scipy 1.16.2__cp311-cp311-win_arm64.whl

scipy/ndimage/tests/test_ni_support.py

@@ -0,0 +1,78 @@
+import pytest
+
+import numpy as np
+from .._ni_support import _get_output
+
+
+@pytest.mark.parametrize(
+    'dtype',
+    [
+        # String specifiers
+        'f4', 'float32', 'complex64', 'complex128',
+        # Type and dtype specifiers
+        np.float32, float, np.dtype('f4'),
+        # Derive from input
+        None,
+    ],
+)
+def test_get_output_basic(dtype):
+    shape = (2, 3)
+
+    input_ = np.zeros(shape, dtype='float32')
+
+    # For None, derive dtype from input
+    expected_dtype = 'float32' if dtype is None else dtype
+
+    # Output is dtype-specifier, retrieve shape from input
+    result = _get_output(dtype, input_)
+    assert result.shape == shape
+    assert result.dtype == np.dtype(expected_dtype)
+
+    # Output is dtype specifier, with explicit shape, overriding input
+    result = _get_output(dtype, input_, shape=(3, 2))
+    assert result.shape == (3, 2)
+    assert result.dtype == np.dtype(expected_dtype)
+
+    # Output is pre-allocated array, return directly
+    output = np.zeros(shape, dtype=dtype)
+    result = _get_output(output, input_)
+    assert result is output
+
+
+@pytest.mark.thread_unsafe
+def test_get_output_complex():
+    shape = (2, 3)
+
+    input_ = np.zeros(shape)
+
+    # None, promote input type to complex
+    result = _get_output(None, input_, complex_output=True)
+    assert result.shape == shape
+    assert result.dtype == np.dtype('complex128')
+
+    # Explicit type, promote type to complex
+    with pytest.warns(UserWarning, match='promoting specified output dtype to complex'):
+        result = _get_output(float, input_, complex_output=True)
+    assert result.shape == shape
+    assert result.dtype == np.dtype('complex128')
+
+    # String specifier, simply verify complex output
+    result = _get_output('complex64', input_, complex_output=True)
+    assert result.shape == shape
+    assert result.dtype == np.dtype('complex64')
+
+
+def test_get_output_error_cases():
+    input_ = np.zeros((2, 3), 'float32')
+
+    # Two separate paths can raise the same error
+    with pytest.raises(RuntimeError, match='output must have complex dtype'):
+        _get_output('float32', input_, complex_output=True)
+    with pytest.raises(RuntimeError, match='output must have complex dtype'):
+        _get_output(np.zeros((2, 3)), input_, complex_output=True)
+
+    with pytest.raises(RuntimeError, match='output must have numeric dtype'):
+        _get_output('void', input_)
+
+    with pytest.raises(RuntimeError, match='shape not correct'):
+        _get_output(np.zeros((3, 2)), input_)

scipy/ndimage/tests/test_splines.py

@@ -0,0 +1,70 @@
+"""Tests for spline filtering."""
+import pytest
+
+import numpy as np
+from scipy._lib._array_api import assert_almost_equal
+
+from scipy import ndimage
+
+skip_xp_backends = pytest.mark.skip_xp_backends
+pytestmark = [skip_xp_backends(cpu_only=True, exceptions=['cupy', 'jax.numpy'])]
+
+
+def get_spline_knot_values(order):
+    """Knot values to the right of a B-spline's center."""
+    knot_values = {0: [1],
+                   1: [1],
+                   2: [6, 1],
+                   3: [4, 1],
+                   4: [230, 76, 1],
+                   5: [66, 26, 1]}
+
+    return knot_values[order]
+
+
+def make_spline_knot_matrix(xp, n, order, mode='mirror'):
+    """Matrix to invert to find the spline coefficients."""
+    knot_values = get_spline_knot_values(order)
+
+    # NB: do computations with numpy, convert to xp as the last step only
+
+    matrix = np.zeros((n, n))
+    for diag, knot_value in enumerate(knot_values):
+        indices = np.arange(diag, n)
+        if diag == 0:
+            matrix[indices, indices] = knot_value
+        else:
+            matrix[indices, indices - diag] = knot_value
+            matrix[indices - diag, indices] = knot_value
+
+    knot_values_sum = knot_values[0] + 2 * sum(knot_values[1:])
+
+    if mode == 'mirror':
+        start, step = 1, 1
+    elif mode == 'reflect':
+        start, step = 0, 1
+    elif mode == 'grid-wrap':
+        start, step = -1, -1
+    else:
+        raise ValueError(f'unsupported mode {mode}')
+
+    for row in range(len(knot_values) - 1):
+        for idx, knot_value in enumerate(knot_values[row + 1:]):
+            matrix[row, start + step*idx] += knot_value
+            matrix[-row - 1, -start - 1 - step*idx] += knot_value
+
+    return xp.asarray(matrix / knot_values_sum)
+
+
+@pytest.mark.parametrize('order', [0, 1, 2, 3, 4, 5])
+@pytest.mark.parametrize('mode', ['mirror', 'grid-wrap', 'reflect'])
+def test_spline_filter_vs_matrix_solution(order, mode, xp):
+    n = 100
+    eye = xp.eye(n, dtype=xp.float64)
+    spline_filter_axis_0 = ndimage.spline_filter1d(eye, axis=0, order=order,
+                                                   mode=mode)
+    spline_filter_axis_1 = ndimage.spline_filter1d(eye, axis=1, order=order,
+                                                   mode=mode)
+    matrix = make_spline_knot_matrix(xp, n, order, mode=mode)
+    assert_almost_equal(eye, spline_filter_axis_0 @ matrix)
+    assert_almost_equal(eye, spline_filter_axis_1 @ matrix.T)

