isotonic.py 13.2 KB
# Authors: Fabian Pedregosa <fabian@fseoane.net>
#          Alexandre Gramfort <alexandre.gramfort@inria.fr>
#          Nelle Varoquaux <nelle.varoquaux@gmail.com>
# License: BSD 3 clause

import numpy as np
from scipy import interpolate
from scipy.stats import spearmanr
import warnings
import math

from .base import BaseEstimator, TransformerMixin, RegressorMixin
from .utils import check_array, check_consistent_length
from .utils.validation import _check_sample_weight, _deprecate_positional_args
from ._isotonic import _inplace_contiguous_isotonic_regression, _make_unique


__all__ = ['check_increasing', 'isotonic_regression',
           'IsotonicRegression']


def check_increasing(x, y):
    """Determine whether y is monotonically correlated with x.

    y is found increasing or decreasing with respect to x based on a Spearman
    correlation test.

    Parameters
    ----------
    x : array-like of shape (n_samples,)
            Training data.

    y : array-like of shape (n_samples,)
        Training target.

    Returns
    -------
    increasing_bool : boolean
        Whether the relationship is increasing or decreasing.

    Notes
    -----
    The Spearman correlation coefficient is estimated from the data, and the
    sign of the resulting estimate is used as the result.

    In the event that the 95% confidence interval based on Fisher transform
    spans zero, a warning is raised.

    References
    ----------
    Fisher transformation. Wikipedia.
    https://en.wikipedia.org/wiki/Fisher_transformation
    """

    # Calculate Spearman rho estimate and set return accordingly.
    rho, _ = spearmanr(x, y)
    increasing_bool = rho >= 0

    # Run Fisher transform to get the rho CI, but handle rho=+/-1
    if rho not in [-1.0, 1.0] and len(x) > 3:
        F = 0.5 * math.log((1. + rho) / (1. - rho))
        F_se = 1 / math.sqrt(len(x) - 3)

        # Use a 95% CI, i.e., +/-1.96 S.E.
        # https://en.wikipedia.org/wiki/Fisher_transformation
        rho_0 = math.tanh(F - 1.96 * F_se)
        rho_1 = math.tanh(F + 1.96 * F_se)

        # Warn if the CI spans zero.
        if np.sign(rho_0) != np.sign(rho_1):
            warnings.warn("Confidence interval of the Spearman "
                          "correlation coefficient spans zero. "
                          "Determination of ``increasing`` may be "
                          "suspect.")

    return increasing_bool


@_deprecate_positional_args
def isotonic_regression(y, *, sample_weight=None, y_min=None, y_max=None,
                        increasing=True):
    """Solve the isotonic regression model.

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

    Parameters
    ----------
    y : array-like of shape (n_samples,)
        The data.

    sample_weight : array-like of shape (n_samples,), default=None
        Weights on each point of the regression.
        If None, weight is set to 1 (equal weights).

    y_min : float, default=None
        Lower bound on the lowest predicted value (the minimum value may
        still be higher). If not set, defaults to -inf.

    y_max : float, default=None
        Upper bound on the highest predicted value (the maximum may still be
        lower). If not set, defaults to +inf.

    increasing : boolean, optional, default: True
        Whether to compute ``y_`` is increasing (if set to True) or decreasing
        (if set to False)

    Returns
    -------
    y_ : list of floats
        Isotonic fit of y.

    References
    ----------
    "Active set algorithms for isotonic regression; A unifying framework"
    by Michael J. Best and Nilotpal Chakravarti, section 3.
    """
    order = np.s_[:] if increasing else np.s_[::-1]
    y = check_array(y, ensure_2d=False, dtype=[np.float64, np.float32])
    y = np.array(y[order], dtype=y.dtype)
    sample_weight = _check_sample_weight(sample_weight, y, dtype=y.dtype)
    sample_weight = np.ascontiguousarray(sample_weight[order])

    _inplace_contiguous_isotonic_regression(y, sample_weight)
    if y_min is not None or y_max is not None:
        # Older versions of np.clip don't accept None as a bound, so use np.inf
        if y_min is None:
            y_min = -np.inf
        if y_max is None:
            y_max = np.inf
        np.clip(y, y_min, y_max, y)
    return y[order]


class IsotonicRegression(RegressorMixin, TransformerMixin, BaseEstimator):
    """Isotonic regression model.

