Source code for matplotlib.colors

"""
A module for converting numbers or color arguments to *RGB* or *RGBA*.

*RGB* and *RGBA* are sequences of, respectively, 3 or 4 floats in the
range 0-1.

This module includes functions and classes for color specification conversions,
and for mapping numbers to colors in a 1-D array of colors called a colormap.

Mapping data onto colors using a colormap typically involves two steps: a data
array is first mapped onto the range 0-1 using a subclass of `Normalize`,
then this number is mapped to a color using a subclass of `Colormap`.  Two
subclasses of `Colormap` provided here:  `LinearSegmentedColormap`, which uses
piecewise-linear interpolation to define colormaps, and `ListedColormap`, which
makes a colormap from a list of colors.

.. seealso::

  :doc:`/tutorials/colors/colormap-manipulation` for examples of how to
  make colormaps and

  :doc:`/tutorials/colors/colormaps` for a list of built-in colormaps.

  :doc:`/tutorials/colors/colormapnorms` for more details about data
  normalization

  More colormaps are available at palettable_.

The module also provides functions for checking whether an object can be
interpreted as a color (`is_color_like`), for converting such an object
to an RGBA tuple (`to_rgba`) or to an HTML-like hex string in the
"#rrggbb" format (`to_hex`), and a sequence of colors to an (n, 4)
RGBA array (`to_rgba_array`).  Caching is used for efficiency.

Colors that Matplotlib recognizes are listed at
:doc:`/tutorials/colors/colors`.

.. _palettable: https://jiffyclub.github.io/palettable/
.. _xkcd color survey: https://xkcd.com/color/rgb/
"""

import base64
from collections.abc import Sized, Sequence, Mapping
import functools
import importlib
import inspect
import io
import itertools
from numbers import Number
import re
from PIL import Image
from PIL.PngImagePlugin import PngInfo

import matplotlib as mpl
import numpy as np
from matplotlib import _api, _cm, cbook, scale
from ._color_data import BASE_COLORS, TABLEAU_COLORS, CSS4_COLORS, XKCD_COLORS


class _ColorMapping(dict):
    def __init__(self, mapping):
        super().__init__(mapping)
        self.cache = {}

    def __setitem__(self, key, value):
        super().__setitem__(key, value)
        self.cache.clear()

    def __delitem__(self, key):
        super().__delitem__(key)
        self.cache.clear()


_colors_full_map = {}
# Set by reverse priority order.
_colors_full_map.update(XKCD_COLORS)
_colors_full_map.update({k.replace('grey', 'gray'): v
                         for k, v in XKCD_COLORS.items()
                         if 'grey' in k})
_colors_full_map.update(CSS4_COLORS)
_colors_full_map.update(TABLEAU_COLORS)
_colors_full_map.update({k.replace('gray', 'grey'): v
                         for k, v in TABLEAU_COLORS.items()
                         if 'gray' in k})
_colors_full_map.update(BASE_COLORS)
_colors_full_map = _ColorMapping(_colors_full_map)

_REPR_PNG_SIZE = (512, 64)


def get_named_colors_mapping():
    """Return the global mapping of names to named colors."""
    return _colors_full_map


class ColorSequenceRegistry(Mapping):
    r"""
    Container for sequences of colors that are known to Matplotlib by name.

    The universal registry instance is `matplotlib.color_sequences`. There
    should be no need for users to instantiate `.ColorSequenceRegistry`
    themselves.

    Read access uses a dict-like interface mapping names to lists of colors::

        import matplotlib as mpl
        cmap = mpl.color_sequences['tab10']

    The returned lists are copies, so that their modification does not change
    the global definition of the color sequence.

    Additional color sequences can be added via
    `.ColorSequenceRegistry.register`::

        mpl.color_sequences.register('rgb', ['r', 'g', 'b'])
    """

    _BUILTIN_COLOR_SEQUENCES = {
        'tab10': _cm._tab10_data,
        'tab20': _cm._tab20_data,
        'tab20b': _cm._tab20b_data,
        'tab20c': _cm._tab20c_data,
        'Pastel1': _cm._Pastel1_data,
        'Pastel2': _cm._Pastel2_data,
        'Paired': _cm._Paired_data,
        'Accent': _cm._Accent_data,
        'Dark2': _cm._Dark2_data,
        'Set1': _cm._Set1_data,
        'Set2': _cm._Set1_data,
        'Set3': _cm._Set1_data,
    }

    def __init__(self):
        self._color_sequences = {**self._BUILTIN_COLOR_SEQUENCES}

    def __getitem__(self, item):
        try:
            return list(self._color_sequences[item])
        except KeyError:
            raise KeyError(f"{item!r} is not a known color sequence name")

    def __iter__(self):
        return iter(self._color_sequences)

    def __len__(self):
        return len(self._color_sequences)

    def __str__(self):
        return ('ColorSequenceRegistry; available colormaps:\n' +
                ', '.join(f"'{name}'" for name in self))

    def register(self, name, color_list):
        """
        Register a new color sequence.

        The color sequence registry stores a copy of the given *color_list*, so
        that future changes to the original list do not affect the registered
        color sequence. Think of this as the registry taking a snapshot
        of *color_list* at registration.

        Parameters
        ----------
        name : str
            The name for the color sequence.

        color_list : list of colors
            An iterable returning valid Matplotlib colors when iterating over.
            Note however that the returned color sequence will always be a
            list regardless of the input type.

        """
        if name in self._BUILTIN_COLOR_SEQUENCES:
            raise ValueError(f"{name!r} is a reserved name for a builtin "
                             "color sequence")

        color_list = list(color_list)  # force copy and coerce type to list
        for color in color_list:
            try:
                to_rgba(color)
            except ValueError:
                raise ValueError(
                    f"{color!r} is not a valid color specification")

        self._color_sequences[name] = color_list

    def unregister(self, name):
        """
        Remove a sequence from the registry.

        You cannot remove built-in color sequences.

        If the name is not registered, returns with no error.
        """
        if name in self._BUILTIN_COLOR_SEQUENCES:
            raise ValueError(
                f"Cannot unregister builtin color sequence {name!r}")
        self._color_sequences.pop(name, None)


_color_sequences = ColorSequenceRegistry()


def _sanitize_extrema(ex):
    if ex is None:
        return ex
    try:
        ret = ex.item()
    except AttributeError:
        ret = float(ex)
    return ret


def _is_nth_color(c):
    """Return whether *c* can be interpreted as an item in the color cycle."""
    return isinstance(c, str) and re.match(r"\AC[0-9]+\Z", c)


def is_color_like(c):
    """Return whether *c* can be interpreted as an RGB(A) color."""
    # Special-case nth color syntax because it cannot be parsed during setup.
    if _is_nth_color(c):
        return True
    try:
        to_rgba(c)
    except ValueError:
        return False
    else:
        return True


def _has_alpha_channel(c):
    """Return whether *c* is a color with an alpha channel."""
    # 4-element sequences are interpreted as r, g, b, a
    return not isinstance(c, str) and len(c) == 4


def _check_color_like(**kwargs):
    """
    For each *key, value* pair in *kwargs*, check that *value* is color-like.
    """
    for k, v in kwargs.items():
        if not is_color_like(v):
            raise ValueError(f"{v!r} is not a valid value for {k}")


def same_color(c1, c2):
    """
    Return whether the colors *c1* and *c2* are the same.

    *c1*, *c2* can be single colors or lists/arrays of colors.
    """
    c1 = to_rgba_array(c1)
    c2 = to_rgba_array(c2)
    n1 = max(c1.shape[0], 1)  # 'none' results in shape (0, 4), but is 1-elem
    n2 = max(c2.shape[0], 1)  # 'none' results in shape (0, 4), but is 1-elem

    if n1 != n2:
        raise ValueError('Different number of elements passed.')
    # The following shape test is needed to correctly handle comparisons with
    # 'none', which results in a shape (0, 4) array and thus cannot be tested
    # via value comparison.
    return c1.shape == c2.shape and (c1 == c2).all()


def to_rgba(c, alpha=None):
    """
    Convert *c* to an RGBA color.

    Parameters
    ----------
    c : Matplotlib color or ``np.ma.masked``

    alpha : float, optional
        If *alpha* is given, force the alpha value of the returned RGBA tuple
        to *alpha*.

        If None, the alpha value from *c* is used. If *c* does not have an
        alpha channel, then alpha defaults to 1.

        *alpha* is ignored for the color value ``"none"`` (case-insensitive),
        which always maps to ``(0, 0, 0, 0)``.

    Returns
    -------
    tuple
        Tuple of floats ``(r, g, b, a)``, where each channel (red, green, blue,
        alpha) can assume values between 0 and 1.
    """
    # Special-case nth color syntax because it should not be cached.
    if _is_nth_color(c):
        prop_cycler = mpl.rcParams['axes.prop_cycle']
        colors = prop_cycler.by_key().get('color', ['k'])
        c = colors[int(c[1:]) % len(colors)]
    try:
        rgba = _colors_full_map.cache[c, alpha]
    except (KeyError, TypeError):  # Not in cache, or unhashable.
        rgba = None
    if rgba is None:  # Suppress exception chaining of cache lookup failure.
        rgba = _to_rgba_no_colorcycle(c, alpha)
        try:
            _colors_full_map.cache[c, alpha] = rgba
        except TypeError:
            pass
    return rgba


def _to_rgba_no_colorcycle(c, alpha=None):
    """
    Convert *c* to an RGBA color, with no support for color-cycle syntax.

    If *alpha* is given, force the alpha value of the returned RGBA tuple
    to *alpha*. Otherwise, the alpha value from *c* is used, if it has alpha
    information, or defaults to 1.

