sklearn/ensemble/_gb.py

"""Gradient Boosted Regression Trees.

This module contains methods for fitting gradient boosted regression trees for
both classification and regression.

The module structure is the following:

- The ``BaseGradientBoosting`` base class implements a common ``fit`` method
  for all the estimators in the module. Regression and classification
  only differ in the concrete ``LossFunction`` used.

- ``GradientBoostingClassifier`` implements gradient boosting for
  classification problems.

- ``GradientBoostingRegressor`` implements gradient boosting for
  regression problems.
"""

# Authors: Peter Prettenhofer, Scott White, Gilles Louppe, Emanuele Olivetti,
#          Arnaud Joly, Jacob Schreiber
# License: BSD 3 clause

import math
import warnings
from abc import ABCMeta, abstractmethod
from numbers import Integral, Real
from time import time

import numpy as np
from scipy.sparse import csc_matrix, csr_matrix, issparse

from .._loss.loss import (
    _LOSSES,
    AbsoluteError,
    ExponentialLoss,
    HalfBinomialLoss,
    HalfMultinomialLoss,
    HalfSquaredError,
    HuberLoss,
    PinballLoss,
)
from ..base import ClassifierMixin, RegressorMixin, _fit_context, is_classifier
from ..dummy import DummyClassifier, DummyRegressor
from ..exceptions import NotFittedError
from ..model_selection import train_test_split
from ..preprocessing import LabelEncoder
from ..tree import DecisionTreeRegressor
from ..tree._tree import DOUBLE, DTYPE, TREE_LEAF
from ..utils import check_array, check_random_state, column_or_1d
from ..utils._param_validation import HasMethods, Interval, StrOptions
from ..utils.multiclass import check_classification_targets
from ..utils.stats import _weighted_percentile
from ..utils.validation import _check_sample_weight, check_is_fitted
from ._base import BaseEnsemble
from ._gradient_boosting import _random_sample_mask, predict_stage, predict_stages

_LOSSES = _LOSSES.copy()
_LOSSES.update(
    {
        "quantile": PinballLoss,
        "huber": HuberLoss,
    }
)


def _safe_divide(numerator, denominator):
    """Prevents overflow and division by zero."""
    # This is used for classifiers where the denominator might become zero exatly.
    # For instance for log loss, HalfBinomialLoss, if proba=0 or proba=1 exactly, then
    # denominator = hessian = 0, and we should set the node value in the line search to
    # zero as there is no improvement of the loss possible.
    # For numerical safety, we do this already for extremely tiny values.
    if abs(denominator) < 1e-150:
        return 0.0
    else:
        # Cast to Python float to trigger Python errors, e.g. ZeroDivisionError,
        # without relying on `np.errstate` that is not supported by Pyodide.
        result = float(numerator) / float(denominator)
        # Cast to Python float to trigger a ZeroDivisionError without relying
        # on `np.errstate` that is not supported by Pyodide.
        result = float(numerator) / float(denominator)
        if math.isinf(result):
            warnings.warn("overflow encountered in _safe_divide", RuntimeWarning)
        return result


def _init_raw_predictions(X, estimator, loss, use_predict_proba):
    """Return the initial raw predictions.

    Parameters
    ----------
    X : ndarray of shape (n_samples, n_features)
        The data array.
    estimator : object
        The estimator to use to compute the predictions.
    loss : BaseLoss
        An instance of a loss function class.
    use_predict_proba : bool
        Whether estimator.predict_proba is used instead of estimator.predict.

    Returns
    -------
    raw_predictions : ndarray of shape (n_samples, K)
        The initial raw predictions. K is equal to 1 for binary
        classification and regression, and equal to the number of classes
        for multiclass classification. ``raw_predictions`` is casted
        into float64.
    """
    # TODO: Use loss.fit_intercept_only where appropriate instead of
    # DummyRegressor which is the default given by the `init` parameter,
    # see also _init_state.
    if use_predict_proba:
        # Our parameter validation, set via _fit_context and _parameter_constraints
        # already guarantees that estimator has a predict_proba method.
        predictions = estimator.predict_proba(X)
        if not loss.is_multiclass:
            predictions = predictions[:, 1]  # probability of positive class
        eps = np.finfo(np.float32).eps  # FIXME: This is quite large!
        predictions = np.clip(predictions, eps, 1 - eps, dtype=np.float64)
    else:
        predictions = estimator.predict(X).astype(np.float64)

    if predictions.ndim == 1:
        return loss.link.link(predictions).reshape(-1, 1)
    else:
        return loss.link.link(predictions)


def _update_terminal_regions(
    loss,
    tree,
    X,
    y,
    neg_gradient,
    raw_prediction,
    sample_weight,
    sample_mask,
    learning_rate=0.1,
    k=0,
):
    """Update the leaf values to be predicted by the tree and raw_prediction.

    The current raw predictions of the model (of this stage) are updated.

    Additionally, the terminal regions (=leaves) of the given tree are updated as well.
    This corresponds to the line search step in "Greedy Function Approximation" by
    Friedman, Algorithm 1 step 5.

    Update equals:
        argmin_{x} loss(y_true, raw_prediction_old + x * tree.value)

    For non-trivial cases like the Binomial loss, the update has no closed formula and
    is an approximation, again, see the Friedman paper.

    Also note that the update formula for the SquaredError is the identity. Therefore,
    in this case, the leaf values don't need an update and only the raw_predictions are
    updated (with the learning rate included).

    Parameters
    ----------
    loss : BaseLoss
    tree : tree.Tree
        The tree object.
    X : ndarray of shape (n_samples, n_features)
        The data array.
    y : ndarray of shape (n_samples,)
        The target labels.
    neg_gradient : ndarray of shape (n_samples,)
        The negative gradient.
    raw_prediction : ndarray of shape (n_samples, n_trees_per_iteration)
        The raw predictions (i.e. values from the tree leaves) of the
        tree ensemble at iteration ``i - 1``.
    sample_weight : ndarray of shape (n_samples,)
        The weight of each sample.
    sample_mask : ndarray of shape (n_samples,)
        The sample mask to be used.
    learning_rate : float, default=0.1
        Learning rate shrinks the contribution of each tree by
         ``learning_rate``.
    k : int, default=0
        The index of the estimator being updated.
    """
    # compute leaf for each sample in ``X``.
    terminal_regions = tree.apply(X)

    if not isinstance(loss, HalfSquaredError):
        # mask all which are not in sample mask.
        masked_terminal_regions = terminal_regions.copy()
        masked_terminal_regions[~sample_mask] = -1

        if isinstance(loss, HalfBinomialLoss):

            def compute_update(y_, indices, neg_gradient, raw_prediction, k):
                # Make a single Newton-Raphson step, see "Additive Logistic Regression:
                # A Statistical View of Boosting" FHT00 and note that we use a slightly
                # different version (factor 2) of "F" with proba=expit(raw_prediction).
                # Our node estimate is given by:
                #    sum(w * (y - prob)) / sum(w * prob * (1 - prob))
                # we take advantage that: y - prob = neg_gradient
                neg_g = neg_gradient.take(indices, axis=0)
                prob = y_ - neg_g
                # numerator = negative gradient = y - prob
                numerator = np.average(neg_g, weights=sw)
                # denominator = hessian = prob * (1 - prob)
                denominator = np.average(prob * (1 - prob), weights=sw)
                return _safe_divide(numerator, denominator)

        elif isinstance(loss, HalfMultinomialLoss):

            def compute_update(y_, indices, neg_gradient, raw_prediction, k):
                # we take advantage that: y - prob = neg_gradient
                neg_g = neg_gradient.take(indices, axis=0)
                prob = y_ - neg_g
                K = loss.n_classes
                # numerator = negative gradient * (k - 1) / k
                # Note: The factor (k - 1)/k appears in the original papers "Greedy
                # Function Approximation" by Friedman and "Additive Logistic
                # Regression" by Friedman, Hastie, Tibshirani. This factor is, however,
                # wrong or at least arbitrary as it directly multiplies the
                # learning_rate. We keep it for backward compatibility.
                numerator = np.average(neg_g, weights=sw)
                numerator *= (K - 1) / K
                # denominator = (diagonal) hessian = prob * (1 - prob)
                denominator = np.average(prob * (1 - prob), weights=sw)
                return _safe_divide(numerator, denominator)

        elif isinstance(loss, ExponentialLoss):

            def compute_update(y_, indices, neg_gradient, raw_prediction, k):
                neg_g = neg_gradient.take(indices, axis=0)
                # numerator = negative gradient = y * exp(-raw) - (1-y) * exp(raw)
                numerator = np.average(neg_g, weights=sw)
                # denominator = hessian = y * exp(-raw) + (1-y) * exp(raw)
                # if y=0: hessian = exp(raw) = -neg_g
                #    y=1: hessian = exp(-raw) = neg_g
                hessian = neg_g.copy()
                hessian[y_ == 0] *= -1
                denominator = np.average(hessian, weights=sw)
                return _safe_divide(numerator, denominator)

        else:

            def compute_update(y_, indices, neg_gradient, raw_prediction, k):
                return loss.fit_intercept_only(
                    y_true=y_ - raw_prediction[indices, k],
                    sample_weight=sw,
                )

        # update each leaf (= perform line search)
        for leaf in np.nonzero(tree.children_left == TREE_LEAF)[0]:
            indices = np.nonzero(masked_terminal_regions == leaf)[
                0
            ]  # of terminal regions
            y_ = y.take(indices, axis=0)
            sw = None if sample_weight is None else sample_weight[indices]
            update = compute_update(y_, indices, neg_gradient, raw_prediction, k)

            # TODO: Multiply here by learning rate instead of everywhere else.
            tree.value[leaf, 0, 0] = update

    # update predictions (both in-bag and out-of-bag)
    raw_prediction[:, k] += learning_rate * tree.value[:, 0, 0].take(
        terminal_regions, axis=0
    )


def set_huber_delta(loss, y_true, raw_prediction, sample_weight=None):
    """Calculate and set self.closs.delta based on self.quantile."""
    abserr = np.abs(y_true - raw_prediction.squeeze())
    # sample_weight is always a ndarray, never None.
    delta = _weighted_percentile(abserr, sample_weight, 100 * loss.quantile)
    loss.closs.delta = float(delta)


class VerboseReporter:
    """Reports verbose output to stdout.

