling factor for the step size power series, by default 1.0.

• n_restart (int) – The number of steps after which to restart the power series, by default -1 (no restart).

• disable_gradient_scaling (bool) – If true, skip the normalization of the gradient, by default False.

• iterative_scaling (bool) – If true, the power series is scaled by the iteration k without checking whether this actually decreases the objective, by default False.

• bypass_objective_decrease (bool) – If true, the check whether the objective function value actually decreases is bypassed, by default False.

Methods

perturbation_step(x)	Perform n_red perturbation steps on the input array.
post_step(x, *args, **kwargs)	If required perform any form of step after each perturbation in each iteration.
pre_step(x, *args, **kwargs)	Resets the power series after n steps.
reset()	Reset the perturbation to its initial state.

perturbation_step(x)

Perform n_red perturbation steps on the input array.

Parameters:

x (ndarray[tuple[Any, ...], dtype[TypeVar(_ScalarT, bound= generic)]]) – The input array to be perturbed.

Return type:

ndarray

Returns:

The perturbed array after applying the reduction steps.

post_step(x, *args, **kwargs)

If required perform any form of step after each perturbation in each iteration.

This method is intended to be overridden by subclasses to implement specific post-step logic. By default, it does nothing.

Parameters:

x (ndarray[tuple[Any, ...], dtype[TypeVar(_ScalarT, bound= generic)]]) – Current iterate.

Return type:

None

pre_step(x, *args, **kwargs)

Resets the power series after n steps.

Parameters:

x (ndarray[tuple[Any, ...], dtype[TypeVar(_ScalarT, bound= generic)]]) – Current iterate.

Return type:

None

reset()

Reset the perturbation to its initial state.

Return type:

None