Fixed database typo and removed unnecessary class identifier.

venv/Lib/site-packages/scipy/special/_precompute/expn_asy.py

@@ -0,0 +1,59 @@
+"""Precompute the polynomials for the asymptotic expansion of the
+generalized exponential integral.
+
+Sources
+-------
+[1] NIST, Digital Library of Mathematical Functions,
+    https://dlmf.nist.gov/8.20#ii
+
+"""
+import os
+from numpy.testing import suppress_warnings
+
+try:
+    # Can remove when sympy #11255 is resolved; see
+    # https://github.com/sympy/sympy/issues/11255
+    with suppress_warnings() as sup:
+        sup.filter(DeprecationWarning, "inspect.getargspec.. is deprecated")
+        import sympy  # type: ignore[import]
+        from sympy import Poly
+        x = sympy.symbols('x')
+except ImportError:
+    pass
+
+
+def generate_A(K):
+    A = [Poly(1, x)]
+    for k in range(K):
+        A.append(Poly(1 - 2*k*x, x)*A[k] + Poly(x*(x + 1))*A[k].diff())
+    return A
+
+
+WARNING = """\
+/* This file was automatically generated by _precompute/expn_asy.py.
+ * Do not edit it manually!
+ */
+"""
+
+
+def main():
+    print(__doc__)
+    fn = os.path.join('..', 'cephes', 'expn.h')
+
+    K = 12
+    A = generate_A(K)
+    with open(fn + '.new', 'w') as f:
+        f.write(WARNING)
+        f.write("#define nA {}\n".format(len(A)))
+        for k, Ak in enumerate(A):
+            tmp = ', '.join([str(x.evalf(18)) for x in Ak.coeffs()])
+            f.write("static const double A{}[] = {{{}}};\n".format(k, tmp))
+        tmp = ", ".join(["A{}".format(k) for k in range(K + 1)])
+        f.write("static const double *A[] = {{{}}};\n".format(tmp))
+        tmp = ", ".join([str(Ak.degree()) for Ak in A])
+        f.write("static const int Adegs[] = {{{}}};\n".format(tmp))
+    os.rename(fn + '.new', fn)
+
+
+if __name__ == "__main__":
+    main()

venv/Lib/site-packages/scipy/special/_precompute/gammainc_asy.py

@@ -0,0 +1,115 @@
+"""
+Precompute coefficients of Temme's asymptotic expansion for gammainc.
+
+This takes about 8 hours to run on a 2.3 GHz Macbook Pro with 4GB ram.
+
+Sources:
+[1] NIST, "Digital Library of Mathematical Functions",
+    https://dlmf.nist.gov/
+
+"""
+import os
+from scipy.special._precompute.utils import lagrange_inversion
+
+try:
+    import mpmath as mp  # type: ignore[import]
+except ImportError:
+    pass
+
+
+def compute_a(n):
+    """a_k from DLMF 5.11.6"""
+    a = [mp.sqrt(2)/2]
+    for k in range(1, n):
+        ak = a[-1]/k
+        for j in range(1, len(a)):
+            ak -= a[j]*a[-j]/(j + 1)
+        ak /= a[0]*(1 + mp.mpf(1)/(k + 1))
+        a.append(ak)
+    return a
+
+
+def compute_g(n):
+    """g_k from DLMF 5.11.3/5.11.5"""
+    a = compute_a(2*n)
+    g = [mp.sqrt(2)*mp.rf(0.5, k)*a[2*k] for k in range(n)]
+    return g
+
+
+def eta(lam):
+    """Function from DLMF 8.12.1 shifted to be centered at 0."""
+    if lam > 0:
+        return mp.sqrt(2*(lam - mp.log(lam + 1)))
+    elif lam < 0:
+        return -mp.sqrt(2*(lam - mp.log(lam + 1)))
+    else:
+        return 0
+
+
+def compute_alpha(n):
+    """alpha_n from DLMF 8.12.13"""
+    coeffs = mp.taylor(eta, 0, n - 1)
+    return lagrange_inversion(coeffs)
+
+
+def compute_d(K, N):
+    """d_{k, n} from DLMF 8.12.12"""
+    M = N + 2*K
+    d0 = [-mp.mpf(1)/3]
+    alpha = compute_alpha(M + 2)
+    for n in range(1, M):
+        d0.append((n + 2)*alpha[n+2])
+    d = [d0]
+    g = compute_g(K)
+    for k in range(1, K):
+        dk = []
+        for n in range(M - 2*k):
+            dk.append((-1)**k*g[k]*d[0][n] + (n + 2)*d[k-1][n+2])
+        d.append(dk)
+    for k in range(K):
+        d[k] = d[k][:N]
+    return d
+
+
+header = \
+r"""/* This file was automatically generated by _precomp/gammainc.py.
+ * Do not edit it manually!
+ */
+
+#ifndef IGAM_H
+#define IGAM_H
+
+#define K {}
+#define N {}
+
+static const double d[K][N] =
+{{"""
+
+footer = \
+r"""
+#endif
+"""
+
+def main():
+    print(__doc__)
+    K = 25
+    N = 25
+    with mp.workdps(50):
+        d = compute_d(K, N)
+    fn = os.path.join(os.path.dirname(__file__), '..', 'cephes', 'igam.h')
+    with open(fn + '.new', 'w') as f:
+        f.write(header.format(K, N))
+        for k, row in enumerate(d):
+            row = map(lambda x: mp.nstr(x, 17, min_fixed=0, max_fixed=0), row)
+            f.write('{')
+            f.write(", ".join(row))
+            if k < K - 1:
+                f.write('},\n')
+            else:
+                f.write('}};\n')
+        f.write(footer)
+    os.rename(fn + '.new', fn)
+
+
+if __name__ == "__main__":
+    main()

