Module evolution 

Evolutionary hyperparameter search for super::OptimizationStrategy::Evolutionary (F15).

The previous implementation pushed 20 random configurations into state.population and then never read it again — the population was write-only, and every suggestion after the first was a mutation of the single best configuration, so there was no recombination and no selection pressure beyond “the best so far”.

This module implements real selection, crossover and mutation over the population, all bounded by the search space.

Structs

EvolutionConfig

Tunables of the evolutionary strategy.

Functions

breed

Produce one offspring by tournament-selecting two parents, recombining them per parameter, and mutating the result inside the declared bounds.

survivor_selection

Trim population to the fittest population_size individuals (higher score first). Called after every recorded evaluation so the population is a real, bounded, elitist pool rather than an append-only log.