Multistart swap local search

Track: exact-K binary selection. Method ID: swap.

Starts from top returns, diagonal objective costs, and seeded random subsets. Repeatedly applies the best improving exchange of one selected and one unselected asset.

Dependencies: NumPy; no extra dependencies.

Key parameters: seed, restarts (at least two), max_swaps, and an optional candidate_callback.

Guarantees and limits: Every exchange preserves K. An uncapped descent reaches a one-swap local optimum within numerical tolerance; the swap cap can stop earlier. Neither implies a global optimum.

Example

Use the shared inputs.

from vqe_portfolio.baselines import local_search_binary_baseline

result = local_search_binary_baseline(mu, Sigma, lam=4.0, k=2, seed=0, restarts=2)