data.buffer.middleware.priority

priority

PriorityExperienceReplay

class ding.data.buffer.middleware.priority.PriorityExperienceReplay(buffer: Buffer, IS_weight: bool = True, priority_power_factor: float = 0.6, IS_weight_power_factor: float = 0.4, IS_weight_anneal_train_iter: int = 100000)

Overview:

The middleware that implements priority experience replay (PER).

__init__(buffer: Buffer, IS_weight: bool = True, priority_power_factor: float = 0.6, IS_weight_power_factor: float = 0.4, IS_weight_anneal_train_iter: int = 100000) → None

Arguments:

• buffer (Buffer): The buffer to use PER.

• IS_weight (bool): Whether use importance sampling or not.

• priority_power_factor (float): The factor that adjust the sensitivity between the sampling probability and the priority level.

• IS_weight_power_factor (float): The factor that adjust the sensitivity between the sample rarity and sampling probability in importance sampling.

• IS_weight_anneal_train_iter (float): The factor that controls the increasing of IS_weight_power_factor during training.