    Parameters
    ----------
    verbose : int
        Verbosity level. If ``verbose==1`` output is printed once in a while
        (when iteration mod verbose_mod is zero).; if larger than 1 then output
        is printed for each update.
    """

    def __init__(self, verbose):
        self.verbose = verbose

    def init(self, est, begin_at_stage=0):
        """Initialize reporter

        Parameters
        ----------
        est : Estimator
            The estimator

        begin_at_stage : int, default=0
            stage at which to begin reporting
        """
        # header fields and line format str
        header_fields = ["Iter", "Train Loss"]
        verbose_fmt = ["{iter:>10d}", "{train_score:>16.4f}"]
        # do oob?
        if est.subsample < 1:
            header_fields.append("OOB Improve")
            verbose_fmt.append("{oob_impr:>16.4f}")
        header_fields.append("Remaining Time")
        verbose_fmt.append("{remaining_time:>16s}")

        # print the header line
        print(("%10s " + "%16s " * (len(header_fields) - 1)) % tuple(header_fields))

        self.verbose_fmt = " ".join(verbose_fmt)
        # plot verbose info each time i % verbose_mod == 0
        self.verbose_mod = 1
        self.start_time = time()
        self.begin_at_stage = begin_at_stage

    def update(self, j, est):
        """Update reporter with new iteration.

        Parameters
        ----------
        j : int
            The new iteration.
        est : Estimator
            The estimator.
        """
        do_oob = est.subsample < 1
        # we need to take into account if we fit additional estimators.
        i = j - self.begin_at_stage  # iteration relative to the start iter
        if (i + 1) % self.verbose_mod == 0:
            oob_impr = est.oob_improvement_[j] if do_oob else 0
            remaining_time = (
                (est.n_estimators - (j + 1)) * (time() - self.start_time) / float(i + 1)
            )
            if remaining_time > 60:
                remaining_time = "{0:.2f}m".format(remaining_time / 60.0)
            else:
                remaining_time = "{0:.2f}s".format(remaining_time)
            print(
                self.verbose_fmt.format(
                    iter=j + 1,
                    train_score=est.train_score_[j],
                    oob_impr=oob_impr,
                    remaining_time=remaining_time,
                )
            )
            if self.verbose == 1 and ((i + 1) // (self.verbose_mod * 10) > 0):
                # adjust verbose frequency (powers of 10)
                self.verbose_mod *= 10


class BaseGradientBoosting(BaseEnsemble, metaclass=ABCMeta):
    """Abstract base class for Gradient Boosting."""

    _parameter_constraints: dict = {
        **DecisionTreeRegressor._parameter_constraints,
        "learning_rate": [Interval(Real, 0.0, None, closed="left")],
        "n_estimators": [Interval(Integral, 1, None, closed="left")],
        "criterion": [StrOptions({"friedman_mse", "squared_error"})],
        "subsample": [Interval(Real, 0.0, 1.0, closed="right")],
        "verbose": ["verbose"],
        "warm_start": ["boolean"],
        "validation_fraction": [Interval(Real, 0.0, 1.0, closed="neither")],
        "n_iter_no_change": [Interval(Integral, 1, None, closed="left"), None],
        "tol": [Interval(Real, 0.0, None, closed="left")],
    }
    _parameter_constraints.pop("splitter")
    _parameter_constraints.pop("monotonic_cst")

    @abstractmethod
    def __init__(
        self,
        *,
        loss,
        learning_rate,
        n_estimators,
        criterion,
        min_samples_split,
        min_samples_leaf,
        min_weight_fraction_leaf,
        max_depth,
        min_impurity_decrease,
        init,
        subsample,
        max_features,
        ccp_alpha,
        random_state,
        alpha=0.9,
        verbose=0,
        max_leaf_nodes=None,
        warm_start=False,
        validation_fraction=0.1,
        n_iter_no_change=None,
        tol=1e-4,
    ):
        self.n_estimators = n_estimators
        self.learning_rate = learning_rate
        self.loss = loss
        self.criterion = criterion
        self.min_samples_split = min_samples_split
        self.min_samples_leaf = min_samples_leaf
        self.min_weight_fraction_leaf = min_weight_fraction_leaf
        self.subsample = subsample
        self.max_features = max_features
        self.max_depth = max_depth
        self.min_impurity_decrease = min_impurity_decrease
        self.ccp_alpha = ccp_alpha
        self.init = init
        self.random_state = random_state
        self.alpha = alpha
        self.verbose = verbose
        self.max_leaf_nodes = max_leaf_nodes
        self.warm_start = warm_start
        self.validation_fraction = validation_fraction
        self.n_iter_no_change = n_iter_no_change
        self.tol = tol

    @abstractmethod
    def _encode_y(self, y=None, sample_weight=None):
        """Called by fit to validate and encode y."""

    @abstractmethod
    def _get_loss(self, sample_weight):
        """Get loss object from sklearn._loss.loss."""

    def _fit_stage(
        self,
        i,
        X,
        y,
        raw_predictions,
        sample_weight,
        sample_mask,
        random_state,
        X_csc=None,
        X_csr=None,
    ):
        """Fit another stage of ``n_trees_per_iteration_`` trees."""
        original_y = y

        if isinstance(self._loss, HuberLoss):
            set_huber_delta(
                loss=self._loss,
                y_true=y,
                raw_prediction=raw_predictions,
                sample_weight=sample_weight,
            )
        # TODO: Without oob, i.e. with self.subsample = 1.0, we could call
        # self._loss.loss_gradient and use it to set train_score_.
        # But note that train_score_[i] is the score AFTER fitting the i-th tree.
        # Note: We need the negative gradient!
        neg_gradient = -self._loss.gradient(
            y_true=y,
            raw_prediction=raw_predictions,
            sample_weight=None,  # We pass sample_weights to the tree directly.
        )
        # 2-d views of shape (n_samples, n_trees_per_iteration_) or (n_samples, 1)
        # on neg_gradient to simplify the loop over n_trees_per_iteration_.
        if neg_gradient.ndim == 1:
            neg_g_view = neg_gradient.reshape((-1, 1))
        else:
            neg_g_view = neg_gradient

        for k in range(self.n_trees_per_iteration_):
            if self._loss.is_multiclass:
                y = np.array(original_y == k, dtype=np.float64)

            # induce regression tree on the negative gradient
            tree = DecisionTreeRegressor(
                criterion=self.criterion,
                splitter="best",
                max_depth=self.max_depth,
                min_samples_split=self.min_samples_split,
                min_samples_leaf=self.min_samples_leaf,
                min_weight_fraction_leaf=self.min_weight_fraction_leaf,
                min_impurity_decrease=self.min_impurity_decrease,
                max_features=self.max_features,
                max_leaf_nodes=self.max_leaf_nodes,
                random_state=random_state,
                ccp_alpha=self.ccp_alpha,
            )

            if self.subsample < 1.0:
                # no inplace multiplication!
                sample_weight = sample_weight * sample_mask.astype(np.float64)

            X = X_csc if X_csc is not None else X
            tree.fit(
                X, neg_g_view[:, k], sample_weight=sample_weight, check_input=False
            )

            # update tree leaves
            X_for_tree_update = X_csr if X_csr is not None else X
            _update_terminal_regions(
                self._loss,
                tree.tree_,
                X_for_tree_update,
                y,
                neg_g_view[:, k],
                raw_predictions,
                sample_weight,
                sample_mask,
                learning_rate=self.learning_rate,
                k=k,
            )

            # add tree to ensemble
            self.estimators_[i, k] = tree

        return raw_predictions

    def _set_max_features(self):
        """Set self.max_features_."""
        if isinstance(self.max_features, str):
            if self.max_features == "auto":
                if is_classifier(self):
                    max_features = max(1, int(np.sqrt(self.n_features_in_)))
                else:
                    max_features = self.n_features_in_
            elif self.max_features == "sqrt":
                max_features = max(1, int(np.sqrt(self.n_features_in_)))
            else:  # self.max_features == "log2"
                max_features = max(1, int(np.log2(self.n_features_in_)))
        elif self.max_features is None:
            max_features = self.n_features_in_
        elif isinstance(self.max_features, Integral):
            max_features = self.max_features
        else:  # float
            max_features = max(1, int(self.max_features * self.n_features_in_))

        self.max_features_ = max_features

    def _init_state(self):
        """Initialize model state and allocate model state data structures."""

