Sferes

Sferes is an open-source C++ library for evolutionary computation and neuroevolution. It provides a modular framework that enables researchers and students to prototype genetic algorithms, neuroevolution, and other bio-inspired optimization methods. The core idea is to separate problem definition from the evolution mechanics, so users can plug in their own genome representations, fitness functions, and phenotypic mappings while reusing common evolution operators and infrastructure.

The library organizes code around genomes, phenotypes, fitness evaluations, populations, and evolution loops. Users implement a genome type and a phenotype transformation, then define a fitness function that evaluates individuals. Sferes supplies flexible templates for selection, mutation, crossover, and reproduction, plus utilities for running multiple experiments, collecting statistics, and saving checkpoints. It also includes support for parallel evaluation and multi-objective optimization, and provides facilities to log performance and visualize results.

Sferes includes example experiments, documentation, and a set of ready-made components to aid rapid development of new approaches, including evolvable neural networks and simple robot controllers. It is designed to be portable across platforms with common C++ toolchains and standard libraries, and to integrate with existing project workflows.

As a research-oriented tool, sferes has been used to study evolvable neural networks, behavior learning, and the optimization of complex tasks. Its design emphasizes reusability, extensibility, and clarity, making it a reference point for experimental setups in evolutionary computation and related fields.