    *alpha* is ignored for the color value ``"none"`` (case-insensitive),
    which always maps to ``(0, 0, 0, 0)``.
    """
    orig_c = c
    if c is np.ma.masked:
        return (0., 0., 0., 0.)
    if isinstance(c, str):
        if c.lower() == "none":
            return (0., 0., 0., 0.)
        # Named color.
        try:
            # This may turn c into a non-string, so we check again below.
            c = _colors_full_map[c]
        except KeyError:
            if len(orig_c) != 1:
                try:
                    c = _colors_full_map[c.lower()]
                except KeyError:
                    pass
    if isinstance(c, str):
        # hex color in #rrggbb format.
        match = re.match(r"\A#[a-fA-F0-9]{6}\Z", c)
        if match:
            return (tuple(int(n, 16) / 255
                          for n in [c[1:3], c[3:5], c[5:7]])
                    + (alpha if alpha is not None else 1.,))
        # hex color in #rgb format, shorthand for #rrggbb.
        match = re.match(r"\A#[a-fA-F0-9]{3}\Z", c)
        if match:
            return (tuple(int(n, 16) / 255
                          for n in [c[1]*2, c[2]*2, c[3]*2])
                    + (alpha if alpha is not None else 1.,))
        # hex color with alpha in #rrggbbaa format.
        match = re.match(r"\A#[a-fA-F0-9]{8}\Z", c)
        if match:
            color = [int(n, 16) / 255
                     for n in [c[1:3], c[3:5], c[5:7], c[7:9]]]
            if alpha is not None:
                color[-1] = alpha
            return tuple(color)
        # hex color with alpha in #rgba format, shorthand for #rrggbbaa.
        match = re.match(r"\A#[a-fA-F0-9]{4}\Z", c)
        if match:
            color = [int(n, 16) / 255
                     for n in [c[1]*2, c[2]*2, c[3]*2, c[4]*2]]
            if alpha is not None:
                color[-1] = alpha
            return tuple(color)
        # string gray.
        try:
            c = float(c)
        except ValueError:
            pass
        else:
            if not (0 <= c <= 1):
                raise ValueError(
                    f"Invalid string grayscale value {orig_c!r}. "
                    f"Value must be within 0-1 range")
            return c, c, c, alpha if alpha is not None else 1.
        raise ValueError(f"Invalid RGBA argument: {orig_c!r}")
    # turn 2-D array into 1-D array
    if isinstance(c, np.ndarray):
        if c.ndim == 2 and c.shape[0] == 1:
            c = c.reshape(-1)
    # tuple color.
    if not np.iterable(c):
        raise ValueError(f"Invalid RGBA argument: {orig_c!r}")
    if len(c) not in [3, 4]:
        raise ValueError("RGBA sequence should have length 3 or 4")
    if not all(isinstance(x, Number) for x in c):
        # Checks that don't work: `map(float, ...)`, `np.array(..., float)` and
        # `np.array(...).astype(float)` would all convert "0.5" to 0.5.
        raise ValueError(f"Invalid RGBA argument: {orig_c!r}")
    # Return a tuple to prevent the cached value from being modified.
    c = tuple(map(float, c))
    if len(c) == 3 and alpha is None:
        alpha = 1
    if alpha is not None:
        c = c[:3] + (alpha,)
    if any(elem < 0 or elem > 1 for elem in c):
        raise ValueError("RGBA values should be within 0-1 range")
    return c


def to_rgba_array(c, alpha=None):
    """
    Convert *c* to a (n, 4) array of RGBA colors.

    Parameters
    ----------
    c : Matplotlib color or array of colors
        If *c* is a masked array, an `~numpy.ndarray` is returned with a
        (0, 0, 0, 0) row for each masked value or row in *c*.

    alpha : float or sequence of floats, optional
        If *alpha* is given, force the alpha value of the returned RGBA tuple
        to *alpha*.

        If None, the alpha value from *c* is used. If *c* does not have an
        alpha channel, then alpha defaults to 1.

        *alpha* is ignored for the color value ``"none"`` (case-insensitive),
        which always maps to ``(0, 0, 0, 0)``.

        If *alpha* is a sequence and *c* is a single color, *c* will be
        repeated to match the length of *alpha*.

    Returns
    -------
    array
        (n, 4) array of RGBA colors,  where each channel (red, green, blue,
        alpha) can assume values between 0 and 1.
    """
    # Special-case inputs that are already arrays, for performance.  (If the
    # array has the wrong kind or shape, raise the error during one-at-a-time
    # conversion.)
    if np.iterable(alpha):
        alpha = np.asarray(alpha).ravel()
    if (isinstance(c, np.ndarray) and c.dtype.kind in "if"
            and c.ndim == 2 and c.shape[1] in [3, 4]):
        mask = c.mask.any(axis=1) if np.ma.is_masked(c) else None
        c = np.ma.getdata(c)
        if np.iterable(alpha):
            if c.shape[0] == 1 and alpha.shape[0] > 1:
                c = np.tile(c, (alpha.shape[0], 1))
            elif c.shape[0] != alpha.shape[0]:
                raise ValueError("The number of colors must match the number"
                                 " of alpha values if there are more than one"
                                 " of each.")
        if c.shape[1] == 3:
            result = np.column_stack([c, np.zeros(len(c))])
            result[:, -1] = alpha if alpha is not None else 1.
        elif c.shape[1] == 4:
            result = c.copy()
            if alpha is not None:
                result[:, -1] = alpha
        if mask is not None:
            result[mask] = 0
        if np.any((result < 0) | (result > 1)):
            raise ValueError("RGBA values should be within 0-1 range")
        return result
    # Handle single values.
    # Note that this occurs *after* handling inputs that are already arrays, as
    # `to_rgba(c, alpha)` (below) is expensive for such inputs, due to the need
    # to format the array in the ValueError message(!).
    if cbook._str_lower_equal(c, "none"):
        return np.zeros((0, 4), float)
    try:
        if np.iterable(alpha):
            return np.array([to_rgba(c, a) for a in alpha], float)
        else:
            return np.array([to_rgba(c, alpha)], float)
    except (ValueError, TypeError):
        pass

    if isinstance(c, str):
        raise ValueError(f"{c!r} is not a valid color value.")

    if len(c) == 0:
        return np.zeros((0, 4), float)

    # Quick path if the whole sequence can be directly converted to a numpy
    # array in one shot.
    if isinstance(c, Sequence):
        lens = {len(cc) if isinstance(cc, (list, tuple)) else -1 for cc in c}
        if lens == {3}:
            rgba = np.column_stack([c, np.ones(len(c))])
        elif lens == {4}:
            rgba = np.array(c)
        else:
            rgba = np.array([to_rgba(cc) for cc in c])
    else:
        rgba = np.array([to_rgba(cc) for cc in c])

    if alpha is not None:
        rgba[:, 3] = alpha
    return rgba


def to_rgb(c):
    """Convert *c* to an RGB color, silently dropping the alpha channel."""
    return to_rgba(c)[:3]


def to_hex(c, keep_alpha=False):
    """
    Convert *c* to a hex color.

    Parameters
    ----------
    c : :doc:`color </tutorials/colors/colors>` or `numpy.ma.masked`

    keep_alpha : bool, default: False
      If False, use the ``#rrggbb`` format, otherwise use ``#rrggbbaa``.

    Returns
    -------
    str
      ``#rrggbb`` or ``#rrggbbaa`` hex color string
    """
    c = to_rgba(c)
    if not keep_alpha:
        c = c[:3]
    return "#" + "".join(format(round(val * 255), "02x") for val in c)


### Backwards-compatible color-conversion API


cnames = CSS4_COLORS
hexColorPattern = re.compile(r"\A#[a-fA-F0-9]{6}\Z")
rgb2hex = to_hex
hex2color = to_rgb


class ColorConverter:
    """
    A class only kept for backwards compatibility.

    Its functionality is entirely provided by module-level functions.
    """
    colors = _colors_full_map
    cache = _colors_full_map.cache
    to_rgb = staticmethod(to_rgb)
    to_rgba = staticmethod(to_rgba)
    to_rgba_array = staticmethod(to_rgba_array)


colorConverter = ColorConverter()


### End of backwards-compatible color-conversion API


def _create_lookup_table(N, data, gamma=1.0):
    r"""
    Create an *N* -element 1D lookup table.

    This assumes a mapping :math:`f : [0, 1] \rightarrow [0, 1]`. The returned
    data is an array of N values :math:`y = f(x)` where x is sampled from
    [0, 1].

    By default (*gamma* = 1) x is equidistantly sampled from [0, 1]. The
    *gamma* correction factor :math:`\gamma` distorts this equidistant
    sampling by :math:`x \rightarrow x^\gamma`.

    Parameters
    ----------
    N : int
        The number of elements of the created lookup table; at least 1.

    data : (M, 3) array-like or callable
        Defines the mapping :math:`f`.

        If a (M, 3) array-like, the rows define values (x, y0, y1).  The x
        values must start with x=0, end with x=1, and all x values be in
        increasing order.

        A value between :math:`x_i` and :math:`x_{i+1}` is mapped to the range
        :math:`y^1_{i-1} \ldots y^0_i` by linear interpolation.

        For the simple case of a y-continuous mapping, y0 and y1 are identical.

        The two values of y are to allow for discontinuous mapping functions.
        E.g. a sawtooth with a period of 0.2 and an amplitude of 1 would be::

            [(0, 1, 0), (0.2, 1, 0), (0.4, 1, 0), ..., [(1, 1, 0)]

        In the special case of ``N == 1``, by convention the returned value
        is y0 for x == 1.

        If *data* is a callable, it must accept and return numpy arrays::

           data(x : ndarray) -> ndarray

        and map values between 0 - 1 to 0 - 1.

    gamma : float
        Gamma correction factor for input distribution x of the mapping.

        See also https://en.wikipedia.org/wiki/Gamma_correction.

    Returns
    -------
    array
        The lookup table where ``lut[x * (N-1)]`` gives the closest value
        for values of x between 0 and 1.

    Notes
    -----
    This function is internally used for `.LinearSegmentedColormap`.
    """

    if callable(data):
        xind = np.linspace(0, 1, N) ** gamma
        lut = np.clip(np.array(data(xind), dtype=float), 0, 1)
        return lut

    try:
        adata = np.array(data)
    except Exception as err:
        raise TypeError("data must be convertible to an array") from err
    _api.check_shape((None, 3), data=adata)

    x = adata[:, 0]
    y0 = adata[:, 1]
    y1 = adata[:, 2]

    if x[0] != 0. or x[-1] != 1.0:
        raise ValueError(
            "data mapping points must start with x=0 and end with x=1")
    if (np.diff(x) < 0).any():
        raise ValueError("data mapping points must have x in increasing order")
    # begin generation of lookup table
    if N == 1:
        # convention: use the y = f(x=1) value for a 1-element lookup table
        lut = np.array(y0[-1])
    else:
        x = x * (N - 1)
        xind = (N - 1) * np.linspace(0, 1, N) ** gamma
        ind = np.searchsorted(x, xind)[1:-1]

        distance = (xind[1:-1] - x[ind - 1]) / (x[ind] - x[ind - 1])
        lut = np.concatenate([
            [y1[0]],
            distance * (y0[ind] - y1[ind - 1]) + y1[ind - 1],
            [y0[-1]],
        ])
    # ensure that the lut is confined to values between 0 and 1 by clipping it
    return np.clip(lut, 0.0, 1.0)


class Colormap:
    """
    Baseclass for all scalar to RGBA mappings.

    Typically, Colormap instances are used to convert data values (floats)
    from the interval ``[0, 1]`` to the RGBA color that the respective
    Colormap represents. For scaling of data into the ``[0, 1]`` interval see
    `matplotlib.colors.Normalize`. Subclasses of `matplotlib.cm.ScalarMappable`
    make heavy use of this ``data -> normalize -> map-to-color`` processing
    chain.
    """

    def __init__(self, name, N=256):
        """
        Parameters
        ----------
        name : str
            The name of the colormap.
        N : int
            The number of RGB quantization levels.
        """
        self.name = name
        self.N = int(N)  # ensure that N is always int
        self._rgba_bad = (0.0, 0.0, 0.0, 0.0)  # If bad, don't paint anything.
        self._rgba_under = None
        self._rgba_over = None
        self._i_under = self.N
        self._i_over = self.N + 1
        self._i_bad = self.N + 2
        self._isinit = False
        #: When this colormap exists on a scalar mappable and colorbar_extend
        #: is not False, colorbar creation will pick up ``colorbar_extend`` as
        #: the default value for the ``extend`` keyword in the
        #: `matplotlib.colorbar.Colorbar` constructor.
        self.colorbar_extend = False

    def __call__(self, X, alpha=None, bytes=False):
        """
        Parameters
        ----------
        X : float or int, `~numpy.ndarray` or scalar
            The data value(s) to convert to RGBA.
            For floats, *X* should be in the interval ``[0.0, 1.0]`` to
            return the RGBA values ``X*100`` percent along the Colormap line.
            For integers, *X* should be in the interval ``[0, Colormap.N)`` to
            return RGBA values *indexed* from the Colormap with index ``X``.
        alpha : float or array-like or None
            Alpha must be a scalar between 0 and 1, a sequence of such
            floats with shape matching X, or None.
        bytes : bool
            If False (default), the returned RGBA values will be floats in the
            interval ``[0, 1]`` otherwise they will be uint8s in the interval
            ``[0, 255]``.