        self.init_ = self.init
        if self.init_ is None:
            if is_classifier(self):
                self.init_ = DummyClassifier(strategy="prior")
            elif isinstance(self._loss, (AbsoluteError, HuberLoss)):
                self.init_ = DummyRegressor(strategy="quantile", quantile=0.5)
            elif isinstance(self._loss, PinballLoss):
                self.init_ = DummyRegressor(strategy="quantile", quantile=self.alpha)
            else:
                self.init_ = DummyRegressor(strategy="mean")

        self.estimators_ = np.empty(
            (self.n_estimators, self.n_trees_per_iteration_), dtype=object
        )
        self.train_score_ = np.zeros((self.n_estimators,), dtype=np.float64)
        # do oob?
        if self.subsample < 1.0:
            self.oob_improvement_ = np.zeros((self.n_estimators), dtype=np.float64)
            self.oob_scores_ = np.zeros((self.n_estimators), dtype=np.float64)
            self.oob_score_ = np.nan

    def _clear_state(self):
        """Clear the state of the gradient boosting model."""
        if hasattr(self, "estimators_"):
            self.estimators_ = np.empty((0, 0), dtype=object)
        if hasattr(self, "train_score_"):
            del self.train_score_
        if hasattr(self, "oob_improvement_"):
            del self.oob_improvement_
        if hasattr(self, "oob_scores_"):
            del self.oob_scores_
        if hasattr(self, "oob_score_"):
            del self.oob_score_
        if hasattr(self, "init_"):
            del self.init_
        if hasattr(self, "_rng"):
            del self._rng

    def _resize_state(self):
        """Add additional ``n_estimators`` entries to all attributes."""
        # self.n_estimators is the number of additional est to fit
        total_n_estimators = self.n_estimators
        if total_n_estimators < self.estimators_.shape[0]:
            raise ValueError(
                "resize with smaller n_estimators %d < %d"
                % (total_n_estimators, self.estimators_[0])
            )

        self.estimators_ = np.resize(
            self.estimators_, (total_n_estimators, self.n_trees_per_iteration_)
        )
        self.train_score_ = np.resize(self.train_score_, total_n_estimators)
        if self.subsample < 1 or hasattr(self, "oob_improvement_"):
            # if do oob resize arrays or create new if not available
            if hasattr(self, "oob_improvement_"):
                self.oob_improvement_ = np.resize(
                    self.oob_improvement_, total_n_estimators
                )
                self.oob_scores_ = np.resize(self.oob_scores_, total_n_estimators)
                self.oob_score_ = np.nan
            else:
                self.oob_improvement_ = np.zeros(
                    (total_n_estimators,), dtype=np.float64
                )
                self.oob_scores_ = np.zeros((total_n_estimators,), dtype=np.float64)
                self.oob_score_ = np.nan

    def _is_fitted(self):
        return len(getattr(self, "estimators_", [])) > 0

    def _check_initialized(self):
        """Check that the estimator is initialized, raising an error if not."""
        check_is_fitted(self)

    @_fit_context(
        # GradientBoosting*.init is not validated yet
        prefer_skip_nested_validation=False
    )
    def fit(self, X, y, sample_weight=None, monitor=None):
        """Fit the gradient boosting model.

        Parameters
        ----------
        X : {array-like, sparse matrix} of shape (n_samples, n_features)
            The input samples. Internally, it will be converted to
            ``dtype=np.float32`` and if a sparse matrix is provided
            to a sparse ``csr_matrix``.

        y : array-like of shape (n_samples,)
            Target values (strings or integers in classification, real numbers
            in regression)
            For classification, labels must correspond to classes.

        sample_weight : array-like of shape (n_samples,), default=None
            Sample weights. If None, then samples are equally weighted. Splits
            that would create child nodes with net zero or negative weight are
            ignored while searching for a split in each node. In the case of
            classification, splits are also ignored if they would result in any
            single class carrying a negative weight in either child node.

        monitor : callable, default=None
            The monitor is called after each iteration with the current
            iteration, a reference to the estimator and the local variables of
            ``_fit_stages`` as keyword arguments ``callable(i, self,
            locals())``. If the callable returns ``True`` the fitting procedure
            is stopped. The monitor can be used for various things such as
            computing held-out estimates, early stopping, model introspect, and
            snapshotting.

        Returns
        -------
        self : object
            Fitted estimator.
        """
        if not self.warm_start:
            self._clear_state()

        # Check input
        # Since check_array converts both X and y to the same dtype, but the
        # trees use different types for X and y, checking them separately.

        X, y = self._validate_data(
            X, y, accept_sparse=["csr", "csc", "coo"], dtype=DTYPE, multi_output=True
        )
        sample_weight_is_none = sample_weight is None
        sample_weight = _check_sample_weight(sample_weight, X)
        if sample_weight_is_none:
            y = self._encode_y(y=y, sample_weight=None)
        else:
            y = self._encode_y(y=y, sample_weight=sample_weight)
        y = column_or_1d(y, warn=True)  # TODO: Is this still required?

        self._set_max_features()

        # self.loss is guaranteed to be a string
        self._loss = self._get_loss(sample_weight=sample_weight)

        if self.n_iter_no_change is not None:
            stratify = y if is_classifier(self) else None
            (
                X_train,
                X_val,
                y_train,
                y_val,
                sample_weight_train,
                sample_weight_val,
            ) = train_test_split(
                X,
                y,
                sample_weight,
                random_state=self.random_state,
                test_size=self.validation_fraction,
                stratify=stratify,
            )
            if is_classifier(self):
                if self.n_classes_ != np.unique(y_train).shape[0]:
                    # We choose to error here. The problem is that the init
                    # estimator would be trained on y, which has some missing
                    # classes now, so its predictions would not have the
                    # correct shape.
                    raise ValueError(
                        "The training data after the early stopping split "
                        "is missing some classes. Try using another random "
                        "seed."
                    )
        else:
            X_train, y_train, sample_weight_train = X, y, sample_weight
            X_val = y_val = sample_weight_val = None

        n_samples = X_train.shape[0]

        # First time calling fit.
        if not self._is_fitted():
            # init state
            self._init_state()

            # fit initial model and initialize raw predictions
            if self.init_ == "zero":
                raw_predictions = np.zeros(
                    shape=(n_samples, self.n_trees_per_iteration_),
                    dtype=np.float64,
                )
            else:
                # XXX clean this once we have a support_sample_weight tag
                if sample_weight_is_none:
                    self.init_.fit(X_train, y_train)
                else:
                    msg = (
                        "The initial estimator {} does not support sample "
                        "weights.".format(self.init_.__class__.__name__)
                    )
                    try:
                        self.init_.fit(
                            X_train, y_train, sample_weight=sample_weight_train
                        )
                    except TypeError as e:
                        if "unexpected keyword argument 'sample_weight'" in str(e):
                            # regular estimator without SW support
                            raise ValueError(msg) from e
                        else:  # regular estimator whose input checking failed
                            raise
                    except ValueError as e:
                        if (
                            "pass parameters to specific steps of "
                            "your pipeline using the "
                            "stepname__parameter"
                            in str(e)
                        ):  # pipeline
                            raise ValueError(msg) from e
                        else:  # regular estimator whose input checking failed
                            raise

                raw_predictions = _init_raw_predictions(
                    X_train, self.init_, self._loss, is_classifier(self)
                )

            begin_at_stage = 0

            # The rng state must be preserved if warm_start is True
            self._rng = check_random_state(self.random_state)

        # warm start: this is not the first time fit was called
        else:
            # add more estimators to fitted model
            # invariant: warm_start = True
            if self.n_estimators < self.estimators_.shape[0]:
                raise ValueError(
                    "n_estimators=%d must be larger or equal to "
                    "estimators_.shape[0]=%d when "
                    "warm_start==True" % (self.n_estimators, self.estimators_.shape[0])
                )
            begin_at_stage = self.estimators_.shape[0]
            # The requirements of _raw_predict
            # are more constrained than fit. It accepts only CSR
            # matrices. Finite values have already been checked in _validate_data.
            X_train = check_array(
                X_train,
                dtype=DTYPE,
                order="C",
                accept_sparse="csr",
                force_all_finite=False,
            )
            raw_predictions = self._raw_predict(X_train)
            self._resize_state()

        # fit the boosting stages
        n_stages = self._fit_stages(
            X_train,
            y_train,
            raw_predictions,
            sample_weight_train,
            self._rng,
            X_val,
            y_val,
            sample_weight_val,
            begin_at_stage,
            monitor,
        )

        # change shape of arrays after fit (early-stopping or additional ests)
        if n_stages != self.estimators_.shape[0]:
            self.estimators_ = self.estimators_[:n_stages]
            self.train_score_ = self.train_score_[:n_stages]
            if hasattr(self, "oob_improvement_"):
                # OOB scores were computed
                self.oob_improvement_ = self.oob_improvement_[:n_stages]
                self.oob_scores_ = self.oob_scores_[:n_stages]
                self.oob_score_ = self.oob_scores_[-1]
        self.n_estimators_ = n_stages
        return self

    def _fit_stages(
        self,
        X,
        y,
        raw_predictions,
        sample_weight,
        random_state,
        X_val,
        y_val,
        sample_weight_val,
        begin_at_stage=0,
        monitor=None,
    ):
        """Iteratively fits the stages.