scipy/odr/__init__.py

@@ -0,0 +1,131 @@
+"""
+=================================================
+Orthogonal distance regression (:mod:`scipy.odr`)
+=================================================
+
+.. currentmodule:: scipy.odr
+
+Package Content
+===============
+
+.. autosummary::
+   :toctree: generated/
+
+   Data          -- The data to fit.
+   RealData      -- Data with weights as actual std. dev.s and/or covariances.
+   Model         -- Stores information about the function to be fit.
+   ODR           -- Gathers all info & manages the main fitting routine.
+   Output        -- Result from the fit.
+   odr           -- Low-level function for ODR.
+
+   OdrWarning    -- Warning about potential problems when running ODR.
+   OdrError      -- Error exception.
+   OdrStop       -- Stop exception.
+
+   polynomial    -- Factory function for a general polynomial model.
+   exponential   -- Exponential model
+   multilinear   -- Arbitrary-dimensional linear model
+   unilinear     -- Univariate linear model
+   quadratic     -- Quadratic model
+
+Usage information
+=================
+
+Introduction
+------------
+
+Why Orthogonal Distance Regression (ODR)? Sometimes one has
+measurement errors in the explanatory (a.k.a., "independent")
+variable(s), not just the response (a.k.a., "dependent") variable(s).
+Ordinary Least Squares (OLS) fitting procedures treat the data for
+explanatory variables as fixed, i.e., not subject to error of any kind.
+Furthermore, OLS procedures require that the response variables be an
+explicit function of the explanatory variables; sometimes making the
+equation explicit is impractical and/or introduces errors.  ODR can
+handle both of these cases with ease, and can even reduce to the OLS
+case if that is sufficient for the problem.
+
+ODRPACK is a FORTRAN-77 library for performing ODR with possibly
+non-linear fitting functions. It uses a modified trust-region
+Levenberg-Marquardt-type algorithm [1]_ to estimate the function
+parameters.  The fitting functions are provided by Python functions
+operating on NumPy arrays. The required derivatives may be provided
+by Python functions as well, or may be estimated numerically. ODRPACK
+can do explicit or implicit ODR fits, or it can do OLS. Input and
+output variables may be multidimensional. Weights can be provided to
+account for different variances of the observations, and even
+covariances between dimensions of the variables.
+
+The `scipy.odr` package offers an object-oriented interface to
+ODRPACK, in addition to the low-level `odr` function.
+
+Additional background information about ODRPACK can be found in the
+`ODRPACK User's Guide
+<https://docs.scipy.org/doc/external/odrpack_guide.pdf>`_, reading
+which is recommended.
+
+Basic usage
+-----------
+
+1. Define the function you want to fit against.::
+
+       def f(B, x):
+           '''Linear function y = m*x + b'''
+           # B is a vector of the parameters.
+           # x is an array of the current x values.
+           # x is in the same format as the x passed to Data or RealData.
+           #
+           # Return an array in the same format as y passed to Data or RealData.
+           return B[0]*x + B[1]
+
+2. Create a Model.::
+
+       linear = Model(f)
+
+3. Create a Data or RealData instance.::
+
+       mydata = Data(x, y, wd=1./power(sx,2), we=1./power(sy,2))
+
+   or, when the actual covariances are known::
+
+       mydata = RealData(x, y, sx=sx, sy=sy)
+
+4. Instantiate ODR with your data, model and initial parameter estimate.::
+
+       myodr = ODR(mydata, linear, beta0=[1., 2.])
+
+5. Run the fit.::
+
+       myoutput = myodr.run()
+
+6. Examine output.::
+
+       myoutput.pprint()
+
+
+References
+----------
+.. [1] P. T. Boggs and J. E. Rogers, "Orthogonal Distance Regression,"
+   in "Statistical analysis of measurement error models and
+   applications: proceedings of the AMS-IMS-SIAM joint summer research
+   conference held June 10-16, 1989," Contemporary Mathematics,
+   vol. 112, pg. 186, 1990.
+
+"""
+# version: 0.7
+# author: Robert Kern <robert.kern@gmail.com>
+# date: 2006-09-21
+
+from ._odrpack import *
+from ._models import *
+from . import _add_newdocs
+
+# Deprecated namespaces, to be removed in v2.0.0
+from . import models, odrpack
+
+__all__ = [s for s in dir()
+           if not (s.startswith('_') or s in ('odr_stop', 'odr_error'))]
+
+from scipy._lib._testutils import PytestTester
+test = PytestTester(__name__)
+del PytestTester

