Multiple-Frequencies Population-Based Training

This section covers Multiple-Frequencies Population-Based Training (MF-PBT), the multi-frequency selection strategy for AgileRL’s evolutionary hyperparameter optimisation loop. MF-PBT is a drop-in replacement for tournament selection and is algorithm-agnostic, so the tutorial below demonstrates it on a single representative algorithm (PPO) and the same wiring applies everywhere else.

LunarLander gif

Lunar Lander with PPO & MF-PBT