        For each stage it computes the progress (OOB, train score)
        and delegates to ``_fit_stage``.
        Returns the number of stages fit; might differ from ``n_estimators``
        due to early stopping.
        """
        n_samples = X.shape[0]
        do_oob = self.subsample < 1.0
        sample_mask = np.ones((n_samples,), dtype=bool)
        n_inbag = max(1, int(self.subsample * n_samples))

        if self.verbose:
            verbose_reporter = VerboseReporter(verbose=self.verbose)
            verbose_reporter.init(self, begin_at_stage)

        X_csc = csc_matrix(X) if issparse(X) else None
        X_csr = csr_matrix(X) if issparse(X) else None

        if self.n_iter_no_change is not None:
            loss_history = np.full(self.n_iter_no_change, np.inf)
            # We create a generator to get the predictions for X_val after
            # the addition of each successive stage
            y_val_pred_iter = self._staged_raw_predict(X_val, check_input=False)

        # Older versions of GBT had its own loss functions. With the new common
        # private loss function submodule _loss, we often are a factor of 2
        # away from the old version. Here we keep backward compatibility for
        # oob_scores_ and oob_improvement_, even if the old way is quite
        # inconsistent (sometimes the gradient is half the gradient, sometimes
        # not).
        if isinstance(
            self._loss,
            (
                HalfSquaredError,
                HalfBinomialLoss,
            ),
        ):
            factor = 2
        else:
            factor = 1

        # perform boosting iterations
        i = begin_at_stage
        for i in range(begin_at_stage, self.n_estimators):
            # subsampling
            if do_oob:
                sample_mask = _random_sample_mask(n_samples, n_inbag, random_state)
                y_oob_masked = y[~sample_mask]
                sample_weight_oob_masked = sample_weight[~sample_mask]
                if i == 0:  # store the initial loss to compute the OOB score
                    initial_loss = factor * self._loss(
                        y_true=y_oob_masked,
                        raw_prediction=raw_predictions[~sample_mask],
                        sample_weight=sample_weight_oob_masked,
                    )

            # fit next stage of trees
            raw_predictions = self._fit_stage(
                i,
                X,
                y,
                raw_predictions,
                sample_weight,
                sample_mask,
                random_state,
                X_csc=X_csc,
                X_csr=X_csr,
            )

            # track loss
            if do_oob:
                self.train_score_[i] = factor * self._loss(
                    y_true=y[sample_mask],
                    raw_prediction=raw_predictions[sample_mask],
                    sample_weight=sample_weight[sample_mask],
                )
                self.oob_scores_[i] = factor * self._loss(
                    y_true=y_oob_masked,
                    raw_prediction=raw_predictions[~sample_mask],
                    sample_weight=sample_weight_oob_masked,
                )
                previous_loss = initial_loss if i == 0 else self.oob_scores_[i - 1]
                self.oob_improvement_[i] = previous_loss - self.oob_scores_[i]
                self.oob_score_ = self.oob_scores_[-1]
            else:
                # no need to fancy index w/ no subsampling
                self.train_score_[i] = factor * self._loss(
                    y_true=y,
                    raw_prediction=raw_predictions,
                    sample_weight=sample_weight,
                )

            if self.verbose > 0:
                verbose_reporter.update(i, self)

            if monitor is not None:
                early_stopping = monitor(i, self, locals())
                if early_stopping:
                    break

            # We also provide an early stopping based on the score from
            # validation set (X_val, y_val), if n_iter_no_change is set
            if self.n_iter_no_change is not None:
                # By calling next(y_val_pred_iter), we get the predictions
                # for X_val after the addition of the current stage
                validation_loss = factor * self._loss(
                    y_val, next(y_val_pred_iter), sample_weight_val
                )

                # Require validation_score to be better (less) than at least
                # one of the last n_iter_no_change evaluations
                if np.any(validation_loss + self.tol < loss_history):
                    loss_history[i % len(loss_history)] = validation_loss
                else:
                    break

        return i + 1

    def _make_estimator(self, append=True):
        # we don't need _make_estimator
        raise NotImplementedError()

    def _raw_predict_init(self, X):
        """Check input and compute raw predictions of the init estimator."""
        self._check_initialized()
        X = self.estimators_[0, 0]._validate_X_predict(X, check_input=True)
        if self.init_ == "zero":
            raw_predictions = np.zeros(
                shape=(X.shape[0], self.n_trees_per_iteration_), dtype=np.float64
            )
        else:
            raw_predictions = _init_raw_predictions(
                X, self.init_, self._loss, is_classifier(self)
            )
        return raw_predictions

    def _raw_predict(self, X):
        """Return the sum of the trees raw predictions (+ init estimator)."""
        check_is_fitted(self)
        raw_predictions = self._raw_predict_init(X)
        predict_stages(self.estimators_, X, self.learning_rate, raw_predictions)
        return raw_predictions

    def _staged_raw_predict(self, X, check_input=True):
        """Compute raw predictions of ``X`` for each iteration.

        This method allows monitoring (i.e. determine error on testing set)
        after each stage.

        Parameters
        ----------
        X : {array-like, sparse matrix} of shape (n_samples, n_features)
            The input samples. Internally, it will be converted to
            ``dtype=np.float32`` and if a sparse matrix is provided
            to a sparse ``csr_matrix``.

        check_input : bool, default=True
            If False, the input arrays X will not be checked.

        Returns
        -------
        raw_predictions : generator of ndarray of shape (n_samples, k)
            The raw predictions of the input samples. The order of the
            classes corresponds to that in the attribute :term:`classes_`.
            Regression and binary classification are special cases with
            ``k == 1``, otherwise ``k==n_classes``.
        """
        if check_input:
            X = self._validate_data(
                X, dtype=DTYPE, order="C", accept_sparse="csr", reset=False
            )
        raw_predictions = self._raw_predict_init(X)
        for i in range(self.estimators_.shape[0]):
            predict_stage(self.estimators_, i, X, self.learning_rate, raw_predictions)
            yield raw_predictions.copy()

    @property
    def feature_importances_(self):
        """The impurity-based feature importances.

        The higher, the more important the feature.
        The importance of a feature is computed as the (normalized)
        total reduction of the criterion brought by that feature.  It is also
        known as the Gini importance.

        Warning: impurity-based feature importances can be misleading for
        high cardinality features (many unique values). See
        :func:`sklearn.inspection.permutation_importance` as an alternative.

        Returns
        -------
        feature_importances_ : ndarray of shape (n_features,)
            The values of this array sum to 1, unless all trees are single node
            trees consisting of only the root node, in which case it will be an
            array of zeros.
        """
        self._check_initialized()

        relevant_trees = [
            tree
            for stage in self.estimators_
            for tree in stage
            if tree.tree_.node_count > 1
        ]
        if not relevant_trees:
            # degenerate case where all trees have only one node
            return np.zeros(shape=self.n_features_in_, dtype=np.float64)

        relevant_feature_importances = [
            tree.tree_.compute_feature_importances(normalize=False)
            for tree in relevant_trees
        ]
        avg_feature_importances = np.mean(
            relevant_feature_importances, axis=0, dtype=np.float64
        )
        return avg_feature_importances / np.sum(avg_feature_importances)

    def _compute_partial_dependence_recursion(self, grid, target_features):
        """Fast partial dependence computation.

        Parameters
        ----------
        grid : ndarray of shape (n_samples, n_target_features)
            The grid points on which the partial dependence should be
            evaluated.
        target_features : ndarray of shape (n_target_features,)
            The set of target features for which the partial dependence
            should be evaluated.

        Returns
        -------
        averaged_predictions : ndarray of shape \
                (n_trees_per_iteration_, n_samples)
            The value of the partial dependence function on each grid point.
        """
        if self.init is not None:
            warnings.warn(
                "Using recursion method with a non-constant init predictor "
                "will lead to incorrect partial dependence values. "
                "Got init=%s."
                % self.init,
                UserWarning,
            )
        grid = np.asarray(grid, dtype=DTYPE, order="C")
        n_estimators, n_trees_per_stage = self.estimators_.shape
        averaged_predictions = np.zeros(
            (n_trees_per_stage, grid.shape[0]), dtype=np.float64, order="C"
        )
        for stage in range(n_estimators):
            for k in range(n_trees_per_stage):
                tree = self.estimators_[stage, k].tree_
                tree.compute_partial_dependence(
                    grid, target_features, averaged_predictions[k]
                )
        averaged_predictions *= self.learning_rate

        return averaged_predictions

    def apply(self, X):
        """Apply trees in the ensemble to X, return leaf indices.

        .. versionadded:: 0.17

        Parameters
        ----------
        X : {array-like, sparse matrix} of shape (n_samples, n_features)
            The input samples. Internally, its dtype will be converted to
            ``dtype=np.float32``. If a sparse matrix is provided, it will
            be converted to a sparse ``csr_matrix``.

        Returns
        -------
        X_leaves : array-like of shape (n_samples, n_estimators, n_classes)
            For each datapoint x in X and for each tree in the ensemble,
            return the index of the leaf x ends up in each estimator.
            In the case of binary classification n_classes is 1.
        """

        self._check_initialized()
        X = self.estimators_[0, 0]._validate_X_predict(X, check_input=True)

        # n_classes will be equal to 1 in the binary classification or the
        # regression case.
        n_estimators, n_classes = self.estimators_.shape
        leaves = np.zeros((X.shape[0], n_estimators, n_classes))

        for i in range(n_estimators):
            for j in range(n_classes):
                estimator = self.estimators_[i, j]
                leaves[:, i, j] = estimator.apply(X, check_input=False)

        return leaves


class GradientBoostingClassifier(ClassifierMixin, BaseGradientBoosting):
    """Gradient Boosting for classification.