        Returns
        -------
        Tuple of RGBA values if X is scalar, otherwise an array of
        RGBA values with a shape of ``X.shape + (4, )``.
        """
        if not self._isinit:
            self._init()

        # Take the bad mask from a masked array, or in all other cases defer
        # np.isnan() to after we have converted to an array.
        mask_bad = X.mask if np.ma.is_masked(X) else None
        xa = np.array(X, copy=True)
        if mask_bad is None:
            mask_bad = np.isnan(xa)
        if not xa.dtype.isnative:
            xa = xa.byteswap().newbyteorder()  # Native byteorder is faster.
        if xa.dtype.kind == "f":
            xa *= self.N
            # Negative values are out of range, but astype(int) would
            # truncate them towards zero.
            xa[xa < 0] = -1
            # xa == 1 (== N after multiplication) is not out of range.
            xa[xa == self.N] = self.N - 1
            # Avoid converting large positive values to negative integers.
            np.clip(xa, -1, self.N, out=xa)
        with np.errstate(invalid="ignore"):
            # We need this cast for unsigned ints as well as floats
            xa = xa.astype(int)
        # Set the over-range indices before the under-range;
        # otherwise the under-range values get converted to over-range.
        xa[xa > self.N - 1] = self._i_over
        xa[xa < 0] = self._i_under
        xa[mask_bad] = self._i_bad

        lut = self._lut
        if bytes:
            lut = (lut * 255).astype(np.uint8)

        rgba = lut.take(xa, axis=0, mode='clip')

        if alpha is not None:
            alpha = np.clip(alpha, 0, 1)
            if bytes:
                alpha *= 255  # Will be cast to uint8 upon assignment.
            if alpha.shape not in [(), xa.shape]:
                raise ValueError(
                    f"alpha is array-like but its shape {alpha.shape} does "
                    f"not match that of X {xa.shape}")
            rgba[..., -1] = alpha

            # If the "bad" color is all zeros, then ignore alpha input.
            if (lut[-1] == 0).all() and np.any(mask_bad):
                if np.iterable(mask_bad) and mask_bad.shape == xa.shape:
                    rgba[mask_bad] = (0, 0, 0, 0)
                else:
                    rgba[..., :] = (0, 0, 0, 0)

        if not np.iterable(X):
            rgba = tuple(rgba)
        return rgba

    def __copy__(self):
        cls = self.__class__
        cmapobject = cls.__new__(cls)
        cmapobject.__dict__.update(self.__dict__)
        if self._isinit:
            cmapobject._lut = np.copy(self._lut)
        return cmapobject

    def __eq__(self, other):
        if (not isinstance(other, Colormap) or self.name != other.name or
                self.colorbar_extend != other.colorbar_extend):
            return False
        # To compare lookup tables the Colormaps have to be initialized
        if not self._isinit:
            self._init()
        if not other._isinit:
            other._init()
        return np.array_equal(self._lut, other._lut)

[docs]    def get_bad(self):
        """Get the color for masked values."""
        if not self._isinit:
            self._init()
        return np.array(self._lut[self._i_bad])

[docs]    def set_bad(self, color='k', alpha=None):
        """Set the color for masked values."""
        self._rgba_bad = to_rgba(color, alpha)
        if self._isinit:
            self._set_extremes()

[docs]    def get_under(self):
        """Get the color for low out-of-range values."""
        if not self._isinit:
            self._init()
        return np.array(self._lut[self._i_under])

[docs]    def set_under(self, color='k', alpha=None):
        """Set the color for low out-of-range values."""
        self._rgba_under = to_rgba(color, alpha)
        if self._isinit:
            self._set_extremes()

[docs]    def get_over(self):
        """Get the color for high out-of-range values."""
        if not self._isinit:
            self._init()
        return np.array(self._lut[self._i_over])

[docs]    def set_over(self, color='k', alpha=None):
        """Set the color for high out-of-range values."""
        self._rgba_over = to_rgba(color, alpha)
        if self._isinit:
            self._set_extremes()

[docs]    def set_extremes(self, *, bad=None, under=None, over=None):
        """
        Set the colors for masked (*bad*) values and, when ``norm.clip =
        False``, low (*under*) and high (*over*) out-of-range values.
        """
        if bad is not None:
            self.set_bad(bad)
        if under is not None:
            self.set_under(under)
        if over is not None:
            self.set_over(over)

[docs]    def with_extremes(self, *, bad=None, under=None, over=None):
        """
        Return a copy of the colormap, for which the colors for masked (*bad*)
        values and, when ``norm.clip = False``, low (*under*) and high (*over*)
        out-of-range values, have been set accordingly.
        """
        new_cm = self.copy()
        new_cm.set_extremes(bad=bad, under=under, over=over)
        return new_cm

    def _set_extremes(self):
        if self._rgba_under:
            self._lut[self._i_under] = self._rgba_under
        else:
            self._lut[self._i_under] = self._lut[0]
        if self._rgba_over:
            self._lut[self._i_over] = self._rgba_over
        else:
            self._lut[self._i_over] = self._lut[self.N - 1]
        self._lut[self._i_bad] = self._rgba_bad

    def _init(self):
        """Generate the lookup table, ``self._lut``."""
        raise NotImplementedError("Abstract class only")

[docs]    def is_gray(self):
        """Return whether the colormap is grayscale."""
        if not self._isinit:
            self._init()
        return (np.all(self._lut[:, 0] == self._lut[:, 1]) and
                np.all(self._lut[:, 0] == self._lut[:, 2]))

[docs]    def resampled(self, lutsize):
        """Return a new colormap with *lutsize* entries."""
        if hasattr(self, '_resample'):
            _api.warn_external(
                "The ability to resample a color map is now public API "
                f"However the class {type(self)} still only implements "
                "the previous private _resample method.  Please update "
                "your class."
            )
            return self._resample(lutsize)

        raise NotImplementedError()

    def reversed(self, name=None):
        """
        Return a reversed instance of the Colormap.

        .. note:: This function is not implemented for the base class.

        Parameters
        ----------
        name : str, optional
            The name for the reversed colormap. If None, the
            name is set to ``self.name + "_r"``.

        See Also
        --------
        LinearSegmentedColormap.reversed
        ListedColormap.reversed
        """
        raise NotImplementedError()

    def _repr_png_(self):
        """Generate a PNG representation of the Colormap."""
        X = np.tile(np.linspace(0, 1, _REPR_PNG_SIZE[0]),
                    (_REPR_PNG_SIZE[1], 1))
        pixels = self(X, bytes=True)
        png_bytes = io.BytesIO()
        title = self.name + ' colormap'
        author = f'Matplotlib v{mpl.__version__}, https://matplotlib.org'
        pnginfo = PngInfo()
        pnginfo.add_text('Title', title)
        pnginfo.add_text('Description', title)
        pnginfo.add_text('Author', author)
        pnginfo.add_text('Software', author)
        Image.fromarray(pixels).save(png_bytes, format='png', pnginfo=pnginfo)
        return png_bytes.getvalue()

    def _repr_html_(self):
        """Generate an HTML representation of the Colormap."""
        png_bytes = self._repr_png_()
        png_base64 = base64.b64encode(png_bytes).decode('ascii')
        def color_block(color):
            hex_color = to_hex(color, keep_alpha=True)
            return (f'<div title="{hex_color}" '
                    'style="display: inline-block; '
                    'width: 1em; height: 1em; '
                    'margin: 0; '
                    'vertical-align: middle; '
                    'border: 1px solid #555; '
                    f'background-color: {hex_color};"></div>')

        return ('<div style="vertical-align: middle;">'
                f'<strong>{self.name}</strong> '
                '</div>'
                '<div class="cmap"><img '
                f'alt="{self.name} colormap" '
                f'title="{self.name}" '
                'style="border: 1px solid #555;" '
                f'src="data:image/png;base64,{png_base64}"></div>'
                '<div style="vertical-align: middle; '
                f'max-width: {_REPR_PNG_SIZE[0]+2}px; '
                'display: flex; justify-content: space-between;">'
                '<div style="float: left;">'
                f'{color_block(self.get_under())} under'
                '</div>'
                '<div style="margin: 0 auto; display: inline-block;">'
                f'bad {color_block(self.get_bad())}'
                '</div>'
                '<div style="float: right;">'
                f'over {color_block(self.get_over())}'
                '</div>')

[docs]    def copy(self):
        """Return a copy of the colormap."""
        return self.__copy__()


class LinearSegmentedColormap(Colormap):
    """
    Colormap objects based on lookup tables using linear segments.

    The lookup table is generated using linear interpolation for each
    primary color, with the 0-1 domain divided into any number of
    segments.
    """

    def __init__(self, name, segmentdata, N=256, gamma=1.0):
        """
        Create colormap from linear mapping segments

        segmentdata argument is a dictionary with a red, green and blue
        entries. Each entry should be a list of *x*, *y0*, *y1* tuples,
        forming rows in a table. Entries for alpha are optional.

        Example: suppose you want red to increase from 0 to 1 over
        the bottom half, green to do the same over the middle half,
        and blue over the top half.  Then you would use::

            cdict = {'red':   [(0.0,  0.0, 0.0),
                               (0.5,  1.0, 1.0),
                               (1.0,  1.0, 1.0)],

                     'green': [(0.0,  0.0, 0.0),
                               (0.25, 0.0, 0.0),
                               (0.75, 1.0, 1.0),
                               (1.0,  1.0, 1.0)],

                     'blue':  [(0.0,  0.0, 0.0),
                               (0.5,  0.0, 0.0),
                               (1.0,  1.0, 1.0)]}

        Each row in the table for a given color is a sequence of
        *x*, *y0*, *y1* tuples.  In each sequence, *x* must increase
        monotonically from 0 to 1.  For any input value *z* falling
        between *x[i]* and *x[i+1]*, the output value of a given color
        will be linearly interpolated between *y1[i]* and *y0[i+1]*::

            row i:   x  y0  y1
                           /
                          /
            row i+1: x  y0  y1

        Hence y0 in the first row and y1 in the last row are never used.

