Module hyperparameter_tuner 

Hyperparameter Tuner — Bayesian optimization and random/grid search.

Provides HyperparameterTuner supporting:

• Random Search — sample random configurations with xorshift64 PRNG
• Grid Search — enumerate all discrete/categorical/continuous combinations
• Bayesian Optimization — Gaussian process surrogate with UCB acquisition

Example

use ipfrs_tensorlogic::{
    HyperparameterTuner, TunerConfig, HpSpec, HpType, TuningStrategy,
};

let spec = HpSpec {
    name: "lr".to_string(),
    hp_type: HpType::Continuous { lo: 1e-4, hi: 1e-1 },
    log_scale: true,
};
let config = TunerConfig {
    specs: vec![spec],
    maximize: false,
    seed: 42,
};
let mut tuner = HyperparameterTuner::new(config);
let mut rng = 42u64;
let results = tuner.run_random_search(5, |cfg| {
    // Simulated scorer: extract lr, return a dummy loss
    use ipfrs_tensorlogic::HpValue;
    if let Some(HpValue::Float(lr)) = cfg.get("lr") {
        lr.abs()
    } else {
        f64::MAX
    }
}, &mut rng);
assert_eq!(results.len(), 5);

Structs

HpConfig

A complete hyperparameter configuration (map from name to value).

HpSpec

Specification for a single hyperparameter.

HyperparameterTuner

Hyperparameter tuner supporting random search, grid search, and simplified Bayesian optimization with UCB acquisition.

TunerConfig

Configuration for HyperparameterTuner.

TunerStats

Aggregate statistics over all recorded trials.

TuningResult

The result of a single hyperparameter tuning trial.

Enums

HpTunerError

Errors that can arise during hyperparameter tuning.

HpType

The domain type of a single hyperparameter.

HpValue

The value of a single hyperparameter in a concrete configuration.

TuningStrategy

How the tuner explores the hyperparameter space.