algorithms.util.action_perturbator.MultiDimActionPerturbator

class algorithms.util.action_perturbator.MultiDimActionPerturbator(noise: float, probability: float)

Bases: ActionPerturbator

Custom algorithm, with a probability to noise each dimension.

__init__(noise: float, probability: float)

Methods

__init__(noise, probability)
perturb(action[, clip])	Add a noise to the proposed action.

perturb(action, clip=True)

Add a noise to the proposed action.

Parameters:

• action (np.ndarray) – A 1D vector of values representing the action.

• clip – Controls whether to clip the noised values in [0,1].

Returns:

A vector of noised values, with the same shape as action.

Return type:

np.ndarray