注意

点击此处下载完整示例代码或通过 Binder 在浏览器中运行此示例

限制现有组件的超参数数量

以下示例演示了如何用一个新组件替换现有组件，该新组件实现了相同的分类器，但使用了不同的超参数。

from typing import Optional

from ConfigSpace.configuration_space import ConfigurationSpace
from ConfigSpace.hyperparameters import (
    UniformIntegerHyperparameter,
    UniformFloatHyperparameter,
)

from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split

from autosklearn.askl_typing import FEAT_TYPE_TYPE
import autosklearn.classification
import autosklearn.pipeline.components.classification
from autosklearn.pipeline.components.classification import (
    AutoSklearnClassificationAlgorithm,
)
from autosklearn.pipeline.constants import DENSE, UNSIGNED_DATA, PREDICTIONS, SPARSE

继承 auto-sklearn 的随机森林分类器

# This classifier only has one of the hyperparameter's of auto-sklearn's
# default parametrization (``max_features``). Instead, it also
# tunes the number of estimators (``n_estimators``).


class CustomRandomForest(AutoSklearnClassificationAlgorithm):
    def __init__(self, n_estimators, max_features, random_state=None):
        self.n_estimators = n_estimators
        self.max_features = max_features
        self.random_state = random_state

    def fit(self, X, y):
        from sklearn.ensemble import RandomForestClassifier

        self.n_estimators = int(self.n_estimators)

        if self.max_features not in ("sqrt", "log2", "auto"):
            max_features = int(X.shape[1] ** float(self.max_features))
        else:
            max_features = self.max_features

        self.estimator = RandomForestClassifier(
            n_estimators=self.n_estimators,
            max_features=max_features,
            random_state=self.random_state,
        )
        self.estimator.fit(X, y)
        return self

    def predict(self, X):
        if self.estimator is None:
            raise NotImplementedError()
        return self.estimator.predict(X)

    def predict_proba(self, X):
        if self.estimator is None:
            raise NotImplementedError()
        return self.estimator.predict_proba(X)

    @staticmethod
    def get_properties(dataset_properties=None):
        return {
            "shortname": "RF",
            "name": "Random Forest Classifier",
            "handles_regression": False,
            "handles_classification": True,
            "handles_multiclass": True,
            "handles_multilabel": True,
            "handles_multioutput": False,
            "is_deterministic": True,
            "input": (DENSE, SPARSE, UNSIGNED_DATA),
            "output": (PREDICTIONS,),
        }

    @staticmethod
    def get_hyperparameter_search_space(
        feat_type: Optional[FEAT_TYPE_TYPE] = None, dataset_properties=None
    ):
        cs = ConfigurationSpace()

        # The maximum number of features used in the forest is calculated as m^max_features, where
        # m is the total number of features, and max_features is the hyperparameter specified below.
        # The default is 0.5, which yields sqrt(m) features as max_features in the estimator. This
        # corresponds with Geurts' heuristic.
        max_features = UniformFloatHyperparameter(
            "max_features", 0.0, 1.0, default_value=0.5
        )
        n_estimators = UniformIntegerHyperparameter(
            "n_estimators", 10, 1000, default_value=100
        )

        cs.add_hyperparameters([max_features, n_estimators])
        return cs


# Add custom random forest classifier component to auto-sklearn.
autosklearn.pipeline.components.classification.add_classifier(CustomRandomForest)
cs = CustomRandomForest.get_hyperparameter_search_space()
print(cs)

Configuration space object:
  Hyperparameters:
    max_features, Type: UniformFloat, Range: [0.0, 1.0], Default: 0.5
    n_estimators, Type: UniformInteger, Range: [10, 1000], Default: 100

数据加载

X, y = load_breast_cancer(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y)

将随机森林分类器拟合到数据

clf = autosklearn.classification.AutoSklearnClassifier(
    time_left_for_this_task=30,
    per_run_time_limit=10,
    # Here we exclude auto-sklearn's default random forest component
    exclude={"classifier": ["random_forest"]},
    # Bellow two flags are provided to speed up calculations
    # Not recommended for a real implementation
    initial_configurations_via_metalearning=0,
    smac_scenario_args={"runcount_limit": 1},
)
clf.fit(X_train, y_train)

AutoSklearnClassifier(ensemble_class=<class 'autosklearn.ensembles.ensemble_selection.EnsembleSelection'>,
                      exclude={'classifier': ['random_forest']},
                      initial_configurations_via_metalearning=0,
                      per_run_time_limit=10,
                      smac_scenario_args={'runcount_limit': 1},
                      time_left_for_this_task=30)