from abc import ABCMeta, abstractmethod
from typing import Any, ClassVar

from numpy import ndarray

from .._typing import ArrayLike, Float, Int, MatrixLike
from ..dummy import DummyClassifier, DummyRegressor
from ..tree._tree import Tree

class LossFunction(metaclass=ABCMeta):
    K: int = ...

    is_multi_class: ClassVar[bool] = ...

    def __init__(self, n_classes: Int) -> None: ...
    @abstractmethod
    def init_estimator(self): ...
    @abstractmethod
    def __call__(
        self,
        y: ArrayLike,
        raw_predictions: MatrixLike,
        sample_weight: None | ArrayLike = None,
    ): ...
    @abstractmethod
    def negative_gradient(self, y: ArrayLike, raw_predictions: MatrixLike, **kargs): ...
    def update_terminal_regions(
        self,
        tree: Tree,
        X: ArrayLike,
        y: ArrayLike,
        residual: ArrayLike,
        raw_predictions: MatrixLike,
        sample_weight: ArrayLike,
        sample_mask: ArrayLike,
        learning_rate: Float = 0.1,
        k: Int = 0,
    ) -> None: ...
    @abstractmethod
    def get_init_raw_predictions(self, X: ArrayLike, estimator: Any) -> ndarray: ...

class RegressionLossFunction(LossFunction, metaclass=ABCMeta):
    def __init__(self) -> None: ...
    def check_init_estimator(self, estimator: Any): ...
    def get_init_raw_predictions(self, X: ArrayLike, estimator: Any) -> ndarray: ...

class LeastSquaresError(RegressionLossFunction):
    def init_estimator(self) -> DummyRegressor: ...
    def __call__(
        self,
        y: ArrayLike,
        raw_predictions: MatrixLike,
        sample_weight: None | ArrayLike = None,
    ) -> Float: ...
    def negative_gradient(self, y: ArrayLike, raw_predictions: ArrayLike, **kargs) -> ndarray: ...
    def update_terminal_regions(
        self,
        tree: Tree,
        X: ArrayLike,
        y: ArrayLike,
        residual: ArrayLike,
        raw_predictions: MatrixLike,
        sample_weight: ArrayLike,
        sample_mask: ArrayLike,
        learning_rate: Float = 0.1,
        k: Int = 0,
    ) -> None: ...

class LeastAbsoluteError(RegressionLossFunction):
    def init_estimator(self): ...
    def __call__(
        self,
        y: ArrayLike,
        raw_predictions: MatrixLike,
        sample_weight: None | ArrayLike = None,
    ): ...
    def negative_gradient(self, y: ArrayLike, raw_predictions: MatrixLike, **kargs): ...

class HuberLossFunction(RegressionLossFunction):
    def __init__(self, alpha: Float = 0.9) -> None: ...
    def init_estimator(self): ...
    def __call__(
        self,
        y: ArrayLike,
        raw_predictions: MatrixLike,
        sample_weight: None | ArrayLike = None,
    ): ...
    def negative_gradient(self, y: ArrayLike, raw_predictions: MatrixLike, sample_weight: None | ArrayLike = None, **kargs): ...

class QuantileLossFunction(RegressionLossFunction):
    def __init__(self, alpha: Float = 0.9) -> None: ...
    def init_estimator(self) -> DummyRegressor: ...
    def __call__(
        self,
        y: ArrayLike,
        raw_predictions: MatrixLike,
        sample_weight: None | ArrayLike = None,
    ) -> Float: ...
    def negative_gradient(self, y: ArrayLike, raw_predictions: MatrixLike, **kargs) -> ndarray: ...

class ClassificationLossFunction(LossFunction, metaclass=ABCMeta):
    def check_init_estimator(self, estimator: Any): ...

class BinomialDeviance(ClassificationLossFunction):
    def __init__(self, n_classes: Int) -> None: ...
    def init_estimator(self) -> DummyClassifier: ...
    def __call__(
        self,
        y: ArrayLike,
        raw_predictions: MatrixLike,
        sample_weight: None | ArrayLike = None,
    ) -> Float: ...
    def negative_gradient(self, y: ArrayLike, raw_predictions: MatrixLike, **kargs) -> ndarray: ...
    def get_init_raw_predictions(self, X: ArrayLike, estimator: Any) -> ndarray: ...

class MultinomialDeviance(ClassificationLossFunction):
    is_multi_class: ClassVar[bool] = ...

    def __init__(self, n_classes: Int) -> None: ...
    def init_estimator(self) -> DummyClassifier: ...
    def __call__(
        self,
        y: ArrayLike,
        raw_predictions: MatrixLike,
        sample_weight: None | ArrayLike = None,
    ) -> Float: ...
    def negative_gradient(self, y: ArrayLike, raw_predictions: MatrixLike, k: Int = 0, **kwargs) -> ndarray: ...
    def get_init_raw_predictions(self, X: ArrayLike, estimator: Any) -> ndarray: ...

class ExponentialLoss(ClassificationLossFunction):
    def __init__(self, n_classes: Int) -> None: ...
    def init_estimator(self): ...
    def __call__(
        self,
        y: ArrayLike,
        raw_predictions: MatrixLike,
        sample_weight: None | ArrayLike = None,
    ): ...
    def negative_gradient(self, y: ArrayLike, raw_predictions: MatrixLike, **kargs): ...
    def get_init_raw_predictions(self, X: ArrayLike, estimator: Any) -> ndarray: ...

LOSS_FUNCTIONS: dict = ...