带目标权重的二维 Schaffer 函数

展开复制 examples/3_multi_objective/1_schaffer.py (右上角)

from __future__ import annotations

from typing import Dict, Tuple

import numpy as np
from ConfigSpace import Configuration, ConfigurationSpace
from matplotlib import pyplot as plt

from smac import HyperparameterOptimizationFacade as HPOFacade
from smac import Scenario
from smac.facade import AbstractFacade

__copyright__ = "Copyright 2025, Leibniz University Hanover, Institute of AI"
__license__ = "3-clause BSD"


def schaffer(x: float) -> Tuple[float, float]:
    f1 = np.square(x)
    f2 = np.square(np.sqrt(f1) - 2)

    return f1, f2


def target_function(config: Configuration, seed: int = 0) -> Dict[str, float]:
    f1, f2 = schaffer(config["x"])
    return {"metric1": f1, "metric2": f2}


def plot_from_smac(smac: AbstractFacade) -> None:
    plt.figure()
    configs = smac.runhistory.get_configs()
    incumbents = smac.intensifier.get_incumbents()

    for i, config in enumerate(configs):
        if config in incumbents:
            continue

        label = None
        if i == 0:
            label = "Configuration"

        x = config["x"]
        f1, f2 = schaffer(x)
        plt.scatter(f1, f2, c="blue", alpha=0.1, marker="o", zorder=3000, label=label)

    for i, config in enumerate(incumbents):
        label = None
        if i == 0:
            label = "Incumbent"

        x = config["x"]
        f1, f2 = schaffer(x)
        plt.scatter(f1, f2, c="red", alpha=1, marker="x", zorder=3000, label=label)

    plt.xlabel("f1")
    plt.ylabel("f2")
    plt.title("Schaffer 2D")
    plt.legend()

    plt.show()


if __name__ == "__main__":
    # Simple configspace
    cs = ConfigurationSpace({"x": (-2.0, 2.0)})

    # Scenario object
    scenario = Scenario(
        configspace=cs,
        deterministic=True,  # Only one seed
        n_trials=150,
        objectives=["metric1", "metric2"],
    )

    smac = HPOFacade(
        scenario=scenario,
        target_function=target_function,
        multi_objective_algorithm=HPOFacade.get_multi_objective_algorithm(
            scenario,
            objective_weights=[1, 2],  # Weight metric2 twice as much as metric1
        ),
        overwrite=True,
    )
    incumbents = smac.optimize()

    # Get cost of default configuration
    default_cost = smac.validate(cs.get_default_configuration())
    print(f"Validated costs from default config: \n--- {default_cost}\n")

    print("Validated costs from the Pareto front (incumbents):")
    for incumbent in incumbents:
        cost = smac.validate(incumbent)
        print("---", cost)

    # Plot the evaluated points
    plot_from_smac(smac)

描述

这里展示了一个如何使用多目标优化的简单示例。使用了二维 Schaffer 函数。在图中，你可以看到所有点都在帕累托前沿上。然而，由于我们设置了目标权重，你可以注意到 SMAC 优先优化第二个目标而不是第一个。

代码运行

from __future__ import annotations

from typing import Dict, Tuple

import numpy as np
from ConfigSpace import Configuration, ConfigurationSpace
from matplotlib import pyplot as plt

from smac import HyperparameterOptimizationFacade as HPOFacade
from smac import Scenario
from smac.facade import AbstractFacade

__copyright__ = "Copyright 2025, Leibniz University Hanover, Institute of AI"
__license__ = "3-clause BSD"


def schaffer(x: float) -> Tuple[float, float]:
    f1 = np.square(x)
    f2 = np.square(np.sqrt(f1) - 2)

    return f1, f2


def target_function(config: Configuration, seed: int = 0) -> Dict[str, float]:
    f1, f2 = schaffer(config["x"])
    return {"metric1": f1, "metric2": f2}


def plot_from_smac(smac: AbstractFacade) -> None:
    plt.figure()
    configs = smac.runhistory.get_configs()
    incumbents = smac.intensifier.get_incumbents()

    for i, config in enumerate(configs):
        if config in incumbents:
            continue

        label = None
        if i == 0:
            label = "Configuration"

        x = config["x"]
        f1, f2 = schaffer(x)
        plt.scatter(f1, f2, c="blue", alpha=0.1, marker="o", zorder=3000, label=label)

    for i, config in enumerate(incumbents):
        label = None
        if i == 0:
            label = "Incumbent"

        x = config["x"]
        f1, f2 = schaffer(x)
        plt.scatter(f1, f2, c="red", alpha=1, marker="x", zorder=3000, label=label)

    plt.xlabel("f1")
    plt.ylabel("f2")
    plt.title("Schaffer 2D")
    plt.legend()

    plt.show()


if __name__ == "__main__":
    # Simple configspace
    cs = ConfigurationSpace({"x": (-2.0, 2.0)})

    # Scenario object
    scenario = Scenario(
        configspace=cs,
        deterministic=True,  # Only one seed
        n_trials=150,
        objectives=["metric1", "metric2"],
    )

    smac = HPOFacade(
        scenario=scenario,
        target_function=target_function,
        multi_objective_algorithm=HPOFacade.get_multi_objective_algorithm(
            scenario,
            objective_weights=[1, 2],  # Weight metric2 twice as much as metric1
        ),
        overwrite=True,
    )
    incumbents = smac.optimize()

    # Get cost of default configuration
    default_cost = smac.validate(cs.get_default_configuration())
    print(f"Validated costs from default config: \n--- {default_cost}\n")

    print("Validated costs from the Pareto front (incumbents):")
    for incumbent in incumbents:
        cost = smac.validate(incumbent)
        print("---", cost)

    # Plot the evaluated points
    plot_from_smac(smac)