    This algorithm builds an additive model in a forward stage-wise fashion; it
    allows for the optimization of arbitrary differentiable loss functions. In
    each stage ``n_classes_`` regression trees are fit on the negative gradient
    of the loss function, e.g. binary or multiclass log loss. Binary
    classification is a special case where only a single regression tree is
    induced.

    :class:`sklearn.ensemble.HistGradientBoostingClassifier` is a much faster
    variant of this algorithm for intermediate datasets (`n_samples >= 10_000`).

    Read more in the :ref:`User Guide <gradient_boosting>`.

    Parameters
    ----------
    loss : {'log_loss', 'exponential'}, default='log_loss'
        The loss function to be optimized. 'log_loss' refers to binomial and
        multinomial deviance, the same as used in logistic regression.
        It is a good choice for classification with probabilistic outputs.
        For loss 'exponential', gradient boosting recovers the AdaBoost algorithm.

    learning_rate : float, default=0.1
        Learning rate shrinks the contribution of each tree by `learning_rate`.
        There is a trade-off between learning_rate and n_estimators.
        Values must be in the range `[0.0, inf)`.

    n_estimators : int, default=100
        The number of boosting stages to perform. Gradient boosting
        is fairly robust to over-fitting so a large number usually
        results in better performance.
        Values must be in the range `[1, inf)`.

    subsample : float, default=1.0
        The fraction of samples to be used for fitting the individual base
        learners. If smaller than 1.0 this results in Stochastic Gradient
        Boosting. `subsample` interacts with the parameter `n_estimators`.
        Choosing `subsample < 1.0` leads to a reduction of variance
        and an increase in bias.
        Values must be in the range `(0.0, 1.0]`.

    criterion : {'friedman_mse', 'squared_error'}, default='friedman_mse'
        The function to measure the quality of a split. Supported criteria are
        'friedman_mse' for the mean squared error with improvement score by
        Friedman, 'squared_error' for mean squared error. The default value of
        'friedman_mse' is generally the best as it can provide a better
        approximation in some cases.

        .. versionadded:: 0.18

    min_samples_split : int or float, default=2
        The minimum number of samples required to split an internal node:

        - If int, values must be in the range `[2, inf)`.
        - If float, values must be in the range `(0.0, 1.0]` and `min_samples_split`
          will be `ceil(min_samples_split * n_samples)`.

        .. versionchanged:: 0.18
           Added float values for fractions.

    min_samples_leaf : int or float, default=1
        The minimum number of samples required to be at a leaf node.
        A split point at any depth will only be considered if it leaves at
        least ``min_samples_leaf`` training samples in each of the left and
        right branches.  This may have the effect of smoothing the model,
        especially in regression.

        - If int, values must be in the range `[1, inf)`.
        - If float, values must be in the range `(0.0, 1.0)` and `min_samples_leaf`
          will be `ceil(min_samples_leaf * n_samples)`.

        .. versionchanged:: 0.18
           Added float values for fractions.

    min_weight_fraction_leaf : float, default=0.0
        The minimum weighted fraction of the sum total of weights (of all
        the input samples) required to be at a leaf node. Samples have
        equal weight when sample_weight is not provided.
        Values must be in the range `[0.0, 0.5]`.

    max_depth : int or None, default=3
        Maximum depth of the individual regression estimators. The maximum
        depth limits the number of nodes in the tree. Tune this parameter
        for best performance; the best value depends on the interaction
        of the input variables. If None, then nodes are expanded until
        all leaves are pure or until all leaves contain less than
        min_samples_split samples.
        If int, values must be in the range `[1, inf)`.

    min_impurity_decrease : float, default=0.0
        A node will be split if this split induces a decrease of the impurity
        greater than or equal to this value.
        Values must be in the range `[0.0, inf)`.

        The weighted impurity decrease equation is the following::

            N_t / N * (impurity - N_t_R / N_t * right_impurity
                                - N_t_L / N_t * left_impurity)

        where ``N`` is the total number of samples, ``N_t`` is the number of
        samples at the current node, ``N_t_L`` is the number of samples in the
        left child, and ``N_t_R`` is the number of samples in the right child.

        ``N``, ``N_t``, ``N_t_R`` and ``N_t_L`` all refer to the weighted sum,
        if ``sample_weight`` is passed.

        .. versionadded:: 0.19

    init : estimator or 'zero', default=None
        An estimator object that is used to compute the initial predictions.
        ``init`` has to provide :term:`fit` and :term:`predict_proba`. If
        'zero', the initial raw predictions are set to zero. By default, a
        ``DummyEstimator`` predicting the classes priors is used.

    random_state : int, RandomState instance or None, default=None
        Controls the random seed given to each Tree estimator at each
        boosting iteration.
        In addition, it controls the random permutation of the features at
        each split (see Notes for more details).
        It also controls the random splitting of the training data to obtain a
        validation set if `n_iter_no_change` is not None.
        Pass an int for reproducible output across multiple function calls.
        See :term:`Glossary <random_state>`.

    max_features : {'sqrt', 'log2'}, int or float, default=None
        The number of features to consider when looking for the best split:

        - If int, values must be in the range `[1, inf)`.
        - If float, values must be in the range `(0.0, 1.0]` and the features
          considered at each split will be `max(1, int(max_features * n_features_in_))`.
        - If 'sqrt', then `max_features=sqrt(n_features)`.
        - If 'log2', then `max_features=log2(n_features)`.
        - If None, then `max_features=n_features`.

        Choosing `max_features < n_features` leads to a reduction of variance
        and an increase in bias.

        Note: the search for a split does not stop until at least one
        valid partition of the node samples is found, even if it requires to
        effectively inspect more than ``max_features`` features.

    verbose : int, default=0
        Enable verbose output. If 1 then it prints progress and performance
        once in a while (the more trees the lower the frequency). If greater
        than 1 then it prints progress and performance for every tree.
        Values must be in the range `[0, inf)`.

    max_leaf_nodes : int, default=None
        Grow trees with ``max_leaf_nodes`` in best-first fashion.
        Best nodes are defined as relative reduction in impurity.
        Values must be in the range `[2, inf)`.
        If `None`, then unlimited number of leaf nodes.

    warm_start : bool, default=False
        When set to ``True``, reuse the solution of the previous call to fit
        and add more estimators to the ensemble, otherwise, just erase the
        previous solution. See :term:`the Glossary <warm_start>`.

    validation_fraction : float, default=0.1
        The proportion of training data to set aside as validation set for
        early stopping. Values must be in the range `(0.0, 1.0)`.
        Only used if ``n_iter_no_change`` is set to an integer.

        .. versionadded:: 0.20

    n_iter_no_change : int, default=None
        ``n_iter_no_change`` is used to decide if early stopping will be used
        to terminate training when validation score is not improving. By
        default it is set to None to disable early stopping. If set to a
        number, it will set aside ``validation_fraction`` size of the training
        data as validation and terminate training when validation score is not
        improving in all of the previous ``n_iter_no_change`` numbers of
        iterations. The split is stratified.
        Values must be in the range `[1, inf)`.
        See
        :ref:`sphx_glr_auto_examples_ensemble_plot_gradient_boosting_early_stopping.py`.

        .. versionadded:: 0.20

    tol : float, default=1e-4
        Tolerance for the early stopping. When the loss is not improving
        by at least tol for ``n_iter_no_change`` iterations (if set to a
        number), the training stops.
        Values must be in the range `[0.0, inf)`.

        .. versionadded:: 0.20

    ccp_alpha : non-negative float, default=0.0
        Complexity parameter used for Minimal Cost-Complexity Pruning. The
        subtree with the largest cost complexity that is smaller than
        ``ccp_alpha`` will be chosen. By default, no pruning is performed.
        Values must be in the range `[0.0, inf)`.
        See :ref:`minimal_cost_complexity_pruning` for details.

        .. versionadded:: 0.22

    Attributes
    ----------
    n_estimators_ : int
        The number of estimators as selected by early stopping (if
        ``n_iter_no_change`` is specified). Otherwise it is set to
        ``n_estimators``.

        .. versionadded:: 0.20

    n_trees_per_iteration_ : int
        The number of trees that are built at each iteration. For binary classifiers,
        this is always 1.

        .. versionadded:: 1.4.0

    feature_importances_ : ndarray of shape (n_features,)
        The impurity-based feature importances.
        The higher, the more important the feature.
        The importance of a feature is computed as the (normalized)
        total reduction of the criterion brought by that feature.  It is also
        known as the Gini importance.

        Warning: impurity-based feature importances can be misleading for
        high cardinality features (many unique values). See
        :func:`sklearn.inspection.permutation_importance` as an alternative.

    oob_improvement_ : ndarray of shape (n_estimators,)
        The improvement in loss on the out-of-bag samples
        relative to the previous iteration.
        ``oob_improvement_[0]`` is the improvement in
        loss of the first stage over the ``init`` estimator.
        Only available if ``subsample < 1.0``.

    oob_scores_ : ndarray of shape (n_estimators,)
        The full history of the loss values on the out-of-bag
        samples. Only available if `subsample < 1.0`.