        See Also
        --------
        LinearSegmentedColormap.from_list
            Static method; factory function for generating a smoothly-varying
            LinearSegmentedColormap.
        """
        # True only if all colors in map are identical; needed for contouring.
        self.monochrome = False
        super().__init__(name, N)
        self._segmentdata = segmentdata
        self._gamma = gamma

    def _init(self):
        self._lut = np.ones((self.N + 3, 4), float)
        self._lut[:-3, 0] = _create_lookup_table(
            self.N, self._segmentdata['red'], self._gamma)
        self._lut[:-3, 1] = _create_lookup_table(
            self.N, self._segmentdata['green'], self._gamma)
        self._lut[:-3, 2] = _create_lookup_table(
            self.N, self._segmentdata['blue'], self._gamma)
        if 'alpha' in self._segmentdata:
            self._lut[:-3, 3] = _create_lookup_table(
                self.N, self._segmentdata['alpha'], 1)
        self._isinit = True
        self._set_extremes()

    def set_gamma(self, gamma):
        """Set a new gamma value and regenerate colormap."""
        self._gamma = gamma
        self._init()

    @staticmethod
    def from_list(name, colors, N=256, gamma=1.0):
        """
        Create a `LinearSegmentedColormap` from a list of colors.

        Parameters
        ----------
        name : str
            The name of the colormap.
        colors : array-like of colors or array-like of (value, color)
            If only colors are given, they are equidistantly mapped from the
            range :math:`[0, 1]`; i.e. 0 maps to ``colors[0]`` and 1 maps to
            ``colors[-1]``.
            If (value, color) pairs are given, the mapping is from *value*
            to *color*. This can be used to divide the range unevenly.
        N : int
            The number of RGB quantization levels.
        gamma : float
        """
        if not np.iterable(colors):
            raise ValueError('colors must be iterable')

        if (isinstance(colors[0], Sized) and len(colors[0]) == 2
                and not isinstance(colors[0], str)):
            # List of value, color pairs
            vals, colors = zip(*colors)
        else:
            vals = np.linspace(0, 1, len(colors))

        r, g, b, a = to_rgba_array(colors).T
        cdict = {
            "red": np.column_stack([vals, r, r]),
            "green": np.column_stack([vals, g, g]),
            "blue": np.column_stack([vals, b, b]),
            "alpha": np.column_stack([vals, a, a]),
        }

        return LinearSegmentedColormap(name, cdict, N, gamma)

    def resampled(self, lutsize):
        """Return a new colormap with *lutsize* entries."""
        new_cmap = LinearSegmentedColormap(self.name, self._segmentdata,
                                           lutsize)
        new_cmap._rgba_over = self._rgba_over
        new_cmap._rgba_under = self._rgba_under
        new_cmap._rgba_bad = self._rgba_bad
        return new_cmap

    # Helper ensuring picklability of the reversed cmap.
    @staticmethod
    def _reverser(func, x):
        return func(1 - x)

    def reversed(self, name=None):
        """
        Return a reversed instance of the Colormap.

        Parameters
        ----------
        name : str, optional
            The name for the reversed colormap. If None, the
            name is set to ``self.name + "_r"``.

        Returns
        -------
        LinearSegmentedColormap
            The reversed colormap.
        """
        if name is None:
            name = self.name + "_r"

        # Using a partial object keeps the cmap picklable.
        data_r = {key: (functools.partial(self._reverser, data)
                        if callable(data) else
                        [(1.0 - x, y1, y0) for x, y0, y1 in reversed(data)])
                  for key, data in self._segmentdata.items()}

        new_cmap = LinearSegmentedColormap(name, data_r, self.N, self._gamma)
        # Reverse the over/under values too
        new_cmap._rgba_over = self._rgba_under
        new_cmap._rgba_under = self._rgba_over
        new_cmap._rgba_bad = self._rgba_bad
        return new_cmap


class ListedColormap(Colormap):
    """
    Colormap object generated from a list of colors.

    This may be most useful when indexing directly into a colormap,
    but it can also be used to generate special colormaps for ordinary
    mapping.

    Parameters
    ----------
    colors : list, array
        List of Matplotlib color specifications, or an equivalent Nx3 or Nx4
        floating point array (*N* RGB or RGBA values).
    name : str, optional
        String to identify the colormap.
    N : int, optional
        Number of entries in the map. The default is *None*, in which case
        there is one colormap entry for each element in the list of colors.
        If ::

            N < len(colors)

        the list will be truncated at *N*. If ::

            N > len(colors)

        the list will be extended by repetition.
    """
    def __init__(self, colors, name='from_list', N=None):
        self.monochrome = False  # Are all colors identical? (for contour.py)
        if N is None:
            self.colors = colors
            N = len(colors)
        else:
            if isinstance(colors, str):
                self.colors = [colors] * N
                self.monochrome = True
            elif np.iterable(colors):
                if len(colors) == 1:
                    self.monochrome = True
                self.colors = list(
                    itertools.islice(itertools.cycle(colors), N))
            else:
                try:
                    gray = float(colors)
                except TypeError:
                    pass
                else:
                    self.colors = [gray] * N
                self.monochrome = True
        super().__init__(name, N)

    def _init(self):
        self._lut = np.zeros((self.N + 3, 4), float)
        self._lut[:-3] = to_rgba_array(self.colors)
        self._isinit = True
        self._set_extremes()

    def resampled(self, lutsize):
        """Return a new colormap with *lutsize* entries."""
        colors = self(np.linspace(0, 1, lutsize))
        new_cmap = ListedColormap(colors, name=self.name)
        # Keep the over/under values too
        new_cmap._rgba_over = self._rgba_over
        new_cmap._rgba_under = self._rgba_under
        new_cmap._rgba_bad = self._rgba_bad
        return new_cmap

    def reversed(self, name=None):
        """
        Return a reversed instance of the Colormap.

        Parameters
        ----------
        name : str, optional
            The name for the reversed colormap. If None, the
            name is set to ``self.name + "_r"``.

        Returns
        -------
        ListedColormap
            A reversed instance of the colormap.
        """
        if name is None:
            name = self.name + "_r"

        colors_r = list(reversed(self.colors))
        new_cmap = ListedColormap(colors_r, name=name, N=self.N)
        # Reverse the over/under values too
        new_cmap._rgba_over = self._rgba_under
        new_cmap._rgba_under = self._rgba_over
        new_cmap._rgba_bad = self._rgba_bad
        return new_cmap


class Normalize:
    """
    A class which, when called, linearly normalizes data into the
    ``[0.0, 1.0]`` interval.
    """

    def __init__(self, vmin=None, vmax=None, clip=False):
        """
        Parameters
        ----------
        vmin, vmax : float or None
            If *vmin* and/or *vmax* is not given, they are initialized from the
            minimum and maximum value, respectively, of the first input
            processed; i.e., ``__call__(A)`` calls ``autoscale_None(A)``.

        clip : bool, default: False
            If ``True`` values falling outside the range ``[vmin, vmax]``,
            are mapped to 0 or 1, whichever is closer, and masked values are
            set to 1.  If ``False`` masked values remain masked.

            Clipping silently defeats the purpose of setting the over, under,
            and masked colors in a colormap, so it is likely to lead to
            surprises; therefore the default is ``clip=False``.

        Notes
        -----
        Returns 0 if ``vmin == vmax``.
        """
        self._vmin = _sanitize_extrema(vmin)
        self._vmax = _sanitize_extrema(vmax)
        self._clip = clip
        self._scale = None
        self.callbacks = cbook.CallbackRegistry(signals=["changed"])

    @property
    def vmin(self):
        return self._vmin

    @vmin.setter
    def vmin(self, value):
        value = _sanitize_extrema(value)
        if value != self._vmin:
            self._vmin = value
            self._changed()

    @property
    def vmax(self):
        return self._vmax

    @vmax.setter
    def vmax(self, value):
        value = _sanitize_extrema(value)
        if value != self._vmax:
            self._vmax = value
            self._changed()

    @property
    def clip(self):
        return self._clip

    @clip.setter
    def clip(self, value):
        if value != self._clip:
            self._clip = value
            self._changed()

    def _changed(self):
        """
        Call this whenever the norm is changed to notify all the
        callback listeners to the 'changed' signal.
        """
        self.callbacks.process('changed')

    @staticmethod
    def process_value(value):
        """
        Homogenize the input *value* for easy and efficient normalization.

        *value* can be a scalar or sequence.

        Returns
        -------
        result : masked array
            Masked array with the same shape as *value*.
        is_scalar : bool
            Whether *value* is a scalar.

        Notes
        -----
        Float dtypes are preserved; integer types with two bytes or smaller are
        converted to np.float32, and larger types are converted to np.float64.
        Preserving float32 when possible, and using in-place operations,
        greatly improves speed for large arrays.
        """
        is_scalar = not np.iterable(value)
        if is_scalar:
            value = [value]
        dtype = np.min_scalar_type(value)
        if np.issubdtype(dtype, np.integer) or dtype.type is np.bool_:
            # bool_/int8/int16 -> float32; int32/int64 -> float64
            dtype = np.promote_types(dtype, np.float32)
        # ensure data passed in as an ndarray subclass are interpreted as
        # an ndarray. See issue #6622.
        mask = np.ma.getmask(value)
        data = np.asarray(value)
        result = np.ma.array(data, mask=mask, dtype=dtype, copy=True)
        return result, is_scalar

    def __call__(self, value, clip=None):
        """
        Normalize *value* data in the ``[vmin, vmax]`` interval into the
        ``[0.0, 1.0]`` interval and return it.

        Parameters
        ----------
        value
            Data to normalize.
        clip : bool, optional
            If ``None``, defaults to ``self.clip`` (which defaults to
            ``False``).

        Notes
        -----
        If not already initialized, ``self.vmin`` and ``self.vmax`` are
        initialized using ``self.autoscale_None(value)``.
        """
        if clip is None:
            clip = self.clip

        result, is_scalar = self.process_value(value)

        if self.vmin is None or self.vmax is None:
            self.autoscale_None(result)
        # Convert at least to float, without losing precision.
        (vmin,), _ = self.process_value(self.vmin)
        (vmax,), _ = self.process_value(self.vmax)
        if vmin == vmax:
            result.fill(0)  # Or should it be all masked?  Or 0.5?
        elif vmin > vmax:
            raise ValueError("minvalue must be less than or equal to maxvalue")
        else:
            if clip:
                mask = np.ma.getmask(result)
                result = np.ma.array(np.clip(result.filled(vmax), vmin, vmax),
                                     mask=mask)
            # ma division is very slow; we can take a shortcut
            resdat = result.data
            resdat -= vmin
            resdat /= (vmax - vmin)
            result = np.ma.array(resdat, mask=result.mask, copy=False)
        if is_scalar:
            result = result[0]
        return result

    def inverse(self, value):
        if not self.scaled():
            raise ValueError("Not invertible until both vmin and vmax are set")
        (vmin,), _ = self.process_value(self.vmin)
        (vmax,), _ = self.process_value(self.vmax)

        if np.iterable(value):
            val = np.ma.asarray(value)
            return vmin + val * (vmax - vmin)
        else:
            return vmin + value * (vmax - vmin)

    def autoscale(self, A):
        """Set *vmin*, *vmax* to min, max of *A*."""
        with self.callbacks.blocked():
            # Pause callbacks while we are updating so we only get
            # a single update signal at the end
            self.vmin = self.vmax = None
            self.autoscale_None(A)
        self._changed()

    def autoscale_None(self, A):
        """If vmin or vmax are not set, use the min/max of *A* to set them."""
        A = np.asanyarray(A)
        if self.vmin is None and A.size:
            self.vmin = A.min()
        if self.vmax is None and A.size:
            self.vmax = A.max()

    def scaled(self):
        """Return whether vmin and vmax are set."""
        return self.vmin is not None and self.vmax is not None


class TwoSlopeNorm(Normalize):
    def __init__(self, vcenter, vmin=None, vmax=None):
        """
        Normalize data with a set center.