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

    .. versionadded:: 0.13

    Parameters
    ----------
    y_min : float, default=None
        Lower bound on the lowest predicted value (the minimum value may
        still be higher). If not set, defaults to -inf.

    y_max : float, default=None
        Upper bound on the highest predicted value (the maximum may still be
        lower). If not set, defaults to +inf.

    increasing : bool or 'auto', default=True
        Determines whether the predictions should be constrained to increase
        or decrease with `X`. 'auto' will decide based on the Spearman
        correlation estimate's sign.

    out_of_bounds : str, default="nan"
        The ``out_of_bounds`` parameter handles how `X` values outside of the
        training domain are handled.  When set to "nan", predictions
        will be NaN.  When set to "clip", predictions will be
        set to the value corresponding to the nearest train interval endpoint.
        When set to "raise" a `ValueError` is raised.


    Attributes
    ----------
    X_min_ : float
        Minimum value of input array `X_` for left bound.

    X_max_ : float
        Maximum value of input array `X_` for right bound.

    f_ : function
        The stepwise interpolating function that covers the input domain ``X``.

    increasing_ : bool
        Inferred value for ``increasing``.

    Notes
    -----
    Ties are broken using the secondary method from Leeuw, 1977.

    References
    ----------
    Isotonic Median Regression: A Linear Programming Approach
    Nilotpal Chakravarti
    Mathematics of Operations Research
    Vol. 14, No. 2 (May, 1989), pp. 303-308

    Isotone Optimization in R : Pool-Adjacent-Violators
    Algorithm (PAVA) and Active Set Methods
    Leeuw, Hornik, Mair
    Journal of Statistical Software 2009

    Correctness of Kruskal's algorithms for monotone regression with ties
    Leeuw, Psychometrica, 1977

    Examples
    --------
    >>> from sklearn.datasets import make_regression
    >>> from sklearn.isotonic import IsotonicRegression
    >>> X, y = make_regression(n_samples=10, n_features=1, random_state=41)
    >>> iso_reg = IsotonicRegression().fit(X.flatten(), y)
    >>> iso_reg.predict([.1, .2])
    array([1.8628..., 3.7256...])
    """
    @_deprecate_positional_args
    def __init__(self, *, y_min=None, y_max=None, increasing=True,
                 out_of_bounds='nan'):
        self.y_min = y_min
        self.y_max = y_max
        self.increasing = increasing
        self.out_of_bounds = out_of_bounds

    def _check_fit_data(self, X, y, sample_weight=None):
        if len(X.shape) != 1:
            raise ValueError("X should be a 1d array")

    def _build_f(self, X, y):
        """Build the f_ interp1d function."""

        # Handle the out_of_bounds argument by setting bounds_error
        if self.out_of_bounds not in ["raise", "nan", "clip"]:
            raise ValueError("The argument ``out_of_bounds`` must be in "
                             "'nan', 'clip', 'raise'; got {0}"
                             .format(self.out_of_bounds))

        bounds_error = self.out_of_bounds == "raise"
        if len(y) == 1:
            # single y, constant prediction
            self.f_ = lambda x: y.repeat(x.shape)
        else:
            self.f_ = interpolate.interp1d(X, y, kind='linear',
                                           bounds_error=bounds_error)

    def _build_y(self, X, y, sample_weight, trim_duplicates=True):
        """Build the y_ IsotonicRegression."""
        self._check_fit_data(X, y, sample_weight)

        # Determine increasing if auto-determination requested
        if self.increasing == 'auto':
            self.increasing_ = check_increasing(X, y)
        else:
            self.increasing_ = self.increasing

        # If sample_weights is passed, removed zero-weight values and clean
        # order
        sample_weight = _check_sample_weight(sample_weight, X, dtype=X.dtype)
        mask = sample_weight > 0
        X, y, sample_weight = X[mask], y[mask], sample_weight[mask]

        order = np.lexsort((y, X))
        X, y, sample_weight = [array[order] for array in [X, y, sample_weight]]
        unique_X, unique_y, unique_sample_weight = _make_unique(
            X, y, sample_weight)