        .. versionadded:: 1.3

    oob_score_ : float
        The last value of the loss on the out-of-bag samples. It is
        the same as `oob_scores_[-1]`. Only available if `subsample < 1.0`.

        .. versionadded:: 1.3

    train_score_ : ndarray of shape (n_estimators,)
        The i-th score ``train_score_[i]`` is the loss of the
        model at iteration ``i`` on the in-bag sample.
        If ``subsample == 1`` this is the loss on the training data.

    init_ : estimator
        The estimator that provides the initial predictions. Set via the ``init``
        argument.

    estimators_ : ndarray of DecisionTreeRegressor of \
            shape (n_estimators, ``n_trees_per_iteration_``)
        The collection of fitted sub-estimators. ``n_trees_per_iteration_`` is 1 for
        binary classification, otherwise ``n_classes``.

    classes_ : ndarray of shape (n_classes,)
        The classes labels.

    n_features_in_ : int
        Number of features seen during :term:`fit`.

        .. versionadded:: 0.24

    feature_names_in_ : ndarray of shape (`n_features_in_`,)
        Names of features seen during :term:`fit`. Defined only when `X`
        has feature names that are all strings.

        .. versionadded:: 1.0

    n_classes_ : int
        The number of classes.

    max_features_ : int
        The inferred value of max_features.

    See Also
    --------
    HistGradientBoostingClassifier : Histogram-based Gradient Boosting
        Classification Tree.
    sklearn.tree.DecisionTreeClassifier : A decision tree classifier.
    RandomForestClassifier : A meta-estimator that fits a number of decision
        tree classifiers on various sub-samples of the dataset and uses
        averaging to improve the predictive accuracy and control over-fitting.
    AdaBoostClassifier : A meta-estimator that begins by fitting a classifier
        on the original dataset and then fits additional copies of the
        classifier on the same dataset where the weights of incorrectly
        classified instances are adjusted such that subsequent classifiers
        focus more on difficult cases.

    Notes
    -----
    The features are always randomly permuted at each split. Therefore,
    the best found split may vary, even with the same training data and
    ``max_features=n_features``, if the improvement of the criterion is
    identical for several splits enumerated during the search of the best
    split. To obtain a deterministic behaviour during fitting,
    ``random_state`` has to be fixed.

    References
    ----------
    J. Friedman, Greedy Function Approximation: A Gradient Boosting
    Machine, The Annals of Statistics, Vol. 29, No. 5, 2001.

    J. Friedman, Stochastic Gradient Boosting, 1999

    T. Hastie, R. Tibshirani and J. Friedman.
    Elements of Statistical Learning Ed. 2, Springer, 2009.

    Examples
    --------
    The following example shows how to fit a gradient boosting classifier with
    100 decision stumps as weak learners.

    >>> from sklearn.datasets import make_hastie_10_2
    >>> from sklearn.ensemble import GradientBoostingClassifier

    >>> X, y = make_hastie_10_2(random_state=0)
    >>> X_train, X_test = X[:2000], X[2000:]
    >>> y_train, y_test = y[:2000], y[2000:]

    >>> clf = GradientBoostingClassifier(n_estimators=100, learning_rate=1.0,
    ...     max_depth=1, random_state=0).fit(X_train, y_train)
    >>> clf.score(X_test, y_test)
    0.913...
    """

    _parameter_constraints: dict = {
        **BaseGradientBoosting._parameter_constraints,
        "loss": [StrOptions({"log_loss", "exponential"})],
        "init": [StrOptions({"zero"}), None, HasMethods(["fit", "predict_proba"])],
    }

    def __init__(
        self,
        *,
        loss="log_loss",
        learning_rate=0.1,
        n_estimators=100,
        subsample=1.0,
        criterion="friedman_mse",
        min_samples_split=2,
        min_samples_leaf=1,
        min_weight_fraction_leaf=0.0,
        max_depth=3,
        min_impurity_decrease=0.0,
        init=None,
        random_state=None,
        max_features=None,
        verbose=0,
        max_leaf_nodes=None,
        warm_start=False,
        validation_fraction=0.1,
        n_iter_no_change=None,
        tol=1e-4,
        ccp_alpha=0.0,
    ):
        super().__init__(
            loss=loss,
            learning_rate=learning_rate,
            n_estimators=n_estimators,
            criterion=criterion,
            min_samples_split=min_samples_split,
            min_samples_leaf=min_samples_leaf,
            min_weight_fraction_leaf=min_weight_fraction_leaf,
            max_depth=max_depth,
            init=init,
            subsample=subsample,
            max_features=max_features,
            random_state=random_state,
            verbose=verbose,
            max_leaf_nodes=max_leaf_nodes,
            min_impurity_decrease=min_impurity_decrease,
            warm_start=warm_start,
            validation_fraction=validation_fraction,
            n_iter_no_change=n_iter_no_change,
            tol=tol,
            ccp_alpha=ccp_alpha,
        )

    def _encode_y(self, y, sample_weight):
        # encode classes into 0 ... n_classes - 1 and sets attributes classes_
        # and n_trees_per_iteration_
        check_classification_targets(y)

        label_encoder = LabelEncoder()
        encoded_y_int = label_encoder.fit_transform(y)
        self.classes_ = label_encoder.classes_
        n_classes = self.classes_.shape[0]
        # only 1 tree for binary classification. For multiclass classification,
        # we build 1 tree per class.
        self.n_trees_per_iteration_ = 1 if n_classes <= 2 else n_classes
        encoded_y = encoded_y_int.astype(float, copy=False)

        # From here on, it is additional to the HGBT case.
        # expose n_classes_ attribute
        self.n_classes_ = n_classes
        if sample_weight is None:
            n_trim_classes = n_classes
        else:
            n_trim_classes = np.count_nonzero(np.bincount(encoded_y_int, sample_weight))

        if n_trim_classes < 2:
            raise ValueError(
                "y contains %d class after sample_weight "
                "trimmed classes with zero weights, while a "
                "minimum of 2 classes are required." % n_trim_classes
            )
        return encoded_y

    def _get_loss(self, sample_weight):
        if self.loss == "log_loss":
            if self.n_classes_ == 2:
                return HalfBinomialLoss(sample_weight=sample_weight)
            else:
                return HalfMultinomialLoss(
                    sample_weight=sample_weight, n_classes=self.n_classes_
                )
        elif self.loss == "exponential":
            if self.n_classes_ > 2:
                raise ValueError(
                    f"loss='{self.loss}' is only suitable for a binary classification "
                    f"problem, you have n_classes={self.n_classes_}. "
                    "Please use loss='log_loss' instead."
                )
            else:
                return ExponentialLoss(sample_weight=sample_weight)

    def decision_function(self, X):
        """Compute the decision function of ``X``.

        Parameters
        ----------
        X : {array-like, sparse matrix} of shape (n_samples, n_features)
            The input samples. Internally, it will be converted to
            ``dtype=np.float32`` and if a sparse matrix is provided
            to a sparse ``csr_matrix``.

        Returns
        -------
        score : ndarray of shape (n_samples, n_classes) or (n_samples,)
            The decision function of the input samples, which corresponds to
            the raw values predicted from the trees of the ensemble . The
            order of the classes corresponds to that in the attribute
            :term:`classes_`. Regression and binary classification produce an
            array of shape (n_samples,).
        """
        X = self._validate_data(
            X, dtype=DTYPE, order="C", accept_sparse="csr", reset=False
        )
        raw_predictions = self._raw_predict(X)
        if raw_predictions.shape[1] == 1:
            return raw_predictions.ravel()
        return raw_predictions

    def staged_decision_function(self, X):
        """Compute decision function of ``X`` for each iteration.

        This method allows monitoring (i.e. determine error on testing set)
        after each stage.

        Parameters
        ----------
        X : {array-like, sparse matrix} of shape (n_samples, n_features)
            The input samples. Internally, it will be converted to
            ``dtype=np.float32`` and if a sparse matrix is provided
            to a sparse ``csr_matrix``.

        Yields
        ------
        score : generator of ndarray of shape (n_samples, k)
            The decision function of the input samples, which corresponds to
            the raw values predicted from the trees of the ensemble . The
            classes corresponds to that in the attribute :term:`classes_`.
            Regression and binary classification are special cases with
            ``k == 1``, otherwise ``k==n_classes``.
        """
        yield from self._staged_raw_predict(X)

    def predict(self, X):
        """Predict class for X.

        Parameters
        ----------
        X : {array-like, sparse matrix} of shape (n_samples, n_features)
            The input samples. Internally, it will be converted to
            ``dtype=np.float32`` and if a sparse matrix is provided
            to a sparse ``csr_matrix``.

        Returns
        -------
        y : ndarray of shape (n_samples,)
            The predicted values.
        """
        raw_predictions = self.decision_function(X)
        if raw_predictions.ndim == 1:  # decision_function already squeezed it
            encoded_classes = (raw_predictions >= 0).astype(int)
        else:
            encoded_classes = np.argmax(raw_predictions, axis=1)
        return self.classes_[encoded_classes]

    def staged_predict(self, X):
        """Predict class at each stage for X.

        This method allows monitoring (i.e. determine error on testing set)
        after each stage.

        Parameters
        ----------
        X : {array-like, sparse matrix} of shape (n_samples, n_features)
            The input samples. Internally, it will be converted to
            ``dtype=np.float32`` and if a sparse matrix is provided
            to a sparse ``csr_matrix``.