        Useful when mapping data with an unequal rates of change around a
        conceptual center, e.g., data that range from -2 to 4, with 0 as
        the midpoint.

        Parameters
        ----------
        vcenter : float
            The data value that defines ``0.5`` in the normalization.
        vmin : float, optional
            The data value that defines ``0.0`` in the normalization.
            Defaults to the min value of the dataset.
        vmax : float, optional
            The data value that defines ``1.0`` in the normalization.
            Defaults to the max value of the dataset.

        Examples
        --------
        This maps data value -4000 to 0., 0 to 0.5, and +10000 to 1.0; data
        between is linearly interpolated::

            >>> import matplotlib.colors as mcolors
            >>> offset = mcolors.TwoSlopeNorm(vmin=-4000.,
                                              vcenter=0., vmax=10000)
            >>> data = [-4000., -2000., 0., 2500., 5000., 7500., 10000.]
            >>> offset(data)
            array([0., 0.25, 0.5, 0.625, 0.75, 0.875, 1.0])
        """

        super().__init__(vmin=vmin, vmax=vmax)
        self._vcenter = vcenter
        if vcenter is not None and vmax is not None and vcenter >= vmax:
            raise ValueError('vmin, vcenter, and vmax must be in '
                             'ascending order')
        if vcenter is not None and vmin is not None and vcenter <= vmin:
            raise ValueError('vmin, vcenter, and vmax must be in '
                             'ascending order')

    @property
    def vcenter(self):
        return self._vcenter

    @vcenter.setter
    def vcenter(self, value):
        if value != self._vcenter:
            self._vcenter = value
            self._changed()

    def autoscale_None(self, A):
        """
        Get vmin and vmax, and then clip at vcenter
        """
        super().autoscale_None(A)
        if self.vmin > self.vcenter:
            self.vmin = self.vcenter
        if self.vmax < self.vcenter:
            self.vmax = self.vcenter

    def __call__(self, value, clip=None):
        """
        Map value to the interval [0, 1]. The *clip* argument is unused.
        """
        result, is_scalar = self.process_value(value)
        self.autoscale_None(result)  # sets self.vmin, self.vmax if None

        if not self.vmin <= self.vcenter <= self.vmax:
            raise ValueError("vmin, vcenter, vmax must increase monotonically")
        # note that we must extrapolate for tick locators:
        result = np.ma.masked_array(
            np.interp(result, [self.vmin, self.vcenter, self.vmax],
                      [0, 0.5, 1], left=-np.inf, right=np.inf),
            mask=np.ma.getmask(result))
        if is_scalar:
            result = np.atleast_1d(result)[0]
        return result

    def inverse(self, value):
        if not self.scaled():
            raise ValueError("Not invertible until both vmin and vmax are set")
        (vmin,), _ = self.process_value(self.vmin)
        (vmax,), _ = self.process_value(self.vmax)
        (vcenter,), _ = self.process_value(self.vcenter)
        result = np.interp(value, [0, 0.5, 1], [vmin, vcenter, vmax],
                           left=-np.inf, right=np.inf)
        return result


class CenteredNorm(Normalize):
    def __init__(self, vcenter=0, halfrange=None, clip=False):
        """
        Normalize symmetrical data around a center (0 by default).

        Unlike `TwoSlopeNorm`, `CenteredNorm` applies an equal rate of change
        around the center.

        Useful when mapping symmetrical data around a conceptual center
        e.g., data that range from -2 to 4, with 0 as the midpoint, and
        with equal rates of change around that midpoint.

        Parameters
        ----------
        vcenter : float, default: 0
            The data value that defines ``0.5`` in the normalization.
        halfrange : float, optional
            The range of data values that defines a range of ``0.5`` in the
            normalization, so that *vcenter* - *halfrange* is ``0.0`` and
            *vcenter* + *halfrange* is ``1.0`` in the normalization.
            Defaults to the largest absolute difference to *vcenter* for
            the values in the dataset.
        clip : bool, default: False
            If ``True`` values falling outside the range ``[vmin, vmax]``,
            are mapped to 0 or 1, whichever is closer, and masked values are
            set to 1.  If ``False`` masked values remain masked.

        Examples
        --------
        This maps data values -2 to 0.25, 0 to 0.5, and 4 to 1.0
        (assuming equal rates of change above and below 0.0):

            >>> import matplotlib.colors as mcolors
            >>> norm = mcolors.CenteredNorm(halfrange=4.0)
            >>> data = [-2., 0., 4.]
            >>> norm(data)
            array([0.25, 0.5 , 1.  ])
        """
        super().__init__(vmin=None, vmax=None, clip=clip)
        self._vcenter = vcenter
        # calling the halfrange setter to set vmin and vmax
        self.halfrange = halfrange

    def autoscale(self, A):
        """
        Set *halfrange* to ``max(abs(A-vcenter))``, then set *vmin* and *vmax*.
        """
        A = np.asanyarray(A)
        self.halfrange = max(self._vcenter-A.min(),
                             A.max()-self._vcenter)

    def autoscale_None(self, A):
        """Set *vmin* and *vmax*."""
        A = np.asanyarray(A)
        if self.halfrange is None and A.size:
            self.autoscale(A)

    @property
    def vmin(self):
        return self._vmin

    @vmin.setter
    def vmin(self, value):
        value = _sanitize_extrema(value)
        if value != self._vmin:
            self._vmin = value
            self._vmax = 2*self.vcenter - value
            self._changed()

    @property
    def vmax(self):
        return self._vmax

    @vmax.setter
    def vmax(self, value):
        value = _sanitize_extrema(value)
        if value != self._vmax:
            self._vmax = value
            self._vmin = 2*self.vcenter - value
            self._changed()

    @property
    def vcenter(self):
        return self._vcenter

    @vcenter.setter
    def vcenter(self, vcenter):
        if vcenter != self._vcenter:
            self._vcenter = vcenter
            # Trigger an update of the vmin/vmax values through the setter
            self.halfrange = self.halfrange
            self._changed()

    @property
    def halfrange(self):
        if self.vmin is None or self.vmax is None:
            return None
        return (self.vmax - self.vmin) / 2

    @halfrange.setter
    def halfrange(self, halfrange):
        if halfrange is None:
            self.vmin = None
            self.vmax = None
        else:
            self.vmin = self.vcenter - abs(halfrange)
            self.vmax = self.vcenter + abs(halfrange)


def make_norm_from_scale(scale_cls, base_norm_cls=None, *, init=None):
    """
    Decorator for building a `.Normalize` subclass from a `~.scale.ScaleBase`
    subclass.

    After ::

        @make_norm_from_scale(scale_cls)
        class norm_cls(Normalize):
            ...

    *norm_cls* is filled with methods so that normalization computations are
    forwarded to *scale_cls* (i.e., *scale_cls* is the scale that would be used
    for the colorbar of a mappable normalized with *norm_cls*).

    If *init* is not passed, then the constructor signature of *norm_cls*
    will be ``norm_cls(vmin=None, vmax=None, clip=False)``; these three
    parameters will be forwarded to the base class (``Normalize.__init__``),
    and a *scale_cls* object will be initialized with no arguments (other than
    a dummy axis).

    If the *scale_cls* constructor takes additional parameters, then *init*
    should be passed to `make_norm_from_scale`.  It is a callable which is
    *only* used for its signature.  First, this signature will become the
    signature of *norm_cls*.  Second, the *norm_cls* constructor will bind the
    parameters passed to it using this signature, extract the bound *vmin*,
    *vmax*, and *clip* values, pass those to ``Normalize.__init__``, and
    forward the remaining bound values (including any defaults defined by the
    signature) to the *scale_cls* constructor.
    """

    if base_norm_cls is None:
        return functools.partial(make_norm_from_scale, scale_cls, init=init)

    if isinstance(scale_cls, functools.partial):
        scale_args = scale_cls.args
        scale_kwargs_items = tuple(scale_cls.keywords.items())
        scale_cls = scale_cls.func
    else:
        scale_args = scale_kwargs_items = ()

    if init is None:
        def init(vmin=None, vmax=None, clip=False): pass

    return _make_norm_from_scale(
        scale_cls, scale_args, scale_kwargs_items,
        base_norm_cls, inspect.signature(init))


@functools.lru_cache(None)
def _make_norm_from_scale(
    scale_cls, scale_args, scale_kwargs_items,
    base_norm_cls, bound_init_signature,
):
    """
    Helper for `make_norm_from_scale`.

    This function is split out to enable caching (in particular so that
    different unpickles reuse the same class).  In order to do so,

    - ``functools.partial`` *scale_cls* is expanded into ``func, args, kwargs``
      to allow memoizing returned norms (partial instances always compare
      unequal, but we can check identity based on ``func, args, kwargs``;
    - *init* is replaced by *init_signature*, as signatures are picklable,
      unlike to arbitrary lambdas.
    """

    class Norm(base_norm_cls):
        def __reduce__(self):
            cls = type(self)
            # If the class is toplevel-accessible, it is possible to directly
            # pickle it "by name".  This is required to support norm classes
            # defined at a module's toplevel, as the inner base_norm_cls is
            # otherwise unpicklable (as it gets shadowed by the generated norm
            # class).  If either import or attribute access fails, fall back to
            # the general path.
            try:
                if cls is getattr(importlib.import_module(cls.__module__),
                                  cls.__qualname__):
                    return (_create_empty_object_of_class, (cls,), vars(self))
            except (ImportError, AttributeError):
                pass
            return (_picklable_norm_constructor,
                    (scale_cls, scale_args, scale_kwargs_items,
                     base_norm_cls, bound_init_signature),
                    vars(self))

        def __init__(self, *args, **kwargs):
            ba = bound_init_signature.bind(*args, **kwargs)
            ba.apply_defaults()
            super().__init__(
                **{k: ba.arguments.pop(k) for k in ["vmin", "vmax", "clip"]})
            self._scale = functools.partial(
                scale_cls, *scale_args, **dict(scale_kwargs_items))(
                    axis=None, **ba.arguments)
            self._trf = self._scale.get_transform()

