Nous.Eval.Optimizer.Strategies.Random (nous v0.16.2)

Random search optimization strategy.

Random search samples configurations randomly from the search space. Often surprisingly effective and much faster than grid search for high-dimensional spaces.

Options

• :n_trials - Number of trials to run (default: 100)
• :timeout - Total timeout in ms (default: 3600000 = 1 hour)
• :early_stop - Stop if score reaches threshold
• :verbose - Print progress (default: true)
• :latin_hypercube - Use Latin Hypercube Sampling for better coverage (default: false)

Example

Optimizer.optimize(suite, params,
  strategy: :random,
  n_trials: 50,
  metric: :score
)

When to Use

Random search is recommended when:

• Search space is large (many parameters or wide ranges)
• Some parameters are more important than others (random search explores all)
• You have limited time/budget for optimization
• Grid search would take too long

Latin Hypercube Sampling

Enable latin_hypercube: true for better coverage of the search space. LHS ensures samples are spread evenly across each parameter's range.