Q&A 36.2 — FitnessEvaluator: Multi-Objective & Novelty Search Questions

[web3guru888]

FitnessEvaluator — Q&A Thread

Use this discussion for questions about Phase 36.2 FitnessEvaluator.

Common Questions

Q: When to use NSGA-II vs MOEA/D?

• NSGA-II: 2-3 objectives, no prior knowledge of Pareto front shape
• MOEA/D: 3+ objectives (many-objective), uniform reference directions needed
• SPEA2: when fine-grained density estimation matters

Q: How to set novelty search k parameter?

• k=15 is a robust default (Lehman & Stanley 2011)
• Smaller k: more local novelty (fine-grained exploration)
• Larger k: more global novelty (macro-level diversity)

Q: How to blend fitness and novelty?

• weight_novelty=0.0: pure fitness (standard EA)
• weight_novelty=1.0: pure novelty search
• weight_novelty=0.3-0.5: hybrid, recommended for deceptive landscapes

Q: How to handle constraint violations in multi-objective?

• Feasibility-first: feasible always dominates infeasible
• Penalty-based: add constraint violation as additional objective
• ε-constraint: convert constraints to objectives with tolerance

Related issue: #740
Planning: #738