        Yields
        ------
        y : generator of ndarray of shape (n_samples,)
            The predicted value of the input samples.
        """
        if self.n_classes_ == 2:  # n_trees_per_iteration_ = 1
            for raw_predictions in self._staged_raw_predict(X):
                encoded_classes = (raw_predictions.squeeze() >= 0).astype(int)
                yield self.classes_.take(encoded_classes, axis=0)
        else:
            for raw_predictions in self._staged_raw_predict(X):
                encoded_classes = np.argmax(raw_predictions, axis=1)
                yield self.classes_.take(encoded_classes, axis=0)

    def predict_proba(self, X):
        """Predict class probabilities for X.

        Parameters
        ----------
        X : {array-like, sparse matrix} of shape (n_samples, n_features)
            The input samples. Internally, it will be converted to
            ``dtype=np.float32`` and if a sparse matrix is provided
            to a sparse ``csr_matrix``.

        Returns
        -------
        p : ndarray of shape (n_samples, n_classes)
            The class probabilities of the input samples. The order of the
            classes corresponds to that in the attribute :term:`classes_`.

        Raises
        ------
        AttributeError
            If the ``loss`` does not support probabilities.
        """
        raw_predictions = self.decision_function(X)
        return self._loss.predict_proba(raw_predictions)

    def predict_log_proba(self, X):
        """Predict class log-probabilities for X.

        Parameters
        ----------
        X : {array-like, sparse matrix} of shape (n_samples, n_features)
            The input samples. Internally, it will be converted to
            ``dtype=np.float32`` and if a sparse matrix is provided
            to a sparse ``csr_matrix``.

        Returns
        -------
        p : ndarray of shape (n_samples, n_classes)
            The class log-probabilities of the input samples. The order of the
            classes corresponds to that in the attribute :term:`classes_`.

        Raises
        ------
        AttributeError
            If the ``loss`` does not support probabilities.
        """
        proba = self.predict_proba(X)
        return np.log(proba)

    def staged_predict_proba(self, X):
        """Predict class probabilities at each stage for X.

        This method allows monitoring (i.e. determine error on testing set)
        after each stage.

        Parameters
        ----------
        X : {array-like, sparse matrix} of shape (n_samples, n_features)
            The input samples. Internally, it will be converted to
            ``dtype=np.float32`` and if a sparse matrix is provided
            to a sparse ``csr_matrix``.

        Yields
        ------
        y : generator of ndarray of shape (n_samples,)
            The predicted value of the input samples.
        """
        try:
            for raw_predictions in self._staged_raw_predict(X):
                yield self._loss.predict_proba(raw_predictions)
        except NotFittedError:
            raise
        except AttributeError as e:
            raise AttributeError(
                "loss=%r does not support predict_proba" % self.loss
            ) from e


class GradientBoostingRegressor(RegressorMixin, BaseGradientBoosting):
    """Gradient Boosting for regression.

    This estimator builds an additive model in a forward stage-wise fashion; it
    allows for the optimization of arbitrary differentiable loss functions. In
    each stage a regression tree is fit on the negative gradient of the given
    loss function.

    :class:`sklearn.ensemble.HistGradientBoostingRegressor` is a much faster
    variant of this algorithm for intermediate datasets (`n_samples >= 10_000`).

    Read more in the :ref:`User Guide <gradient_boosting>`.

    Parameters
    ----------
    loss : {'squared_error', 'absolute_error', 'huber', 'quantile'}, \
            default='squared_error'
        Loss function to be optimized. 'squared_error' refers to the squared
        error for regression. 'absolute_error' refers to the absolute error of
        regression and is a robust loss function. 'huber' is a
        combination of the two. 'quantile' allows quantile regression (use
        `alpha` to specify the quantile).

    learning_rate : float, default=0.1
        Learning rate shrinks the contribution of each tree by `learning_rate`.
        There is a trade-off between learning_rate and n_estimators.
        Values must be in the range `[0.0, inf)`.

    n_estimators : int, default=100
        The number of boosting stages to perform. Gradient boosting
        is fairly robust to over-fitting so a large number usually
        results in better performance.
        Values must be in the range `[1, inf)`.

    subsample : float, default=1.0
        The fraction of samples to be used for fitting the individual base
        learners. If smaller than 1.0 this results in Stochastic Gradient
        Boosting. `subsample` interacts with the parameter `n_estimators`.
        Choosing `subsample < 1.0` leads to a reduction of variance
        and an increase in bias.
        Values must be in the range `(0.0, 1.0]`.

    criterion : {'friedman_mse', 'squared_error'}, default='friedman_mse'
        The function to measure the quality of a split. Supported criteria are
        "friedman_mse" for the mean squared error with improvement score by
        Friedman, "squared_error" for mean squared error. The default value of
        "friedman_mse" is generally the best as it can provide a better
        approximation in some cases.

        .. versionadded:: 0.18

    min_samples_split : int or float, default=2
        The minimum number of samples required to split an internal node:

        - If int, values must be in the range `[2, inf)`.
        - If float, values must be in the range `(0.0, 1.0]` and `min_samples_split`
          will be `ceil(min_samples_split * n_samples)`.

        .. versionchanged:: 0.18
           Added float values for fractions.

    min_samples_leaf : int or float, default=1
        The minimum number of samples required to be at a leaf node.
        A split point at any depth will only be considered if it leaves at
        least ``min_samples_leaf`` training samples in each of the left and
        right branches.  This may have the effect of smoothing the model,
        especially in regression.

        - If int, values must be in the range `[1, inf)`.
        - If float, values must be in the range `(0.0, 1.0)` and `min_samples_leaf`
          will be `ceil(min_samples_leaf * n_samples)`.

        .. versionchanged:: 0.18
           Added float values for fractions.

    min_weight_fraction_leaf : float, default=0.0
        The minimum weighted fraction of the sum total of weights (of all
        the input samples) required to be at a leaf node. Samples have
        equal weight when sample_weight is not provided.
        Values must be in the range `[0.0, 0.5]`.

    max_depth : int or None, default=3
        Maximum depth of the individual regression estimators. The maximum
        depth limits the number of nodes in the tree. Tune this parameter
        for best performance; the best value depends on the interaction
        of the input variables. If None, then nodes are expanded until
        all leaves are pure or until all leaves contain less than
        min_samples_split samples.
        If int, values must be in the range `[1, inf)`.

    min_impurity_decrease : float, default=0.0
        A node will be split if this split induces a decrease of the impurity
        greater than or equal to this value.
        Values must be in the range `[0.0, inf)`.

        The weighted impurity decrease equation is the following::

            N_t / N * (impurity - N_t_R / N_t * right_impurity
                                - N_t_L / N_t * left_impurity)

        where ``N`` is the total number of samples, ``N_t`` is the number of
        samples at the current node, ``N_t_L`` is the number of samples in the
        left child, and ``N_t_R`` is the number of samples in the right child.

        ``N``, ``N_t``, ``N_t_R`` and ``N_t_L`` all refer to the weighted sum,
        if ``sample_weight`` is passed.

        .. versionadded:: 0.19

    init : estimator or 'zero', default=None
        An estimator object that is used to compute the initial predictions.
        ``init`` has to provide :term:`fit` and :term:`predict`. If 'zero', the
        initial raw predictions are set to zero. By default a
        ``DummyEstimator`` is used, predicting either the average target value
        (for loss='squared_error'), or a quantile for the other losses.

    random_state : int, RandomState instance or None, default=None
        Controls the random seed given to each Tree estimator at each
        boosting iteration.
        In addition, it controls the random permutation of the features at
        each split (see Notes for more details).
        It also controls the random splitting of the training data to obtain a
        validation set if `n_iter_no_change` is not None.
        Pass an int for reproducible output across multiple function calls.
        See :term:`Glossary <random_state>`.

    max_features : {'sqrt', 'log2'}, int or float, default=None
        The number of features to consider when looking for the best split:

        - If int, values must be in the range `[1, inf)`.
        - If float, values must be in the range `(0.0, 1.0]` and the features
          considered at each split will be `max(1, int(max_features * n_features_in_))`.
        - If "sqrt", then `max_features=sqrt(n_features)`.
        - If "log2", then `max_features=log2(n_features)`.
        - If None, then `max_features=n_features`.

        Choosing `max_features < n_features` leads to a reduction of variance
        and an increase in bias.

        Note: the search for a split does not stop until at least one
        valid partition of the node samples is found, even if it requires to
        effectively inspect more than ``max_features`` features.

    alpha : float, default=0.9
        The alpha-quantile of the huber loss function and the quantile
        loss function. Only if ``loss='huber'`` or ``loss='quantile'``.
        Values must be in the range `(0.0, 1.0)`.

    verbose : int, default=0
        Enable verbose output. If 1 then it prints progress and performance
        once in a while (the more trees the lower the frequency). If greater
        than 1 then it prints progress and performance for every tree.
        Values must be in the range `[0, inf)`.

    max_leaf_nodes : int, default=None
        Grow trees with ``max_leaf_nodes`` in best-first fashion.
        Best nodes are defined as relative reduction in impurity.
        Values must be in the range `[2, inf)`.
        If None, then unlimited number of leaf nodes.

    warm_start : bool, default=False
        When set to ``True``, reuse the solution of the previous call to fit
        and add more estimators to the ensemble, otherwise, just erase the
        previous solution. See :term:`the Glossary <warm_start>`.

    validation_fraction : float, default=0.1
        The proportion of training data to set aside as validation set for
        early stopping. Values must be in the range `(0.0, 1.0)`.
        Only used if ``n_iter_no_change`` is set to an integer.

