Skip to content

Improve speciation algorithm and interface #108

Closed

Description

@HyperCodec

I'm still not satisfied with the new implementation of the speciation feature.

First of all, I think it's bad to require Eq and Hash in the Species type. Perhaps I can fix this by using a BTreeMap, but then I need PartialOrd.

I also don't like the way the algorithm works. It seems naive, slow, and messy.

Activity

  1. self-assigned this
    on Oct 10, 2025
  2. HyperCodec commented on Mar 2, 2026

    @HyperCodec
    OwnerAuthor

    I think we just remove Species altogether, and just have a simple method is_same_species where users can implement whether two genomes are compatible with each other.

  3. HyperCodec commented on Mar 9, 2026

    @HyperCodec
    OwnerAuthor

    going with an implementation based on what was designed for NEAT, as I think an algorithm that is capable of preserving novel structures even in the complex NEAT setup would work well if applied broadly to the rest of genetic algorithms. not using innovation numbers and such tho.

  4. linked a pull request that will close this issueRewrite speciation feature #144on Mar 9, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

enhancementNew feature or request

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions