Reinforcement Learning Algorithms with Python

Evolution strategies

Evolution strategies (ESes) are even easier than GAs as they are primarily based on mutation to create a new population.

Mutation is performed by adding values that have been sampled from a normal distribution to the genotype. A very simple version of ES is obtained by just selecting the most performant individual across the whole population and sampling the next generation from a normal distribution with a fixed standard deviation and a mean equal to that of the best-performing individual.

Outside of the sphere of small problems, using this algorithm is not recommended. This is because following only a single leader and using a fixed standard deviation could prevent potential solutions from exploring a more diverse search ...