Note

Click here to download the full example code or to run this example in your browser via Binder

Extending Auto-Sklearn with Regression Component

The following example demonstrates how to create a new regression component for using in auto-sklearn.

from typing import Optional
from pprint import pprint

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

import sklearn.metrics

from autosklearn.askl_typing import FEAT_TYPE_TYPE
import autosklearn.regression
import autosklearn.pipeline.components.regression
from autosklearn.pipeline.components.base import AutoSklearnRegressionAlgorithm
from autosklearn.pipeline.constants import (
    SPARSE,
    DENSE,
    SIGNED_DATA,
    UNSIGNED_DATA,
    PREDICTIONS,
)

from sklearn.datasets import load_diabetes
from sklearn.model_selection import train_test_split

Implement kernel ridge regression component for auto-sklearn

class KernelRidgeRegression(AutoSklearnRegressionAlgorithm):
    def __init__(self, alpha, kernel, gamma, degree, coef0, random_state=None):
        self.alpha = alpha
        self.kernel = kernel
        self.gamma = gamma
        self.degree = degree
        self.coef0 = coef0
        self.random_state = random_state
        self.estimator = None

    def fit(self, X, y):
        self.alpha = float(self.alpha)
        self.gamma = float(self.gamma)
        self.degree = int(self.degree)
        self.coef0 = float(self.coef0)

        import sklearn.kernel_ridge

        self.estimator = sklearn.kernel_ridge.KernelRidge(
            alpha=self.alpha,
            kernel=self.kernel,
            gamma=self.gamma,
            degree=self.degree,
            coef0=self.coef0,
        )
        self.estimator.fit(X, y)
        return self

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

    @staticmethod
    def get_properties(dataset_properties=None):
        return {
            "shortname": "KRR",
            "name": "Kernel Ridge Regression",
            "handles_regression": True,
            "handles_classification": False,
            "handles_multiclass": False,
            "handles_multilabel": False,
            "handles_multioutput": True,
            "is_deterministic": True,
            "input": (SPARSE, DENSE, UNSIGNED_DATA, SIGNED_DATA),
            "output": (PREDICTIONS,),
        }

    @staticmethod
    def get_hyperparameter_search_space(
        feat_type: Optional[FEAT_TYPE_TYPE] = None, dataset_properties=None
    ):
        cs = ConfigurationSpace()
        alpha = UniformFloatHyperparameter(
            name="alpha", lower=10**-5, upper=1, log=True, default_value=1.0
        )
        kernel = CategoricalHyperparameter(
            name="kernel",
            # We restrict ourselves to two possible kernels for this example
            choices=["polynomial", "rbf"],
            default_value="polynomial",
        )
        gamma = UniformFloatHyperparameter(
            name="gamma", lower=0.00001, upper=1, default_value=0.1, log=True
        )
        degree = UniformIntegerHyperparameter(
            name="degree", lower=2, upper=5, default_value=3
        )
        coef0 = UniformFloatHyperparameter(
            name="coef0",
            lower=1e-2,
            upper=1e2,
            log=True,
            default_value=1,
        )
        cs.add_hyperparameters([alpha, kernel, gamma, degree, coef0])
        degree_condition = EqualsCondition(degree, kernel, "polynomial")
        coef0_condition = EqualsCondition(coef0, kernel, "polynomial")
        cs.add_conditions([degree_condition, coef0_condition])
        return cs


# Add KRR component to auto-sklearn.
autosklearn.pipeline.components.regression.add_regressor(KernelRidgeRegression)
cs = KernelRidgeRegression.get_hyperparameter_search_space()
print(cs)

Configuration space object:
  Hyperparameters:
    alpha, Type: UniformFloat, Range: [1e-05, 1.0], Default: 1.0, on log-scale
    coef0, Type: UniformFloat, Range: [0.01, 100.0], Default: 1.0, on log-scale
    degree, Type: UniformInteger, Range: [2, 5], Default: 3
    gamma, Type: UniformFloat, Range: [1e-05, 1.0], Default: 0.1, on log-scale
    kernel, Type: Categorical, Choices: {polynomial, rbf}, Default: polynomial
  Conditions:
    coef0 | kernel == 'polynomial'
    degree | kernel == 'polynomial'

Generate data

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

Fit the model using KRR

reg = autosklearn.regression.AutoSklearnRegressor(
    time_left_for_this_task=30,
    per_run_time_limit=10,
    include={"regressor": ["KernelRidgeRegression"]},
    # 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": 5},
)
reg.fit(X_train, y_train)

AutoSklearnRegressor(ensemble_class=<class 'autosklearn.ensembles.ensemble_selection.EnsembleSelection'>,
                     include={'regressor': ['KernelRidgeRegression']},
                     initial_configurations_via_metalearning=0,
                     per_run_time_limit=10,
                     smac_scenario_args={'runcount_limit': 5},
                     time_left_for_this_task=30)