Q&A 37.2 — AntColonyOptimizer: Variant Selection, Parameter Tuning & Edge Cases

[web3guru888]

Q&A: AntColonyOptimizer (Phase 37.2)

Configuration Questions

Q1: Which ACO variant should I use?

• AS (Ant System): Simplest, good baseline. All ants deposit pheromone. Best for small instances.
• ACS (Ant Colony System): Best for exploitation. q₀ parameter controls greediness, local pheromone update prevents stagnation. Recommended default.
• MMAS (MAX-MIN Ant System): Only best ant deposits. Pheromone bounds prevent premature convergence. Best for large instances.

Q2: How to set α and β parameters?
Standard: α=1, β=2–5. Higher β emphasizes heuristic information (greedy). Higher α emphasizes learned pheromone trails. For TSP: α=1, β=5 works well (Dorigo & Stützle 2004).

Q3: What evaporation rate ρ should I use?

• AS: ρ = 0.5 (moderate decay)
• ACS: ρ = 0.1 (slow global decay, fast local decay via ξ)
• MMAS: ρ = 0.02 (very slow, bounds prevent stagnation)

Q4: How to adapt ACO for non-TSP problems?
Replace heuristic η_ij with problem-specific info:

• Job scheduling: η = 1/processing_time
• Graph coloring: η = saturation_degree
• Vehicle routing: η = 1/distance + capacity_urgency

Edge Cases

Q5: What if pheromone trails converge to a single path?
Stagnation detection: if all ants produce identical tours for N consecutive iterations:

1. MMAS: pheromone reinitialization to τ_max
2. ACS: local update already diversifies via τ₀ deposit
3. General: increase evaporation rate temporarily

Q6: How to handle asymmetric distance matrices?
Use directed pheromone: τ_ij ≠ τ_ji. Separate pheromone matrices for forward/backward edges. ACS local update applies only to the traversed direction.

Ref: Issue #756 | Wiki: Phase-37-Ant-Colony-Optimizer