Loading…
Glossary term
In reinforcement learning, the strategy that maps situations to actions.
In reinforcement learning, the policy is the agent's strategy: a function that maps each situation to an action, or to probabilities over actions. Learning aims to find the policy with the highest expected reward.
In deep reinforcement learning the policy is a neural network.
| Chosen action | |
|---|---|
| State: low battery | Go charge |
| State: full battery | Explore |
| State: obstacle | Turn left |
The policy is what the learned agent actually does.