Optuna - Hyperparameter Optimization

Jan 27, 2026
python , machine learning , optimization

Optuna is a Python library for automated parameter optimization using adaptive search instead of manual tuning.

It shines whenever:

You have multiple knobs to tune.
The system returns one final numeric score.

This pattern shows up everywhere: machine learning metrics, simulation outcomes, and trading strategy ROI.

What Is a Hyperparameter?

A hyperparameter is a setting you choose before running an evaluation.

Examples:

A model’s learning rate or tree depth.
A trading rule’s lookback window or threshold.
A simulator’s step size or penalty weight.

Unlike learned parameters (like model weights), hyperparameters are not fit directly by gradient descent. You set them, run the system, and observe a final score.

Optuna automates this loop:

Propose a set of hyperparameters.
Run the evaluation.
Record the score.
Use the results to propose better values next time.

Multi-Parameter Systems With One Score

Optuna works especially well for systems that require multiple parameters but produce a single final score.

All you need is an objective function that:

Accepts a trial.
Suggests several parameters.
Returns one numeric score to maximize or minimize.

Minimal Example

import optuna

def objective(trial):
    threshold = trial.suggest_float("threshold", 0.0, 1.0)
    window = trial.suggest_int("window", 5, 50)

    score = evaluate_model(threshold=threshold, window=window)
    return score

study = optuna.create_study(direction="maximize")
study.optimize(objective, n_trials=100)

print(study.best_params)

Example: Trading Strategy ROI Over a Time Series

Trading strategies often depend on several parameters but ultimately produce one score, such as ROI, Sharpe ratio, or drawdown-adjusted return.

Optuna is a great fit for this pattern:

import optuna

def backtest_strategy(prices, window, entry_z, exit_z, fee_bps):
    # Placeholder: implement your backtest here.
    # Return a single score such as ROI.
    return roi_over_time_series(
        prices=prices,
        window=window,
        entry_z=entry_z,
        exit_z=exit_z,
        fee_bps=fee_bps,
    )

def objective(trial):
    window = trial.suggest_int("window", 10, 200)
    entry_z = trial.suggest_float("entry_z", 0.5, 3.0)
    exit_z = trial.suggest_float("exit_z", 0.1, 2.0)
    fee_bps = trial.suggest_float("fee_bps", 0.0, 25.0)

    roi = backtest_strategy(
        prices=load_prices(),
        window=window,
        entry_z=entry_z,
        exit_z=exit_z,
        fee_bps=fee_bps,
    )
    return roi

study = optuna.create_study(direction="maximize")
study.optimize(objective, n_trials=200)

print("Best ROI:", study.best_value)
print("Best params:", study.best_params)