An alternative example to import existing experimental results into Optuna's storage.

quadratic_change_range.py

"""
Optuna example that optimizes a simple quadratic function.

In this example, we demonstrate how to import existing experimental results
and continue the optimization.

We have the following two ways to execute this example:

(1) Execute this code directly.
    $ python quadratic_change_range.py


(2) Execute through CLI.
    $ STUDY_NAME=`optuna create-study --storage sqlite:///example.db`
    $ optuna study optimize quadratic_change_range.py objective --n-trials=100 \
      --study $STUDY_NAME --storage sqlite:///example.db

"""


import optuna


def suggest_all(trial):
    x = trial.suggest_uniform('x', -100, 100)
    y = trial.suggest_int('y', -1, 1)
    return x, y


def suggest_existing_value(trial, x, y):
    _x = trial.suggest_uniform('x', x, x)
    _y = trial.suggest_int('y', y, y)
    return _x, _y


# Define a simple 2-dimensional quadratic function whose minimum value is -1 when (x, y) = (0, -1).
def quadratic(x, y):
    return x ** 2 + y


def objective(trial):
    x, y = suggest_all(trial)
    return quadratic(x, y)


def objective_existing_value(trial, x, y):
    _x, _y = suggest_existing_value(trial, x, y)
    return quadratic(x, y)


if __name__ == '__main__':
    # Let us minimize the objective function above.
    # Caveat 1: use RandomSampler to avoid issue #318.
    study = optuna.create_study(sampler=optuna.samplers.RandomSampler())

    # 1. Run grid search.
    param_value_pairs = []
    step = 40
    for x in range(-100, 100 + step, step):
        for y in [-1, 0, 1]:
            study.optimize(lambda trial: objective_existing_value(trial, x, y),
                           n_trials=1)

    print(study.trials_dataframe())
    print('Best value: {} (params: {})\n'.format(study.best_value, study.best_params))

    # 2. Continue optimization with Optuna.
    # Caveat 2: revert to TPESampler which is Optuna's default sampler.
    study.optimize(objective, n_trials=10)

    print(study.trials_dataframe())
    print('Best value: {} (params: {})\n'.format(study.best_value, study.best_params))