import numpy as np from numpy.testing import assert_allclose, assert_array_equal import pytest from scipy.fft import dct, idct, dctn, idctn, dst, idst, dstn, idstn import scipy.fft as fft from scipy import fftpack import math SQRT_2 = math.sqrt(2) # scipy.fft wraps the fftpack versions but with normalized inverse transforms. # So, the forward transforms and definitions are already thoroughly tested in # fftpack/test_real_transforms.py @pytest.mark.parametrize("forward, backward", [(dct, idct), (dst, idst)]) @pytest.mark.parametrize("type", [1, 2, 3, 4]) @pytest.mark.parametrize("n", [2, 3, 4, 5, 10, 16]) @pytest.mark.parametrize("axis", [0, 1]) @pytest.mark.parametrize("norm", [None, 'backward', 'ortho', 'forward']) @pytest.mark.parametrize("orthogonalize", [False, True]) def test_identity_1d(forward, backward, type, n, axis, norm, orthogonalize): # Test the identity f^-1(f(x)) == x x = np.random.rand(n, n) y = forward(x, type, axis=axis, norm=norm, orthogonalize=orthogonalize) z = backward(y, type, axis=axis, norm=norm, orthogonalize=orthogonalize) assert_allclose(z, x) pad = [(0, 0)] * 2 pad[axis] = (0, 4) y2 = np.pad(y, pad, mode='edge') z2 = backward(y2, type, n, axis, norm, orthogonalize=orthogonalize) assert_allclose(z2, x) @pytest.mark.parametrize("forward, backward", [(dct, idct), (dst, idst)]) @pytest.mark.parametrize("type", [1, 2, 3, 4]) @pytest.mark.parametrize("dtype", [np.float16, np.float32, np.float64, np.complex64, np.complex128]) @pytest.mark.parametrize("axis", [0, 1]) @pytest.mark.parametrize("norm", [None, 'backward', 'ortho', 'forward']) @pytest.mark.parametrize("overwrite_x", [True, False]) def test_identity_1d_overwrite(forward, backward, type, dtype, axis, norm, overwrite_x): # Test the identity f^-1(f(x)) == x x = np.random.rand(7, 8).astype(dtype) x_orig = x.copy() y = forward(x, type, axis=axis, norm=norm, overwrite_x=overwrite_x) y_orig = y.copy() z = backward(y, type, axis=axis, norm=norm, overwrite_x=overwrite_x) if not overwrite_x: assert_allclose(z, x, rtol=1e-6, atol=1e-6) assert_array_equal(x, x_orig) assert_array_equal(y, y_orig) else: assert_allclose(z, x_orig, rtol=1e-6, atol=1e-6) @pytest.mark.parametrize("forward, backward", [(dctn, idctn), (dstn, idstn)]) @pytest.mark.parametrize("type", [1, 2, 3, 4]) @pytest.mark.parametrize("shape, axes", [ ((4, 4), 0), ((4, 4), 1), ((4, 4), None), ((4, 4), (0, 1)), ((10, 12), None), ((10, 12), (0, 1)), ((4, 5, 6), None), ((4, 5, 6), 1), ((4, 5, 6), (0, 2)), ]) @pytest.mark.parametrize("norm", [None, 'backward', 'ortho', 'forward']) @pytest.mark.parametrize("orthogonalize", [False, True]) def test_identity_nd(forward, backward, type, shape, axes, norm, orthogonalize): # Test the identity f^-1(f(x)) == x x = np.random.random(shape) if axes is not None: shape = np.take(shape, axes) y = forward(x, type, axes=axes, norm=norm, orthogonalize=orthogonalize) z = backward(y, type, axes=axes, norm=norm, orthogonalize=orthogonalize) assert_allclose(z, x) if axes is None: pad = [(0, 4)] * x.ndim elif isinstance(axes, int): pad = [(0, 0)] * x.ndim pad[axes] = (0, 4) else: pad = [(0, 0)] * x.ndim for a in axes: pad[a] = (0, 4) y2 = np.pad(y, pad, mode='edge') z2 = backward(y2, type, shape, axes, norm, orthogonalize=orthogonalize) assert_allclose(z2, x) @pytest.mark.parametrize("forward, backward", [(dctn, idctn), (dstn, idstn)]) @pytest.mark.parametrize("type", [1, 2, 3, 4]) @pytest.mark.parametrize("shape, axes", [ ((4, 5), 0), ((4, 5), 1), ((4, 5), None), ]) @pytest.mark.parametrize("dtype", [np.float16, np.float32, np.float64, np.complex64, np.complex128]) @pytest.mark.parametrize("norm", [None, 'backward', 'ortho', 'forward']) @pytest.mark.parametrize("overwrite_x", [False, True]) def test_identity_nd_overwrite(forward, backward, type, shape, axes, dtype, norm, overwrite_x): # Test the identity f^-1(f(x)) == x x = np.random.random(shape).astype(dtype) x_orig = x.copy() if axes is not None: shape = np.take(shape, axes) y = forward(x, type, axes=axes, norm=norm) y_orig = y.copy() z = backward(y, type, axes=axes, norm=norm) if overwrite_x: assert_allclose(z, x_orig, rtol=1e-6, atol=1e-6) else: assert_allclose(z, x, rtol=1e-6, atol=1e-6) assert_array_equal(x, x_orig) assert_array_equal(y, y_orig) @pytest.mark.parametrize("func", ['dct', 'dst', 'dctn', 'dstn']) @pytest.mark.parametrize("type", [1, 2, 3, 4]) @pytest.mark.parametrize("norm", [None, 'backward', 'ortho', 'forward']) def test_fftpack_equivalience(func, type, norm): x = np.random.rand(8, 16) fft_res = getattr(fft, func)(x, type, norm=norm) fftpack_res = getattr(fftpack, func)(x, type, norm=norm) assert_allclose(fft_res, fftpack_res) @pytest.mark.parametrize("func", [dct, dst, dctn, dstn]) @pytest.mark.parametrize("type", [1, 2, 3, 4]) def test_orthogonalize_default(func, type): # Test orthogonalize is the default when norm="ortho", but not otherwise x = np.random.rand(100) for norm, ortho in [ ("forward", False), ("backward", False), ("ortho", True), ]: a = func(x, type=type, norm=norm, orthogonalize=ortho) b = func(x, type=type, norm=norm) assert_allclose(a, b) @pytest.mark.parametrize("norm", ["backward", "ortho", "forward"]) @pytest.mark.parametrize("func, type", [ (dct, 4), (dst, 1), (dst, 4)]) def test_orthogonalize_noop(func, type, norm): # Transforms where orthogonalize is a no-op x = np.random.rand(100) y1 = func(x, type=type, norm=norm, orthogonalize=True) y2 = func(x, type=type, norm=norm, orthogonalize=False) assert_allclose(y1, y2) @pytest.mark.parametrize("norm", ["backward", "ortho", "forward"]) def test_orthogonalize_dct1(norm): x = np.random.rand(100) x2 = x.copy() x2[0] *= SQRT_2 x2[-1] *= SQRT_2 y1 = dct(x, type=1, norm=norm, orthogonalize=True) y2 = dct(x2, type=1, norm=norm, orthogonalize=False) y2[0] /= SQRT_2 y2[-1] /= SQRT_2 assert_allclose(y1, y2) @pytest.mark.parametrize("norm", ["backward", "ortho", "forward"]) @pytest.mark.parametrize("func", [dct, dst]) def test_orthogonalize_dcst2(func, norm): x = np.random.rand(100) y1 = func(x, type=2, norm=norm, orthogonalize=True) y2 = func(x, type=2, norm=norm, orthogonalize=False) y2[0] /= SQRT_2 assert_allclose(y1, y2) @pytest.mark.parametrize("norm", ["backward", "ortho", "forward"]) @pytest.mark.parametrize("func", [dct, dst]) def test_orthogonalize_dcst3(func, norm): x = np.random.rand(100) x2 = x.copy() x2[0] *= SQRT_2 y1 = func(x, type=3, norm=norm, orthogonalize=True) y2 = func(x2, type=3, norm=norm, orthogonalize=False) assert_allclose(y1, y2)