        __init__.__signature__ = bound_init_signature.replace(parameters=[
            inspect.Parameter("self", inspect.Parameter.POSITIONAL_OR_KEYWORD),
            *bound_init_signature.parameters.values()])

        def __call__(self, value, clip=None):
            value, is_scalar = self.process_value(value)
            if self.vmin is None or self.vmax is None:
                self.autoscale_None(value)
            if self.vmin > self.vmax:
                raise ValueError("vmin must be less or equal to vmax")
            if self.vmin == self.vmax:
                return np.full_like(value, 0)
            if clip is None:
                clip = self.clip
            if clip:
                value = np.clip(value, self.vmin, self.vmax)
            t_value = self._trf.transform(value).reshape(np.shape(value))
            t_vmin, t_vmax = self._trf.transform([self.vmin, self.vmax])
            if not np.isfinite([t_vmin, t_vmax]).all():
                raise ValueError("Invalid vmin or vmax")
            t_value -= t_vmin
            t_value /= (t_vmax - t_vmin)
            t_value = np.ma.masked_invalid(t_value, copy=False)
            return t_value[0] if is_scalar else t_value

        def inverse(self, value):
            if not self.scaled():
                raise ValueError("Not invertible until scaled")
            if self.vmin > self.vmax:
                raise ValueError("vmin must be less or equal to vmax")
            t_vmin, t_vmax = self._trf.transform([self.vmin, self.vmax])
            if not np.isfinite([t_vmin, t_vmax]).all():
                raise ValueError("Invalid vmin or vmax")
            value, is_scalar = self.process_value(value)
            rescaled = value * (t_vmax - t_vmin)
            rescaled += t_vmin
            value = (self._trf
                     .inverted()
                     .transform(rescaled)
                     .reshape(np.shape(value)))
            return value[0] if is_scalar else value

        def autoscale_None(self, A):
            # i.e. A[np.isfinite(...)], but also for non-array A's
            in_trf_domain = np.extract(np.isfinite(self._trf.transform(A)), A)
            if in_trf_domain.size == 0:
                in_trf_domain = np.ma.masked
            return super().autoscale_None(in_trf_domain)

    if base_norm_cls is Normalize:
        Norm.__name__ = f"{scale_cls.__name__}Norm"
        Norm.__qualname__ = f"{scale_cls.__qualname__}Norm"
    else:
        Norm.__name__ = base_norm_cls.__name__
        Norm.__qualname__ = base_norm_cls.__qualname__
    Norm.__module__ = base_norm_cls.__module__
    Norm.__doc__ = base_norm_cls.__doc__

    return Norm


def _create_empty_object_of_class(cls):
    return cls.__new__(cls)


def _picklable_norm_constructor(*args):
    return _create_empty_object_of_class(_make_norm_from_scale(*args))


@make_norm_from_scale(
    scale.FuncScale,
    init=lambda functions, vmin=None, vmax=None, clip=False: None)
class FuncNorm(Normalize):
    """
    Arbitrary normalization using functions for the forward and inverse.

    Parameters
    ----------
    functions : (callable, callable)
        two-tuple of the forward and inverse functions for the normalization.
        The forward function must be monotonic.

        Both functions must have the signature ::

           def forward(values: array-like) -> array-like

    vmin, vmax : float or None
        If *vmin* and/or *vmax* is not given, they are initialized from the
        minimum and maximum value, respectively, of the first input
        processed; i.e., ``__call__(A)`` calls ``autoscale_None(A)``.

    clip : bool, default: False
        If ``True`` values falling outside the range ``[vmin, vmax]``,
        are mapped to 0 or 1, whichever is closer, and masked values are
        set to 1.  If ``False`` masked values remain masked.

        Clipping silently defeats the purpose of setting the over, under,
        and masked colors in a colormap, so it is likely to lead to
        surprises; therefore the default is ``clip=False``.
    """


LogNorm = make_norm_from_scale(
    functools.partial(scale.LogScale, nonpositive="mask"))(Normalize)
LogNorm.__name__ = LogNorm.__qualname__ = "LogNorm"
LogNorm.__doc__ = "Normalize a given value to the 0-1 range on a log scale."


@make_norm_from_scale(
    scale.SymmetricalLogScale,
    init=lambda linthresh, linscale=1., vmin=None, vmax=None, clip=False, *,
                base=10: None)
class SymLogNorm(Normalize):
    """
    The symmetrical logarithmic scale is logarithmic in both the
    positive and negative directions from the origin.

    Since the values close to zero tend toward infinity, there is a
    need to have a range around zero that is linear.  The parameter
    *linthresh* allows the user to specify the size of this range
    (-*linthresh*, *linthresh*).

    Parameters
    ----------
    linthresh : float
        The range within which the plot is linear (to avoid having the plot
        go to infinity around zero).
    linscale : float, default: 1
        This allows the linear range (-*linthresh* to *linthresh*) to be
        stretched relative to the logarithmic range. Its value is the
        number of decades to use for each half of the linear range. For
        example, when *linscale* == 1.0 (the default), the space used for
        the positive and negative halves of the linear range will be equal
        to one decade in the logarithmic range.
    base : float, default: 10
    """

    @property
    def linthresh(self):
        return self._scale.linthresh

    @linthresh.setter
    def linthresh(self, value):
        self._scale.linthresh = value


@make_norm_from_scale(
    scale.AsinhScale,
    init=lambda linear_width=1, vmin=None, vmax=None, clip=False: None)
class AsinhNorm(Normalize):
    """
    The inverse hyperbolic sine scale is approximately linear near
    the origin, but becomes logarithmic for larger positive
    or negative values. Unlike the `SymLogNorm`, the transition between
    these linear and logarithmic regions is smooth, which may reduce
    the risk of visual artifacts.

    .. note::

       This API is provisional and may be revised in the future
       based on early user feedback.

    Parameters
    ----------
    linear_width : float, default: 1
        The effective width of the linear region, beyond which
        the transformation becomes asymptotically logarithmic
    """

    @property
    def linear_width(self):
        return self._scale.linear_width

    @linear_width.setter
    def linear_width(self, value):
        self._scale.linear_width = value


class PowerNorm(Normalize):
    """
    Linearly map a given value to the 0-1 range and then apply
    a power-law normalization over that range.
    """
    def __init__(self, gamma, vmin=None, vmax=None, clip=False):
        super().__init__(vmin, vmax, clip)
        self.gamma = gamma

    def __call__(self, value, clip=None):
        if clip is None:
            clip = self.clip

        result, is_scalar = self.process_value(value)

        self.autoscale_None(result)
        gamma = self.gamma
        vmin, vmax = self.vmin, self.vmax
        if vmin > vmax:
            raise ValueError("minvalue must be less than or equal to maxvalue")
        elif vmin == vmax:
            result.fill(0)
        else:
            if clip:
                mask = np.ma.getmask(result)
                result = np.ma.array(np.clip(result.filled(vmax), vmin, vmax),
                                     mask=mask)
            resdat = result.data
            resdat -= vmin
            resdat[resdat < 0] = 0
            np.power(resdat, gamma, resdat)
            resdat /= (vmax - vmin) ** gamma

            result = np.ma.array(resdat, mask=result.mask, copy=False)
        if is_scalar:
            result = result[0]
        return result

    def inverse(self, value):
        if not self.scaled():
            raise ValueError("Not invertible until scaled")
        gamma = self.gamma
        vmin, vmax = self.vmin, self.vmax

        if np.iterable(value):
            val = np.ma.asarray(value)
            return np.ma.power(val, 1. / gamma) * (vmax - vmin) + vmin
        else:
            return pow(value, 1. / gamma) * (vmax - vmin) + vmin


class BoundaryNorm(Normalize):
    """
    Generate a colormap index based on discrete intervals.

    Unlike `Normalize` or `LogNorm`, `BoundaryNorm` maps values to integers
    instead of to the interval 0-1.
    """

    # Mapping to the 0-1 interval could have been done via piece-wise linear
    # interpolation, but using integers seems simpler, and reduces the number
    # of conversions back and forth between int and float.

    def __init__(self, boundaries, ncolors, clip=False, *, extend='neither'):
        """
        Parameters
        ----------
        boundaries : array-like
            Monotonically increasing sequence of at least 2 bin edges:  data
            falling in the n-th bin will be mapped to the n-th color.

        ncolors : int
            Number of colors in the colormap to be used.

        clip : bool, optional
            If clip is ``True``, out of range values are mapped to 0 if they
            are below ``boundaries[0]`` or mapped to ``ncolors - 1`` if they
            are above ``boundaries[-1]``.

            If clip is ``False``, out of range values are mapped to -1 if
            they are below ``boundaries[0]`` or mapped to *ncolors* if they are
            above ``boundaries[-1]``. These are then converted to valid indices
            by `Colormap.__call__`.

        extend : {'neither', 'both', 'min', 'max'}, default: 'neither'
            Extend the number of bins to include one or both of the
            regions beyond the boundaries.  For example, if ``extend``
            is 'min', then the color to which the region between the first
            pair of boundaries is mapped will be distinct from the first
            color in the colormap, and by default a
            `~matplotlib.colorbar.Colorbar` will be drawn with
            the triangle extension on the left or lower end.

        Notes
        -----
        If there are fewer bins (including extensions) than colors, then the
        color index is chosen by linearly interpolating the ``[0, nbins - 1]``
        range onto the ``[0, ncolors - 1]`` range, effectively skipping some
        colors in the middle of the colormap.
        """
        if clip and extend != 'neither':
            raise ValueError("'clip=True' is not compatible with 'extend'")
        super().__init__(vmin=boundaries[0], vmax=boundaries[-1], clip=clip)
        self.boundaries = np.asarray(boundaries)
        self.N = len(self.boundaries)
        if self.N < 2:
            raise ValueError("You must provide at least 2 boundaries "
                             f"(1 region) but you passed in {boundaries!r}")
        self.Ncmap = ncolors
        self.extend = extend

        self._scale = None  # don't use the default scale.

        self._n_regions = self.N - 1  # number of colors needed
        self._offset = 0
        if extend in ('min', 'both'):
            self._n_regions += 1
            self._offset = 1
        if extend in ('max', 'both'):
            self._n_regions += 1
        if self._n_regions > self.Ncmap:
            raise ValueError(f"There are {self._n_regions} color bins "
                             "including extensions, but ncolors = "
                             f"{ncolors}; ncolors must equal or exceed the "
                             "number of bins")

    def __call__(self, value, clip=None):
        """
        This method behaves similarly to `.Normalize.__call__`, except that it
        returns integers or arrays of int16.
        """
        if clip is None:
            clip = self.clip

        xx, is_scalar = self.process_value(value)
        mask = np.ma.getmaskarray(xx)
        # Fill masked values a value above the upper boundary
        xx = np.atleast_1d(xx.filled(self.vmax + 1))
        if clip:
            np.clip(xx, self.vmin, self.vmax, out=xx)
            max_col = self.Ncmap - 1
        else:
            max_col = self.Ncmap
        # this gives us the bins in the lookup table in the range
        # [0, _n_regions - 1]  (the offset is set in the init)
        iret = np.digitize(xx, self.boundaries) - 1 + self._offset
        # if we have more colors than regions, stretch the region
        # index computed above to full range of the color bins.  This
        # will make use of the full range (but skip some of the colors
        # in the middle) such that the first region is mapped to the
        # first color and the last region is mapped to the last color.
        if self.Ncmap > self._n_regions:
            if self._n_regions == 1:
                # special case the 1 region case, pick the middle color
                iret[iret == 0] = (self.Ncmap - 1) // 2
            else:
                # otherwise linearly remap the values from the region index
                # to the color index spaces
                iret = (self.Ncmap - 1) / (self._n_regions - 1) * iret
        # cast to 16bit integers in all cases
        iret = iret.astype(np.int16)
        iret[xx < self.vmin] = -1
        iret[xx >= self.vmax] = max_col
        ret = np.ma.array(iret, mask=mask)
        if is_scalar:
            ret = int(ret[0])  # assume python scalar
        return ret

    def inverse(self, value):
        """
        Raises
        ------
        ValueError
            BoundaryNorm is not invertible, so calling this method will always
            raise an error
        """
        raise ValueError("BoundaryNorm is not invertible")


class NoNorm(Normalize):
    """
    Dummy replacement for `Normalize`, for the case where we want to use
    indices directly in a `~matplotlib.cm.ScalarMappable`.
    """
    def __call__(self, value, clip=None):
        return value

    def inverse(self, value):
        return value


def rgb_to_hsv(arr):
    """
    Convert float RGB values (in the range [0, 1]), in a numpy array to HSV
    values.

