Code source de teachpyx.practice.ml_skl

import numpy as np
from sklearn.base import BaseEstimator, RegressorMixin




class PositiveOrNegativeLinearRegression(RegressorMixin, BaseEstimator):
    """
    Trains a linear regression with coefficients of the same sign.
    The order of inheritance must be ``RegressorMixin, BaseEstimator``
    otherwise the tags are wrong.

    :param epsilon: gradient step
    :param max_iter: number maximum of iterations
    :param positive: only positive weights (or negative if False)

    Tags can be changed.

    .. code-block:: python

        def __sklearn_tags__(self):
            tags = RegressorMixin.__sklearn_tags__(self)
            return tags

            return Tags(
                estimator_type=None,
                target_tags=TargetTags(required=False),
                transformer_tags=None,
                regressor_tags=None,
                classifier_tags=None,
            )

    Or:

    .. code-block:: python

        def __sklearn_tags__(self):
            return Tags(
                estimator_type="regressor",
                classifier_tags=None,
                regressor_tags=RegressorTags(),
                transformer_tags=None,
                target_tags=TargetTags(required=True),
            )
    """

    def __init__(
        self, epsilon: float = 1.0, max_iter: int = 100, positive: bool = True
    ):
        super().__init__()
        self.epsilon = epsilon
        self.max_iter = max_iter
        self.positive = positive



    def fit(self, X, y):
        "Trains."
        theta = np.random.randn(X.shape[1])
        for i in range(self.max_iter):
            epsilon = self.epsilon * 10 / (i + 10)
            grad = X.T @ (X @ theta - y)
            theta = theta - epsilon * grad / X.shape[0]
            theta = np.maximum(theta, 0) if self.positive else np.minimum(theta, 0)
            if np.abs(theta).max() == 0:
                # If all coefficients are null. The algorithm may be stuck.
                theta = np.random.randn(X.shape[1]) * epsilon
        self.theta_ = theta
        return self




    def predict(self, X):
        "Predicts."
        return X @ self.theta_