venv/Lib/site-packages/scipy/special/_precompute/gammainc_data.py

@@ -0,0 +1,124 @@
+"""Compute gammainc and gammaincc for large arguments and parameters
+and save the values to data files for use in tests. We can't just
+compare to mpmath's gammainc in test_mpmath.TestSystematic because it
+would take too long.
+
+Note that mpmath's gammainc is computed using hypercomb, but since it
+doesn't allow the user to increase the maximum number of terms used in
+the series it doesn't converge for many arguments. To get around this
+we copy the mpmath implementation but use more terms.
+
+This takes about 17 minutes to run on a 2.3 GHz Macbook Pro with 4GB
+ram.
+
+Sources:
+[1] Fredrik Johansson and others. mpmath: a Python library for
+    arbitrary-precision floating-point arithmetic (version 0.19),
+    December 2013. http://mpmath.org/.
+
+"""
+import os
+from time import time
+import numpy as np
+from numpy import pi
+
+from scipy.special._mptestutils import mpf2float
+
+try:
+    import mpmath as mp  # type: ignore[import]
+except ImportError:
+    pass
+
+
+def gammainc(a, x, dps=50, maxterms=10**8):
+    """Compute gammainc exactly like mpmath does but allow for more
+    summands in hypercomb. See
+
+    mpmath/functions/expintegrals.py#L134
+
+    in the mpmath github repository.
+
+    """
+    with mp.workdps(dps):
+        z, a, b = mp.mpf(a), mp.mpf(x), mp.mpf(x)
+        G = [z]
+        negb = mp.fneg(b, exact=True)
+
+        def h(z):
+            T1 = [mp.exp(negb), b, z], [1, z, -1], [], G, [1], [1+z], b
+            return (T1,)
+
+        res = mp.hypercomb(h, [z], maxterms=maxterms)
+        return mpf2float(res)
+
+
+def gammaincc(a, x, dps=50, maxterms=10**8):
+    """Compute gammaincc exactly like mpmath does but allow for more
+    terms in hypercomb. See
+
+    mpmath/functions/expintegrals.py#L187
+
+    in the mpmath github repository.
+
+    """
+    with mp.workdps(dps):
+        z, a = a, x
+
+        if mp.isint(z):
+            try:
+                # mpmath has a fast integer path
+                return mpf2float(mp.gammainc(z, a=a, regularized=True))
+            except mp.libmp.NoConvergence:
+                pass
+        nega = mp.fneg(a, exact=True)
+        G = [z]
+        # Use 2F0 series when possible; fall back to lower gamma representation
+        try:
+            def h(z):
+                r = z-1
+                return [([mp.exp(nega), a], [1, r], [], G, [1, -r], [], 1/nega)]
+            return mpf2float(mp.hypercomb(h, [z], force_series=True))
+        except mp.libmp.NoConvergence:
+            def h(z):
+                T1 = [], [1, z-1], [z], G, [], [], 0
+                T2 = [-mp.exp(nega), a, z], [1, z, -1], [], G, [1], [1+z], a
+                return T1, T2
+            return mpf2float(mp.hypercomb(h, [z], maxterms=maxterms))
+
+
+def main():
+    t0 = time()
+    # It would be nice to have data for larger values, but either this
+    # requires prohibitively large precision (dps > 800) or mpmath has
+    # a bug. For example, gammainc(1e20, 1e20, dps=800) returns a
+    # value around 0.03, while the true value should be close to 0.5
+    # (DLMF 8.12.15).
+    print(__doc__)
+    pwd = os.path.dirname(__file__)
+    r = np.logspace(4, 14, 30)
+    ltheta = np.logspace(np.log10(pi/4), np.log10(np.arctan(0.6)), 30)
+    utheta = np.logspace(np.log10(pi/4), np.log10(np.arctan(1.4)), 30)
+
+    regimes = [(gammainc, ltheta), (gammaincc, utheta)]
+    for func, theta in regimes:
+        rg, thetag = np.meshgrid(r, theta)
+        a, x = rg*np.cos(thetag), rg*np.sin(thetag)
+        a, x = a.flatten(), x.flatten()
+        dataset = []
+        for i, (a0, x0) in enumerate(zip(a, x)):
+            if func == gammaincc:
+                # Exploit the fast integer path in gammaincc whenever
+                # possible so that the computation doesn't take too
+                # long
+                a0, x0 = np.floor(a0), np.floor(x0)
+            dataset.append((a0, x0, func(a0, x0)))
+        dataset = np.array(dataset)
+        filename = os.path.join(pwd, '..', 'tests', 'data', 'local',
+                                '{}.txt'.format(func.__name__))
+        np.savetxt(filename, dataset)
+
+    print("{} minutes elapsed".format((time() - t0)/60))
+
+
+if __name__ == "__main__":
+    main()