    Parameters
    ----------
    arr : (..., 3) array-like
       All values must be in the range [0, 1]

    Returns
    -------
    (..., 3) `~numpy.ndarray`
       Colors converted to HSV values in range [0, 1]
    """
    arr = np.asarray(arr)

    # check length of the last dimension, should be _some_ sort of rgb
    if arr.shape[-1] != 3:
        raise ValueError("Last dimension of input array must be 3; "
                         "shape {} was found.".format(arr.shape))

    in_shape = arr.shape
    arr = np.array(
        arr, copy=False,
        dtype=np.promote_types(arr.dtype, np.float32),  # Don't work on ints.
        ndmin=2,  # In case input was 1D.
    )
    out = np.zeros_like(arr)
    arr_max = arr.max(-1)
    ipos = arr_max > 0
    delta = arr.ptp(-1)
    s = np.zeros_like(delta)
    s[ipos] = delta[ipos] / arr_max[ipos]
    ipos = delta > 0
    # red is max
    idx = (arr[..., 0] == arr_max) & ipos
    out[idx, 0] = (arr[idx, 1] - arr[idx, 2]) / delta[idx]
    # green is max
    idx = (arr[..., 1] == arr_max) & ipos
    out[idx, 0] = 2. + (arr[idx, 2] - arr[idx, 0]) / delta[idx]
    # blue is max
    idx = (arr[..., 2] == arr_max) & ipos
    out[idx, 0] = 4. + (arr[idx, 0] - arr[idx, 1]) / delta[idx]

    out[..., 0] = (out[..., 0] / 6.0) % 1.0
    out[..., 1] = s
    out[..., 2] = arr_max

    return out.reshape(in_shape)


def hsv_to_rgb(hsv):
    """
    Convert HSV values to RGB.

    Parameters
    ----------
    hsv : (..., 3) array-like
       All values assumed to be in range [0, 1]

    Returns
    -------
    (..., 3) `~numpy.ndarray`
       Colors converted to RGB values in range [0, 1]
    """
    hsv = np.asarray(hsv)

    # check length of the last dimension, should be _some_ sort of rgb
    if hsv.shape[-1] != 3:
        raise ValueError("Last dimension of input array must be 3; "
                         "shape {shp} was found.".format(shp=hsv.shape))

    in_shape = hsv.shape
    hsv = np.array(
        hsv, copy=False,
        dtype=np.promote_types(hsv.dtype, np.float32),  # Don't work on ints.
        ndmin=2,  # In case input was 1D.
    )

    h = hsv[..., 0]
    s = hsv[..., 1]
    v = hsv[..., 2]

    r = np.empty_like(h)
    g = np.empty_like(h)
    b = np.empty_like(h)

    i = (h * 6.0).astype(int)
    f = (h * 6.0) - i
    p = v * (1.0 - s)
    q = v * (1.0 - s * f)
    t = v * (1.0 - s * (1.0 - f))

    idx = i % 6 == 0
    r[idx] = v[idx]
    g[idx] = t[idx]
    b[idx] = p[idx]

    idx = i == 1
    r[idx] = q[idx]
    g[idx] = v[idx]
    b[idx] = p[idx]

    idx = i == 2
    r[idx] = p[idx]
    g[idx] = v[idx]
    b[idx] = t[idx]

    idx = i == 3
    r[idx] = p[idx]
    g[idx] = q[idx]
    b[idx] = v[idx]

    idx = i == 4
    r[idx] = t[idx]
    g[idx] = p[idx]
    b[idx] = v[idx]

    idx = i == 5
    r[idx] = v[idx]
    g[idx] = p[idx]
    b[idx] = q[idx]

    idx = s == 0
    r[idx] = v[idx]
    g[idx] = v[idx]
    b[idx] = v[idx]

    rgb = np.stack([r, g, b], axis=-1)

    return rgb.reshape(in_shape)


def _vector_magnitude(arr):
    # things that don't work here:
    #  * np.linalg.norm: drops mask from ma.array
    #  * np.sum: drops mask from ma.array unless entire vector is masked
    sum_sq = 0
    for i in range(arr.shape[-1]):
        sum_sq += arr[..., i, np.newaxis] ** 2
    return np.sqrt(sum_sq)


class LightSource:
    """
    Create a light source coming from the specified azimuth and elevation.
    Angles are in degrees, with the azimuth measured
    clockwise from north and elevation up from the zero plane of the surface.

    `shade` is used to produce "shaded" RGB values for a data array.
    `shade_rgb` can be used to combine an RGB image with an elevation map.
    `hillshade` produces an illumination map of a surface.
    """

    def __init__(self, azdeg=315, altdeg=45, hsv_min_val=0, hsv_max_val=1,
                 hsv_min_sat=1, hsv_max_sat=0):
        """
        Specify the azimuth (measured clockwise from south) and altitude
        (measured up from the plane of the surface) of the light source
        in degrees.

        Parameters
        ----------
        azdeg : float, default: 315 degrees (from the northwest)
            The azimuth (0-360, degrees clockwise from North) of the light
            source.
        altdeg : float, default: 45 degrees
            The altitude (0-90, degrees up from horizontal) of the light
            source.

        Notes
        -----
        For backwards compatibility, the parameters *hsv_min_val*,
        *hsv_max_val*, *hsv_min_sat*, and *hsv_max_sat* may be supplied at
        initialization as well.  However, these parameters will only be used if
        "blend_mode='hsv'" is passed into `shade` or `shade_rgb`.
        See the documentation for `blend_hsv` for more details.
        """
        self.azdeg = azdeg
        self.altdeg = altdeg
        self.hsv_min_val = hsv_min_val
        self.hsv_max_val = hsv_max_val
        self.hsv_min_sat = hsv_min_sat
        self.hsv_max_sat = hsv_max_sat

    @property
    def direction(self):
        """The unit vector direction towards the light source."""
        # Azimuth is in degrees clockwise from North. Convert to radians
        # counterclockwise from East (mathematical notation).
        az = np.radians(90 - self.azdeg)
        alt = np.radians(self.altdeg)
        return np.array([
            np.cos(az) * np.cos(alt),
            np.sin(az) * np.cos(alt),
            np.sin(alt)
        ])

    def hillshade(self, elevation, vert_exag=1, dx=1, dy=1, fraction=1.):
        """
        Calculate the illumination intensity for a surface using the defined
        azimuth and elevation for the light source.

        This computes the normal vectors for the surface, and then passes them
        on to `shade_normals`

        Parameters
        ----------
        elevation : 2D array-like
            The height values used to generate an illumination map
        vert_exag : number, optional
            The amount to exaggerate the elevation values by when calculating
            illumination. This can be used either to correct for differences in
            units between the x-y coordinate system and the elevation
            coordinate system (e.g. decimal degrees vs. meters) or to
            exaggerate or de-emphasize topographic effects.
        dx : number, optional
            The x-spacing (columns) of the input *elevation* grid.
        dy : number, optional
            The y-spacing (rows) of the input *elevation* grid.
        fraction : number, optional
            Increases or decreases the contrast of the hillshade.  Values
            greater than one will cause intermediate values to move closer to
            full illumination or shadow (and clipping any values that move
            beyond 0 or 1). Note that this is not visually or mathematically
            the same as vertical exaggeration.

        Returns
        -------
        `~numpy.ndarray`
            A 2D array of illumination values between 0-1, where 0 is
            completely in shadow and 1 is completely illuminated.
        """

        # Because most image and raster GIS data has the first row in the array
        # as the "top" of the image, dy is implicitly negative.  This is
        # consistent to what `imshow` assumes, as well.
        dy = -dy

        # compute the normal vectors from the partial derivatives
        e_dy, e_dx = np.gradient(vert_exag * elevation, dy, dx)

        # .view is to keep subclasses
        normal = np.empty(elevation.shape + (3,)).view(type(elevation))
        normal[..., 0] = -e_dx
        normal[..., 1] = -e_dy
        normal[..., 2] = 1
        normal /= _vector_magnitude(normal)

        return self.shade_normals(normal, fraction)

    def shade_normals(self, normals, fraction=1.):
        """
        Calculate the illumination intensity for the normal vectors of a
        surface using the defined azimuth and elevation for the light source.

        Imagine an artificial sun placed at infinity in some azimuth and
        elevation position illuminating our surface. The parts of the surface
        that slope toward the sun should brighten while those sides facing away
        should become darker.

        Parameters
        ----------
        fraction : number, optional
            Increases or decreases the contrast of the hillshade.  Values
            greater than one will cause intermediate values to move closer to
            full illumination or shadow (and clipping any values that move
            beyond 0 or 1). Note that this is not visually or mathematically
            the same as vertical exaggeration.

        Returns
        -------
        `~numpy.ndarray`
            A 2D array of illumination values between 0-1, where 0 is
            completely in shadow and 1 is completely illuminated.
        """

        intensity = normals.dot(self.direction)

        # Apply contrast stretch
        imin, imax = intensity.min(), intensity.max()
        intensity *= fraction

        # Rescale to 0-1, keeping range before contrast stretch
        # If constant slope, keep relative scaling (i.e. flat should be 0.5,
        # fully occluded 0, etc.)
        if (imax - imin) > 1e-6:
            # Strictly speaking, this is incorrect. Negative values should be
            # clipped to 0 because they're fully occluded. However, rescaling
            # in this manner is consistent with the previous implementation and
            # visually appears better than a "hard" clip.
            intensity -= imin
            intensity /= (imax - imin)
        intensity = np.clip(intensity, 0, 1)

        return intensity

    def shade(self, data, cmap, norm=None, blend_mode='overlay', vmin=None,
              vmax=None, vert_exag=1, dx=1, dy=1, fraction=1, **kwargs):
        """
        Combine colormapped data values with an illumination intensity map
        (a.k.a.  "hillshade") of the values.

