Module Fehu_algorithmsSource

Reinforcement learning algorithms for Fehu.

Each algorithm follows a functional interface:

• Algorithm.init prepares parameters and algorithm state for a given environment;
• Algorithm.step performs a single environment interaction and optimisation update;
• Algorithm.train runs a default training loop that repeatedly calls Algorithm.step.

Available Algorithms

Policy Gradient Methods

• Reinforce: Monte Carlo Policy Gradient (REINFORCE)

Value-Based Methods

• Dqn: Deep Q-Network (DQN)

Future algorithms:

• PPO: More sample efficient, supports continuous actions, industry standard
• SAC: Off-policy actor-critic, excellent for continuous control