        X = unique_X
        y = isotonic_regression(unique_y, sample_weight=unique_sample_weight,
                                y_min=self.y_min, y_max=self.y_max,
                                increasing=self.increasing_)

        # Handle the left and right bounds on X
        self.X_min_, self.X_max_ = np.min(X), np.max(X)

        if trim_duplicates:
            # Remove unnecessary points for faster prediction
            keep_data = np.ones((len(y),), dtype=bool)
            # Aside from the 1st and last point, remove points whose y values
            # are equal to both the point before and the point after it.
            keep_data[1:-1] = np.logical_or(
                np.not_equal(y[1:-1], y[:-2]),
                np.not_equal(y[1:-1], y[2:])
            )
            return X[keep_data], y[keep_data]
        else:
            # The ability to turn off trim_duplicates is only used to it make
            # easier to unit test that removing duplicates in y does not have
            # any impact the resulting interpolation function (besides
            # prediction speed).
            return X, y

    def fit(self, X, y, sample_weight=None):
        """Fit the model using X, y as training data.

        Parameters
        ----------
        X : array-like of shape (n_samples,)
            Training data.

        y : array-like of shape (n_samples,)
            Training target.

        sample_weight : array-like of shape (n_samples,), default=None
            Weights. If set to None, all weights will be set to 1 (equal
            weights).

        Returns
        -------
        self : object
            Returns an instance of self.

        Notes
        -----
        X is stored for future use, as :meth:`transform` needs X to interpolate
        new input data.
        """
        check_params = dict(accept_sparse=False, ensure_2d=False)
        X = check_array(X, dtype=[np.float64, np.float32], **check_params)
        y = check_array(y, dtype=X.dtype, **check_params)
        check_consistent_length(X, y, sample_weight)

        # Transform y by running the isotonic regression algorithm and
        # transform X accordingly.
        X, y = self._build_y(X, y, sample_weight)

        # It is necessary to store the non-redundant part of the training set
        # on the model to make it possible to support model persistence via
        # the pickle module as the object built by scipy.interp1d is not
        # picklable directly.
        self._necessary_X_, self._necessary_y_ = X, y

        # Build the interpolation function
        self._build_f(X, y)
        return self

    def transform(self, T):
        """Transform new data by linear interpolation

        Parameters
        ----------
        T : array-like of shape (n_samples,)
            Data to transform.

        Returns
        -------
        y_pred : ndarray of shape (n_samples,)
            The transformed data
        """

        if hasattr(self, '_necessary_X_'):
            dtype = self._necessary_X_.dtype
        else:
            dtype = np.float64

        T = check_array(T, dtype=dtype, ensure_2d=False)

        if len(T.shape) != 1:
            raise ValueError("Isotonic regression input should be a 1d array")

        # Handle the out_of_bounds argument by clipping if needed
        if self.out_of_bounds not in ["raise", "nan", "clip"]:
            raise ValueError("The argument ``out_of_bounds`` must be in "
                             "'nan', 'clip', 'raise'; got {0}"
                             .format(self.out_of_bounds))

        if self.out_of_bounds == "clip":
            T = np.clip(T, self.X_min_, self.X_max_)

        res = self.f_(T)

        # on scipy 0.17, interp1d up-casts to float64, so we cast back
        res = res.astype(T.dtype)

        return res

    def predict(self, T):
        """Predict new data by linear interpolation.

        Parameters
        ----------
        T : array-like of shape (n_samples,)
            Data to transform.

        Returns
        -------
        y_pred : ndarray of shape (n_samples,)
            Transformed data.
        """
        return self.transform(T)

    def __getstate__(self):
        """Pickle-protocol - return state of the estimator. """
        state = super().__getstate__()
        # remove interpolation method
        state.pop('f_', None)
        return state

    def __setstate__(self, state):
        """Pickle-protocol - set state of the estimator.

        We need to rebuild the interpolation function.
        """
        super().__setstate__(state)
        if hasattr(self, '_necessary_X_') and hasattr(self, '_necessary_y_'):
            self._build_f(self._necessary_X_, self._necessary_y_)

    def _more_tags(self):
        return {'X_types': ['1darray']}