        Parameters
        ----------
        data : 2D array-like
            The height values used to generate a shaded map.
        cmap : `~matplotlib.colors.Colormap`
            The colormap used to color the *data* array. Note that this must be
            a `~matplotlib.colors.Colormap` instance.  For example, rather than
            passing in ``cmap='gist_earth'``, use
            ``cmap=plt.get_cmap('gist_earth')`` instead.
        norm : `~matplotlib.colors.Normalize` instance, optional
            The normalization used to scale values before colormapping. If
            None, the input will be linearly scaled between its min and max.
        blend_mode : {'hsv', 'overlay', 'soft'} or callable, optional
            The type of blending used to combine the colormapped data
            values with the illumination intensity.  Default is
            "overlay".  Note that for most topographic surfaces,
            "overlay" or "soft" appear more visually realistic. If a
            user-defined function is supplied, it is expected to
            combine an MxNx3 RGB array of floats (ranging 0 to 1) with
            an MxNx1 hillshade array (also 0 to 1).  (Call signature
            ``func(rgb, illum, **kwargs)``) Additional kwargs supplied
            to this function will be passed on to the *blend_mode*
            function.
        vmin : float or None, optional
            The minimum value used in colormapping *data*. If *None* the
            minimum value in *data* is used. If *norm* is specified, then this
            argument will be ignored.
        vmax : float or None, optional
            The maximum value used in colormapping *data*. If *None* the
            maximum value in *data* is used. If *norm* is specified, then this
            argument will be ignored.
        vert_exag : number, optional
            The amount to exaggerate the elevation values by when calculating
            illumination. This can be used either to correct for differences in
            units between the x-y coordinate system and the elevation
            coordinate system (e.g. decimal degrees vs. meters) or to
            exaggerate or de-emphasize topography.
        dx : number, optional
            The x-spacing (columns) of the input *elevation* grid.
        dy : number, optional
            The y-spacing (rows) of the input *elevation* grid.
        fraction : number, optional
            Increases or decreases the contrast of the hillshade.  Values
            greater than one will cause intermediate values to move closer to
            full illumination or shadow (and clipping any values that move
            beyond 0 or 1). Note that this is not visually or mathematically
            the same as vertical exaggeration.
        **kwargs
            Additional kwargs are passed on to the *blend_mode* function.

        Returns
        -------
        `~numpy.ndarray`
            An MxNx4 array of floats ranging between 0-1.
        """
        if vmin is None:
            vmin = data.min()
        if vmax is None:
            vmax = data.max()
        if norm is None:
            norm = Normalize(vmin=vmin, vmax=vmax)

        rgb0 = cmap(norm(data))
        rgb1 = self.shade_rgb(rgb0, elevation=data, blend_mode=blend_mode,
                              vert_exag=vert_exag, dx=dx, dy=dy,
                              fraction=fraction, **kwargs)
        # Don't overwrite the alpha channel, if present.
        rgb0[..., :3] = rgb1[..., :3]
        return rgb0

    def shade_rgb(self, rgb, elevation, fraction=1., blend_mode='hsv',
                  vert_exag=1, dx=1, dy=1, **kwargs):
        """
        Use this light source to adjust the colors of the *rgb* input array to
        give the impression of a shaded relief map with the given *elevation*.

        Parameters
        ----------
        rgb : array-like
            An (M, N, 3) RGB array, assumed to be in the range of 0 to 1.
        elevation : array-like
            An (M, N) array of the height values used to generate a shaded map.
        fraction : number
            Increases or decreases the contrast of the hillshade.  Values
            greater than one will cause intermediate values to move closer to
            full illumination or shadow (and clipping any values that move
            beyond 0 or 1). Note that this is not visually or mathematically
            the same as vertical exaggeration.
        blend_mode : {'hsv', 'overlay', 'soft'} or callable, optional
            The type of blending used to combine the colormapped data values
            with the illumination intensity.  For backwards compatibility, this
            defaults to "hsv". Note that for most topographic surfaces,
            "overlay" or "soft" appear more visually realistic. If a
            user-defined function is supplied, it is expected to combine an
            MxNx3 RGB array of floats (ranging 0 to 1) with an MxNx1 hillshade
            array (also 0 to 1).  (Call signature
            ``func(rgb, illum, **kwargs)``)
            Additional kwargs supplied to this function will be passed on to
            the *blend_mode* function.
        vert_exag : number, optional
            The amount to exaggerate the elevation values by when calculating
            illumination. This can be used either to correct for differences in
            units between the x-y coordinate system and the elevation
            coordinate system (e.g. decimal degrees vs. meters) or to
            exaggerate or de-emphasize topography.
        dx : number, optional
            The x-spacing (columns) of the input *elevation* grid.
        dy : number, optional
            The y-spacing (rows) of the input *elevation* grid.
        **kwargs
            Additional kwargs are passed on to the *blend_mode* function.

        Returns
        -------
        `~numpy.ndarray`
            An (m, n, 3) array of floats ranging between 0-1.
        """
        # Calculate the "hillshade" intensity.
        intensity = self.hillshade(elevation, vert_exag, dx, dy, fraction)
        intensity = intensity[..., np.newaxis]

        # Blend the hillshade and rgb data using the specified mode
        lookup = {
                'hsv': self.blend_hsv,
                'soft': self.blend_soft_light,
                'overlay': self.blend_overlay,
                }
        if blend_mode in lookup:
            blend = lookup[blend_mode](rgb, intensity, **kwargs)
        else:
            try:
                blend = blend_mode(rgb, intensity, **kwargs)
            except TypeError as err:
                raise ValueError('"blend_mode" must be callable or one of {}'
                                 .format(lookup.keys)) from err

        # Only apply result where hillshade intensity isn't masked
        if np.ma.is_masked(intensity):
            mask = intensity.mask[..., 0]
            for i in range(3):
                blend[..., i][mask] = rgb[..., i][mask]

        return blend

    def blend_hsv(self, rgb, intensity, hsv_max_sat=None, hsv_max_val=None,
                  hsv_min_val=None, hsv_min_sat=None):
        """
        Take the input data array, convert to HSV values in the given colormap,
        then adjust those color values to give the impression of a shaded
        relief map with a specified light source.  RGBA values are returned,
        which can then be used to plot the shaded image with imshow.

        The color of the resulting image will be darkened by moving the (s, v)
        values (in HSV colorspace) toward (hsv_min_sat, hsv_min_val) in the
        shaded regions, or lightened by sliding (s, v) toward (hsv_max_sat,
        hsv_max_val) in regions that are illuminated.  The default extremes are
        chose so that completely shaded points are nearly black (s = 1, v = 0)
        and completely illuminated points are nearly white (s = 0, v = 1).

        Parameters
        ----------
        rgb : `~numpy.ndarray`
            An MxNx3 RGB array of floats ranging from 0 to 1 (color image).
        intensity : `~numpy.ndarray`
            An MxNx1 array of floats ranging from 0 to 1 (grayscale image).
        hsv_max_sat : number, default: 1
            The maximum saturation value that the *intensity* map can shift the
            output image to.
        hsv_min_sat : number, optional
            The minimum saturation value that the *intensity* map can shift the
            output image to. Defaults to 0.
        hsv_max_val : number, optional
            The maximum value ("v" in "hsv") that the *intensity* map can shift
            the output image to. Defaults to 1.
        hsv_min_val : number, optional
            The minimum value ("v" in "hsv") that the *intensity* map can shift
            the output image to. Defaults to 0.

        Returns
        -------
        `~numpy.ndarray`
            An MxNx3 RGB array representing the combined images.
        """
        # Backward compatibility...
        if hsv_max_sat is None:
            hsv_max_sat = self.hsv_max_sat
        if hsv_max_val is None:
            hsv_max_val = self.hsv_max_val
        if hsv_min_sat is None:
            hsv_min_sat = self.hsv_min_sat
        if hsv_min_val is None:
            hsv_min_val = self.hsv_min_val

        # Expects a 2D intensity array scaled between -1 to 1...
        intensity = intensity[..., 0]
        intensity = 2 * intensity - 1

        # Convert to rgb, then rgb to hsv
        hsv = rgb_to_hsv(rgb[:, :, 0:3])
        hue, sat, val = np.moveaxis(hsv, -1, 0)

        # Modify hsv values (in place) to simulate illumination.
        # putmask(A, mask, B) <=> A[mask] = B[mask]
        np.putmask(sat, (np.abs(sat) > 1.e-10) & (intensity > 0),
                   (1 - intensity) * sat + intensity * hsv_max_sat)
        np.putmask(sat, (np.abs(sat) > 1.e-10) & (intensity < 0),
                   (1 + intensity) * sat - intensity * hsv_min_sat)
        np.putmask(val, intensity > 0,
                   (1 - intensity) * val + intensity * hsv_max_val)
        np.putmask(val, intensity < 0,
                   (1 + intensity) * val - intensity * hsv_min_val)
        np.clip(hsv[:, :, 1:], 0, 1, out=hsv[:, :, 1:])

        # Convert modified hsv back to rgb.
        return hsv_to_rgb(hsv)

    def blend_soft_light(self, rgb, intensity):
        """
        Combine an RGB image with an intensity map using "soft light" blending,
        using the "pegtop" formula.

        Parameters
        ----------
        rgb : `~numpy.ndarray`
            An MxNx3 RGB array of floats ranging from 0 to 1 (color image).
        intensity : `~numpy.ndarray`
            An MxNx1 array of floats ranging from 0 to 1 (grayscale image).

        Returns
        -------
        `~numpy.ndarray`
            An MxNx3 RGB array representing the combined images.
        """
        return 2 * intensity * rgb + (1 - 2 * intensity) * rgb**2

    def blend_overlay(self, rgb, intensity):
        """
        Combine an RGB image with an intensity map using "overlay" blending.

        Parameters
        ----------
        rgb : `~numpy.ndarray`
            An MxNx3 RGB array of floats ranging from 0 to 1 (color image).
        intensity : `~numpy.ndarray`
            An MxNx1 array of floats ranging from 0 to 1 (grayscale image).

        Returns
        -------
        ndarray
            An MxNx3 RGB array representing the combined images.
        """
        low = 2 * intensity * rgb
        high = 1 - 2 * (1 - intensity) * (1 - rgb)
        return np.where(rgb <= 0.5, low, high)


def from_levels_and_colors(levels, colors, extend='neither'):
    """
    A helper routine to generate a cmap and a norm instance which
    behave similar to contourf's levels and colors arguments.

    Parameters
    ----------
    levels : sequence of numbers
        The quantization levels used to construct the `BoundaryNorm`.
        Value ``v`` is quantized to level ``i`` if ``lev[i] <= v < lev[i+1]``.
    colors : sequence of colors
        The fill color to use for each level. If *extend* is "neither" there
        must be ``n_level - 1`` colors. For an *extend* of "min" or "max" add
        one extra color, and for an *extend* of "both" add two colors.
    extend : {'neither', 'min', 'max', 'both'}, optional
        The behaviour when a value falls out of range of the given levels.
        See `~.Axes.contourf` for details.

    Returns
    -------
    cmap : `~matplotlib.colors.Normalize`
    norm : `~matplotlib.colors.Colormap`
    """
    slice_map = {
        'both': slice(1, -1),
        'min': slice(1, None),
        'max': slice(0, -1),
        'neither': slice(0, None),
    }
    _api.check_in_list(slice_map, extend=extend)
    color_slice = slice_map[extend]

    n_data_colors = len(levels) - 1
    n_expected = n_data_colors + color_slice.start - (color_slice.stop or 0)
    if len(colors) != n_expected:
        raise ValueError(
            f'With extend == {extend!r} and {len(levels)} levels, '
            f'expected {n_expected} colors, but got {len(colors)}')

    cmap = ListedColormap(colors[color_slice], N=n_data_colors)

    if extend in ['min', 'both']:
        cmap.set_under(colors[0])
    else:
        cmap.set_under('none')

    if extend in ['max', 'both']:
        cmap.set_over(colors[-1])
    else:
        cmap.set_over('none')

    cmap.colorbar_extend = extend

    norm = BoundaryNorm(levels, ncolors=n_data_colors)
    return cmap, norm