smartgrid.rewards¶
This package contains various reward functions, used to compute a reward signal for each agent, based on their action and the resulting state of the environment.
Reward functions indicate to which degree the agent’s actions were appropriate, or in this case, ethical. By ethical, we mean that they are aligned with one or several moral values. These moral values are encoded in the reward function itself, which guides the learning of agents.
Traditionally, reward functions in the Reinforcement Learning literature are
purely numerical, i.e., based on mathematical expressions. Such functions can
be found in the smartgrid.rewards.numeric package.
Other functions can also be based on symbolic reasoning. (Work in progress)
- members:
Modules
Numeric reward functions are purely based on mathematical expressions. |
|
The Reward abstract class defines a common (standard) API for reward functions. |
|
The RewardCollection is responsible for computing rewards from reward functions. |