Agent Hyperparameter Search

Automatic hyperparameter optimization tools for control algorithms. Supports Optuna and Ray Tune.

Quick Start (Optuna)

import numpy as np
from tensoraerospace.optimization import HyperParamOptimizationOptuna

def objective(trial):
    # Example: minimize the quadratic error of a model with hyperparameters
    lr = trial.suggest_float("lr", 1e-4, 1e-1, log=True)
    gamma = trial.suggest_float("gamma", 0.8, 0.999)
    # ... train/evaluate your model ...
    score = np.random.rand()  # replace with the real quality metric
    return score

opt = HyperParamOptimizationOptuna(direction="minimize")
opt.run_optimization(objective, n_trials=20)

best = opt.get_best_param()
print("Best parameters:", best)
opt.plot_parms(figsize=(12, 4))

Classes

OptunaRay Tune

HyperParamOptimizationOptuna(direction)

Bases: HyperParamOptimizationBase

Hyperparameter optimization using Optuna.

Create an Optuna study for hyperparameter optimization.

Parameters:

Name	Type	Description	Default
direction	str	Optimization direction. One of 'minimize' or 'maximize'.	required

Raises:

Type	Description
ValueError	If direction is not supported.

run_optimization(func, n_trials)

Run hyperparameter search.

Parameters:

Name	Type	Description	Default
func	Callable	Objective function to optimize.	required
n_trials	int	Number of trials to run.	required

get_best_param()

Return the best hyperparameters found by Optuna.

Returns:

Name	Type	Description
dict	dict[str, Any]	Best hyperparameters.

plot_parms(figsize=(15.0, 5.0))

Plot trial values over the optimization history.

Parameters:

Name	Type	Description	Default
figsize	tuple[float, float]	Matplotlib figure size. Defaults to (15, 5).	(15.0, 5.0)

Returns:

Type	Description
Figure	matplotlib.figure.Figure: The created figure.

HyperParamOptimizationRay(direction, metric=None)

Bases: HyperParamOptimizationBase

Hyperparameter optimization using Ray Tune.

Create a Ray Tune optimizer wrapper.

Parameters:

Name	Type	Description	Default
direction	str	Optimization direction. One of 'minimize', 'maximize', 'min', 'max'.	required
metric	Optional[str]	Metric name used to choose the best result when Ray Tune supports it.	None

Raises:

Type	Description
ValueError	If direction is not supported.

run_optimization(func, param_space, tune_config=None, **kwargs)

Run a Ray Tune search.

Parameters:

Name	Type	Description	Default
func	Callable	Trainable (callable) to execute. See Ray Tune docs.	required
param_space	Any	Search space definition passed to Ray Tune.	required
tune_config	TuneConfig | None	Optional Ray Tune configuration. Defaults to tune.TuneConfig(num_samples=5).	None
**kwargs	Any	Additional keyword arguments forwarded to tune.Tuner.	{}

get_best_param()

Return the best configuration from the latest Ray Tune run.

Returns:

Name	Type	Description
dict	dict	Best configuration dictionary.

Raises:

Type	Description
RuntimeError	If optimization has not been run yet or best result cannot be determined.

plot_parms(*args, **kwargs)

Plot optimization history (not implemented yet).

Raises:

Type	Description
NotImplementedError	Always.

Notes

• direction: choose "minimize" or "maximize" for the target metric.
• For Ray Tune, use run_optimization(func, param_space, tune_config=...) and then post-process self.results (see the Ray Tune docs).