Reinforcement learning utils

This module contains the building blocks the algorithms are assembled from. They are collected here rather than inside the algorithms because most of them are shared by several families — a Parameter schedules a learning rate as readily as an exploration coefficient, and the same advantage estimator serves every on-policy actor-critic.

Parameter	This class implements function to manage parameters, such as learning rate.
Optimizer	Base class for gradient optimizers.
StandardizationPreprocessor	Preprocess observations from the environment using a running standardization.
ReplayMemory	This class implements function to manage a replay memory as the one used in "Human-Level Control Through Deep Reinforcement Learning" by Mnih V.
EligibilityTrace	Factory method to create an eligibility trace of the provided type.
RunningStandardization	Compute a running standardization of values according to Welford's online algorithm.

• Parameters
• Optimizers
• Preprocessors
• Replay memory
• Eligibility trace
• Running statistics
• Value functions