scipy/odr/_add_newdocs.py

@@ -0,0 +1,34 @@
+from numpy.lib import add_newdoc
+
+add_newdoc('scipy.odr', 'odr',
+    """
+    odr(fcn, beta0, y, x, we=None, wd=None, fjacb=None, fjacd=None, extra_args=None,
+        ifixx=None, ifixb=None, job=0, iprint=0, errfile=None, rptfile=None, ndigit=0,
+        taufac=0.0, sstol=-1.0, partol=-1.0, maxit=-1, stpb=None, stpd=None, sclb=None,
+        scld=None, work=None, iwork=None, full_output=0)
+
+    Low-level function for ODR.
+
+    See Also
+    --------
+    ODR : The ODR class gathers all information and coordinates the running of the
+          main fitting routine.
+    Model : The Model class stores information about the function you wish to fit.
+    Data : The data to fit.
+    RealData : Data with weights as actual std. dev.s and/or covariances.
+
+    Notes
+    -----
+    This is a function performing the same operation as the `ODR`,
+    `Model`, and `Data` classes together. The parameters of this
+    function are explained in the class documentation.
+
+    """)
+
+add_newdoc('scipy.odr.__odrpack', '_set_exceptions',
+    """
+    _set_exceptions(odr_error, odr_stop)
+
+    Internal function: set exception classes.
+
+    """)

