Hands-On Machine Learning for Algorithmic Trading

Rewards – learning from actions

The reward signal is a single value that's sent to the agent at each time step. The agent's objective is to maximize the total reward received over time. Rewards can also be a stochastic function of the state and the actions. They are typically discounted to facilitate convergence and reflect the time decay of a value.

Rewards are the only way for the agent to learn about the value of its decisions in a given state and to modify the policy accordingly. Due to its critical impact on the agent's learning, the reward signal is often the most challenging part of designing an RL system.

Rewards need to clearly communicate what the agent should accomplish (as opposed to how it should do so) and may require domain ...