        .. versionadded:: 0.20

    n_iter_no_change : int, default=None
        ``n_iter_no_change`` is used to decide if early stopping will be used
        to terminate training when validation score is not improving. By
        default it is set to None to disable early stopping. If set to a
        number, it will set aside ``validation_fraction`` size of the training
        data as validation and terminate training when validation score is not
        improving in all of the previous ``n_iter_no_change`` numbers of
        iterations.
        Values must be in the range `[1, inf)`.
        See
        :ref:`sphx_glr_auto_examples_ensemble_plot_gradient_boosting_early_stopping.py`.

        .. versionadded:: 0.20

    tol : float, default=1e-4
        Tolerance for the early stopping. When the loss is not improving
        by at least tol for ``n_iter_no_change`` iterations (if set to a
        number), the training stops.
        Values must be in the range `[0.0, inf)`.

        .. versionadded:: 0.20

    ccp_alpha : non-negative float, default=0.0
        Complexity parameter used for Minimal Cost-Complexity Pruning. The
        subtree with the largest cost complexity that is smaller than
        ``ccp_alpha`` will be chosen. By default, no pruning is performed.
        Values must be in the range `[0.0, inf)`.
        See :ref:`minimal_cost_complexity_pruning` for details.

        .. versionadded:: 0.22

    Attributes
    ----------
    n_estimators_ : int
        The number of estimators as selected by early stopping (if
        ``n_iter_no_change`` is specified). Otherwise it is set to
        ``n_estimators``.

    n_trees_per_iteration_ : int
        The number of trees that are built at each iteration. For regressors, this is
        always 1.

        .. versionadded:: 1.4.0

    feature_importances_ : ndarray of shape (n_features,)
        The impurity-based feature importances.
        The higher, the more important the feature.
        The importance of a feature is computed as the (normalized)
        total reduction of the criterion brought by that feature.  It is also
        known as the Gini importance.

        Warning: impurity-based feature importances can be misleading for
        high cardinality features (many unique values). See
        :func:`sklearn.inspection.permutation_importance` as an alternative.

    oob_improvement_ : ndarray of shape (n_estimators,)
        The improvement in loss on the out-of-bag samples
        relative to the previous iteration.
        ``oob_improvement_[0]`` is the improvement in
        loss of the first stage over the ``init`` estimator.
        Only available if ``subsample < 1.0``.

    oob_scores_ : ndarray of shape (n_estimators,)
        The full history of the loss values on the out-of-bag
        samples. Only available if `subsample < 1.0`.

        .. versionadded:: 1.3

    oob_score_ : float
        The last value of the loss on the out-of-bag samples. It is
        the same as `oob_scores_[-1]`. Only available if `subsample < 1.0`.

        .. versionadded:: 1.3

    train_score_ : ndarray of shape (n_estimators,)
        The i-th score ``train_score_[i]`` is the loss of the
        model at iteration ``i`` on the in-bag sample.
        If ``subsample == 1`` this is the loss on the training data.

    init_ : estimator
        The estimator that provides the initial predictions. Set via the ``init``
        argument.

    estimators_ : ndarray of DecisionTreeRegressor of shape (n_estimators, 1)
        The collection of fitted sub-estimators.

    n_features_in_ : int
        Number of features seen during :term:`fit`.

        .. versionadded:: 0.24

    feature_names_in_ : ndarray of shape (`n_features_in_`,)
        Names of features seen during :term:`fit`. Defined only when `X`
        has feature names that are all strings.

        .. versionadded:: 1.0

    max_features_ : int
        The inferred value of max_features.

    See Also
    --------
    HistGradientBoostingRegressor : Histogram-based Gradient Boosting
        Classification Tree.
    sklearn.tree.DecisionTreeRegressor : A decision tree regressor.
    sklearn.ensemble.RandomForestRegressor : A random forest regressor.

    Notes
    -----
    The features are always randomly permuted at each split. Therefore,
    the best found split may vary, even with the same training data and
    ``max_features=n_features``, if the improvement of the criterion is
    identical for several splits enumerated during the search of the best
    split. To obtain a deterministic behaviour during fitting,
    ``random_state`` has to be fixed.

    References
    ----------
    J. Friedman, Greedy Function Approximation: A Gradient Boosting
    Machine, The Annals of Statistics, Vol. 29, No. 5, 2001.

    J. Friedman, Stochastic Gradient Boosting, 1999

    T. Hastie, R. Tibshirani and J. Friedman.
    Elements of Statistical Learning Ed. 2, Springer, 2009.

    Examples
    --------
    >>> from sklearn.datasets import make_regression
    >>> from sklearn.ensemble import GradientBoostingRegressor
    >>> from sklearn.model_selection import train_test_split
    >>> X, y = make_regression(random_state=0)
    >>> X_train, X_test, y_train, y_test = train_test_split(
    ...     X, y, random_state=0)
    >>> reg = GradientBoostingRegressor(random_state=0)
    >>> reg.fit(X_train, y_train)
    GradientBoostingRegressor(random_state=0)
    >>> reg.predict(X_test[1:2])
    array([-61...])
    >>> reg.score(X_test, y_test)
    0.4...
    """

    _parameter_constraints: dict = {
        **BaseGradientBoosting._parameter_constraints,
        "loss": [StrOptions({"squared_error", "absolute_error", "huber", "quantile"})],
        "init": [StrOptions({"zero"}), None, HasMethods(["fit", "predict"])],
        "alpha": [Interval(Real, 0.0, 1.0, closed="neither")],
    }

    def __init__(
        self,
        *,
        loss="squared_error",
        learning_rate=0.1,
        n_estimators=100,
        subsample=1.0,
        criterion="friedman_mse",
        min_samples_split=2,
        min_samples_leaf=1,
        min_weight_fraction_leaf=0.0,
        max_depth=3,
        min_impurity_decrease=0.0,
        init=None,
        random_state=None,
        max_features=None,
        alpha=0.9,
        verbose=0,
        max_leaf_nodes=None,
        warm_start=False,
        validation_fraction=0.1,
        n_iter_no_change=None,
        tol=1e-4,
        ccp_alpha=0.0,
    ):
        super().__init__(
            loss=loss,
            learning_rate=learning_rate,
            n_estimators=n_estimators,
            criterion=criterion,
            min_samples_split=min_samples_split,
            min_samples_leaf=min_samples_leaf,
            min_weight_fraction_leaf=min_weight_fraction_leaf,
            max_depth=max_depth,
            init=init,
            subsample=subsample,
            max_features=max_features,
            min_impurity_decrease=min_impurity_decrease,
            random_state=random_state,
            alpha=alpha,
            verbose=verbose,
            max_leaf_nodes=max_leaf_nodes,
            warm_start=warm_start,
            validation_fraction=validation_fraction,
            n_iter_no_change=n_iter_no_change,
            tol=tol,
            ccp_alpha=ccp_alpha,
        )

    def _encode_y(self, y=None, sample_weight=None):
        # Just convert y to the expected dtype
        self.n_trees_per_iteration_ = 1
        y = y.astype(DOUBLE, copy=False)
        return y

    def _get_loss(self, sample_weight):
        if self.loss in ("quantile", "huber"):
            return _LOSSES[self.loss](sample_weight=sample_weight, quantile=self.alpha)
        else:
            return _LOSSES[self.loss](sample_weight=sample_weight)

    def predict(self, X):
        """Predict regression target for X.

        Parameters
        ----------
        X : {array-like, sparse matrix} of shape (n_samples, n_features)
            The input samples. Internally, it will be converted to
            ``dtype=np.float32`` and if a sparse matrix is provided
            to a sparse ``csr_matrix``.

        Returns
        -------
        y : ndarray of shape (n_samples,)
            The predicted values.
        """
        X = self._validate_data(
            X, dtype=DTYPE, order="C", accept_sparse="csr", reset=False
        )
        # In regression we can directly return the raw value from the trees.
        return self._raw_predict(X).ravel()

    def staged_predict(self, X):
        """Predict regression target at each stage for X.

        This method allows monitoring (i.e. determine error on testing set)
        after each stage.

        Parameters
        ----------
        X : {array-like, sparse matrix} of shape (n_samples, n_features)
            The input samples. Internally, it will be converted to
            ``dtype=np.float32`` and if a sparse matrix is provided
            to a sparse ``csr_matrix``.

        Yields
        ------
        y : generator of ndarray of shape (n_samples,)
            The predicted value of the input samples.
        """
        for raw_predictions in self._staged_raw_predict(X):
            yield raw_predictions.ravel()

    def apply(self, X):
        """Apply trees in the ensemble to X, return leaf indices.

        .. versionadded:: 0.17

        Parameters
        ----------
        X : {array-like, sparse matrix} of shape (n_samples, n_features)
            The input samples. Internally, its dtype will be converted to
            ``dtype=np.float32``. If a sparse matrix is provided, it will
            be converted to a sparse ``csr_matrix``.

        Returns
        -------
        X_leaves : array-like of shape (n_samples, n_estimators)
            For each datapoint x in X and for each tree in the ensemble,
            return the index of the leaf x ends up in each estimator.
        """

        leaves = super().apply(X)
        leaves = leaves.reshape(X.shape[0], self.estimators_.shape[0])
        return leaves