Summary

Implement metaheuristics module in aprender-contrib for derivative-free global optimization, following the revised specification in docs/specifications/metaheuristics-spec.md (v1.1).

Scope

Phase 1: Perturbative Core (v0.12.0)

• SearchSpace enum (Continuous, Mixed, Binary, Permutation, Graph)
• PerturbativeMetaheuristic trait
• Differential Evolution (DE/rand/1/bin, JADE, SHADE)
• Particle Swarm Optimization (standard, constriction, SPSO 2011)
• Simulated Annealing (geometric, adaptive cooling)
• Genetic Algorithm (SBX, polynomial mutation)
• Harmony Search
• CEC 2013 benchmark suite

Phase 2: CMA-ES & Binary GA (v0.13.0) [HIGH COMPLEXITY]

• CMA-ES with eigendecomposition, boundary handling, IPOP restart
• Binary GA with BitVec solution type
• Feature selection utilities

Phase 3: Constructive Metaheuristics (v0.14.0)

• ConstructiveMetaheuristic trait
• NeighborhoodSearch trait
• Ant Colony Optimization (AS, MMAS, ACS)
• Tabu Search

Phase 4: ML Integration (v0.15.0)

• HyperoptSearch wrapper
• Neural architecture search primitives

Quality Requirements (EXTREME TDD)

• 95%+ test coverage
• Property tests for convergence on unimodal functions
• Mutation score ≥85%
• Zero clippy warnings
• CEC 2013/2017 benchmark validation

References

• Specification: docs/specifications/metaheuristics-spec.md
• 16 peer-reviewed publications cited
• Toyota Way review incorporated (v1.1)

Labels

epic, enhancement, aprender-contrib