Optimization Results#

Results Vary Between Runs#

Cause: Optimization algorithms are stochastic.

Solution: Set a random seed for reproducibility:

optimizer = HillClimbing(
    search_space=space,
    n_iter=100,
    experiment=objective,
    random_state=42,
)

Optimizer Gets Stuck in Local Optima#

Cause: Local search algorithms (like HillClimbing) can get trapped.

Solutions:

  1. Use a global search algorithm:

    from hyperactive.opt.gfo import RandomSearch, BayesianOptimizer

  2. Use population-based methods:

    from hyperactive.opt.gfo import ParticleSwarmOptimizer, GeneticAlgorithm

  3. Increase exploration in local search:

    optimizer = HillClimbing(
    search_space=space,
    n_iter=100,
    experiment=objective,
    epsilon=0.2,  # Larger steps
)

Best Score is Very Low or Negative#

Check these:

  1. Objective function errors — Make sure your objective doesn’t crash:

    def objective(params):
    try:
        score = evaluate(params)
        return score
    except Exception as e:
        print(f"Error: {e}")  # Debug
        return -np.inf

  2. Sign convention — Hyperactive maximizes. Negate if minimizing:

    def objective(params):
    error = compute_error(params)
    return -error  # Negate for minimization