scipy/odr/_models.py

@@ -0,0 +1,315 @@
+""" Collection of Model instances for use with the odrpack fitting package.
+"""
+import numpy as np
+from scipy.odr._odrpack import Model
+
+__all__ = ['Model', 'exponential', 'multilinear', 'unilinear', 'quadratic',
+           'polynomial']
+
+
+def _lin_fcn(B, x):
+    a, b = B[0], B[1:]
+    b.shape = (b.shape[0], 1)
+
+    return a + (x*b).sum(axis=0)
+
+
+def _lin_fjb(B, x):
+    a = np.ones(x.shape[-1], float)
+    res = np.concatenate((a, x.ravel()))
+    res.shape = (B.shape[-1], x.shape[-1])
+    return res
+
+
+def _lin_fjd(B, x):
+    b = B[1:]
+    b = np.repeat(b, (x.shape[-1],)*b.shape[-1], axis=0)
+    b.shape = x.shape
+    return b
+
+
+def _lin_est(data):
+    # Eh. The answer is analytical, so just return all ones.
+    # Don't return zeros since that will interfere with
+    # ODRPACK's auto-scaling procedures.
+
+    if len(data.x.shape) == 2:
+        m = data.x.shape[0]
+    else:
+        m = 1
+
+    return np.ones((m + 1,), float)
+
+
+def _poly_fcn(B, x, powers):
+    a, b = B[0], B[1:]
+    b.shape = (b.shape[0], 1)
+
+    return a + np.sum(b * np.power(x, powers), axis=0)
+
+
+def _poly_fjacb(B, x, powers):
+    res = np.concatenate((np.ones(x.shape[-1], float),
+                          np.power(x, powers).flat))
+    res.shape = (B.shape[-1], x.shape[-1])
+    return res
+
+
+def _poly_fjacd(B, x, powers):
+    b = B[1:]
+    b.shape = (b.shape[0], 1)
+
+    b = b * powers
+
+    return np.sum(b * np.power(x, powers-1), axis=0)
+
+
+def _exp_fcn(B, x):
+    return B[0] + np.exp(B[1] * x)
+
+
+def _exp_fjd(B, x):
+    return B[1] * np.exp(B[1] * x)
+
+
+def _exp_fjb(B, x):
+    res = np.concatenate((np.ones(x.shape[-1], float), x * np.exp(B[1] * x)))
+    res.shape = (2, x.shape[-1])
+    return res
+
+
+def _exp_est(data):
+    # Eh.
+    return np.array([1., 1.])
+
+
+class _MultilinearModel(Model):
+    r"""
+    Arbitrary-dimensional linear model
+
+    This model is defined by :math:`y=\beta_0 + \sum_{i=1}^m \beta_i x_i`
+
+    Examples
+    --------
+    We can calculate orthogonal distance regression with an arbitrary
+    dimensional linear model:
+
+    >>> from scipy import odr
+    >>> import numpy as np
+    >>> x = np.linspace(0.0, 5.0)
+    >>> y = 10.0 + 5.0 * x
+    >>> data = odr.Data(x, y)
+    >>> odr_obj = odr.ODR(data, odr.multilinear)
+    >>> output = odr_obj.run()
+    >>> print(output.beta)
+    [10.  5.]
+
+    """
+
+    def __init__(self):
+        super().__init__(
+            _lin_fcn, fjacb=_lin_fjb, fjacd=_lin_fjd, estimate=_lin_est,
+            meta={'name': 'Arbitrary-dimensional Linear',
+                  'equ': 'y = B_0 + Sum[i=1..m, B_i * x_i]',
+                  'TeXequ': r'$y=\beta_0 + \sum_{i=1}^m \beta_i x_i$'})
+
+
+multilinear = _MultilinearModel()
+
+
+def polynomial(order):
+    """
+    Factory function for a general polynomial model.
+
+    Parameters
+    ----------
+    order : int or sequence
+        If an integer, it becomes the order of the polynomial to fit. If
+        a sequence of numbers, then these are the explicit powers in the
+        polynomial.
+        A constant term (power 0) is always included, so don't include 0.
+        Thus, polynomial(n) is equivalent to polynomial(range(1, n+1)).
+
+    Returns
+    -------
+    polynomial : Model instance
+        Model instance.
+
+    Examples
+    --------
+    We can fit an input data using orthogonal distance regression (ODR) with
+    a polynomial model:
+
+    >>> import numpy as np
+    >>> import matplotlib.pyplot as plt
+    >>> from scipy import odr
+    >>> x = np.linspace(0.0, 5.0)
+    >>> y = np.sin(x)
+    >>> poly_model = odr.polynomial(3)  # using third order polynomial model
+    >>> data = odr.Data(x, y)
+    >>> odr_obj = odr.ODR(data, poly_model)
+    >>> output = odr_obj.run()  # running ODR fitting
+    >>> poly = np.poly1d(output.beta[::-1])
+    >>> poly_y = poly(x)
+    >>> plt.plot(x, y, label="input data")
+    >>> plt.plot(x, poly_y, label="polynomial ODR")
+    >>> plt.legend()
+    >>> plt.show()
+
+    """
+
+    powers = np.asarray(order)
+    if powers.shape == ():
+        # Scalar.
+        powers = np.arange(1, powers + 1)
+
+    powers.shape = (len(powers), 1)
+    len_beta = len(powers) + 1
+
+    def _poly_est(data, len_beta=len_beta):
+        # Eh. Ignore data and return all ones.
+        return np.ones((len_beta,), float)
+
+    return Model(_poly_fcn, fjacd=_poly_fjacd, fjacb=_poly_fjacb,
+                 estimate=_poly_est, extra_args=(powers,),
+                 meta={'name': 'Sorta-general Polynomial',
+                 'equ': 'y = B_0 + Sum[i=1..%s, B_i * (x**i)]' % (len_beta-1),
+                 'TeXequ': r'$y=\beta_0 + \sum_{i=1}^{%s} \beta_i x^i$' %
+                        (len_beta-1)})
+
+
+class _ExponentialModel(Model):
+    r"""
+    Exponential model
+
+    This model is defined by :math:`y=\beta_0 + e^{\beta_1 x}`
+
+    Examples
+    --------
+    We can calculate orthogonal distance regression with an exponential model:
+
+    >>> from scipy import odr
+    >>> import numpy as np
+    >>> x = np.linspace(0.0, 5.0)
+    >>> y = -10.0 + np.exp(0.5*x)
+    >>> data = odr.Data(x, y)
+    >>> odr_obj = odr.ODR(data, odr.exponential)
+    >>> output = odr_obj.run()
+    >>> print(output.beta)
+    [-10.    0.5]
+
+    """
+
+    def __init__(self):
+        super().__init__(_exp_fcn, fjacd=_exp_fjd, fjacb=_exp_fjb,
+                         estimate=_exp_est,
+                         meta={'name': 'Exponential',
+                               'equ': 'y= B_0 + exp(B_1 * x)',
+                               'TeXequ': r'$y=\beta_0 + e^{\beta_1 x}$'})
+
+
+exponential = _ExponentialModel()
+
+
+def _unilin(B, x):
+    return x*B[0] + B[1]
+
+
+def _unilin_fjd(B, x):
+    return np.ones(x.shape, float) * B[0]
+
+
+def _unilin_fjb(B, x):
+    _ret = np.concatenate((x, np.ones(x.shape, float)))
+    _ret.shape = (2,) + x.shape
+
+    return _ret
+
+
+def _unilin_est(data):
+    return (1., 1.)
+
+
+def _quadratic(B, x):
+    return x*(x*B[0] + B[1]) + B[2]
+
+
+def _quad_fjd(B, x):
+    return 2*x*B[0] + B[1]
+
+
+def _quad_fjb(B, x):
+    _ret = np.concatenate((x*x, x, np.ones(x.shape, float)))
+    _ret.shape = (3,) + x.shape
+
+    return _ret
+
+
+def _quad_est(data):
+    return (1.,1.,1.)
+
+
+class _UnilinearModel(Model):
+    r"""
+    Univariate linear model
+
+    This model is defined by :math:`y = \beta_0 x + \beta_1`
+
+    Examples
+    --------
+    We can calculate orthogonal distance regression with an unilinear model:
+
+    >>> from scipy import odr
+    >>> import numpy as np
+    >>> x = np.linspace(0.0, 5.0)
+    >>> y = 1.0 * x + 2.0
+    >>> data = odr.Data(x, y)
+    >>> odr_obj = odr.ODR(data, odr.unilinear)
+    >>> output = odr_obj.run()
+    >>> print(output.beta)
+    [1. 2.]
+
+    """
+
+    def __init__(self):
+        super().__init__(_unilin, fjacd=_unilin_fjd, fjacb=_unilin_fjb,
+                         estimate=_unilin_est,
+                         meta={'name': 'Univariate Linear',
+                               'equ': 'y = B_0 * x + B_1',
+                               'TeXequ': '$y = \\beta_0 x + \\beta_1$'})
+
+
+unilinear = _UnilinearModel()
+
+
+class _QuadraticModel(Model):
+    r"""
+    Quadratic model
+
+    This model is defined by :math:`y = \beta_0 x^2 + \beta_1 x + \beta_2`
+
+    Examples
+    --------
+    We can calculate orthogonal distance regression with a quadratic model:
+
+    >>> from scipy import odr
+    >>> import numpy as np
+    >>> x = np.linspace(0.0, 5.0)
+    >>> y = 1.0 * x ** 2 + 2.0 * x + 3.0
+    >>> data = odr.Data(x, y)
+    >>> odr_obj = odr.ODR(data, odr.quadratic)
+    >>> output = odr_obj.run()
+    >>> print(output.beta)
+    [1. 2. 3.]
+
+    """
+
+    def __init__(self):
+        super().__init__(
+            _quadratic, fjacd=_quad_fjd, fjacb=_quad_fjb, estimate=_quad_est,
+            meta={'name': 'Quadratic',
+                  'equ': 'y = B_0*x**2 + B_1*x + B_2',
+                  'TeXequ': '$y = \\beta_0 x^2 + \\beta_1 x + \\beta_2'})
+
+
+quadratic = _QuadraticModel()