venv/Lib/site-packages/scipy/special/_precompute/lambertw.py

@@ -0,0 +1,68 @@
+"""Compute a Pade approximation for the principle branch of the
+Lambert W function around 0 and compare it to various other
+approximations.
+
+"""
+import numpy as np
+
+try:
+    import mpmath  # type: ignore[import]
+    import matplotlib.pyplot as plt  # type: ignore[import]
+except ImportError:
+    pass
+
+
+def lambertw_pade():
+    derivs = [mpmath.diff(mpmath.lambertw, 0, n=n) for n in range(6)]
+    p, q = mpmath.pade(derivs, 3, 2)
+    return p, q
+
+
+def main():
+    print(__doc__)
+    with mpmath.workdps(50):
+        p, q = lambertw_pade()
+        p, q = p[::-1], q[::-1]
+        print("p = {}".format(p))
+        print("q = {}".format(q))
+
+    x, y = np.linspace(-1.5, 1.5, 75), np.linspace(-1.5, 1.5, 75)
+    x, y = np.meshgrid(x, y)
+    z = x + 1j*y
+    lambertw_std = []
+    for z0 in z.flatten():
+        lambertw_std.append(complex(mpmath.lambertw(z0)))
+    lambertw_std = np.array(lambertw_std).reshape(x.shape)
+
+    fig, axes = plt.subplots(nrows=3, ncols=1)
+    # Compare Pade approximation to true result
+    p = np.array([float(p0) for p0 in p])
+    q = np.array([float(q0) for q0 in q])
+    pade_approx = np.polyval(p, z)/np.polyval(q, z)
+    pade_err = abs(pade_approx - lambertw_std)
+    axes[0].pcolormesh(x, y, pade_err)
+    # Compare two terms of asymptotic series to true result
+    asy_approx = np.log(z) - np.log(np.log(z))
+    asy_err = abs(asy_approx - lambertw_std)
+    axes[1].pcolormesh(x, y, asy_err)
+    # Compare two terms of the series around the branch point to the
+    # true result
+    p = np.sqrt(2*(np.exp(1)*z + 1))
+    series_approx = -1 + p - p**2/3
+    series_err = abs(series_approx - lambertw_std)
+    im = axes[2].pcolormesh(x, y, series_err)
+
+    fig.colorbar(im, ax=axes.ravel().tolist())
+    plt.show()
+
+    fig, ax = plt.subplots(nrows=1, ncols=1)
+    pade_better = pade_err < asy_err
+    im = ax.pcolormesh(x, y, pade_better)
+    t = np.linspace(-0.3, 0.3)
+    ax.plot(-2.5*abs(t) - 0.2, t, 'r')
+    fig.colorbar(im, ax=ax)
+    plt.show()
+
+
+if __name__ == '__main__':
+    main()

venv/Lib/site-packages/scipy/special/_precompute/loggamma.py

@@ -0,0 +1,43 @@
+"""Precompute series coefficients for log-Gamma."""
+
+try:
+    import mpmath  # type: ignore[import]
+except ImportError:
+    pass
+
+
+def stirling_series(N):
+    with mpmath.workdps(100):
+        coeffs = [mpmath.bernoulli(2*n)/(2*n*(2*n - 1))
+                  for n in range(1, N + 1)]
+    return coeffs
+
+
+def taylor_series_at_1(N):
+    coeffs = []
+    with mpmath.workdps(100):
+        coeffs.append(-mpmath.euler)
+        for n in range(2, N + 1):
+            coeffs.append((-1)**n*mpmath.zeta(n)/n)
+    return coeffs
+
+
+def main():
+    print(__doc__)
+    print()
+    stirling_coeffs = [mpmath.nstr(x, 20, min_fixed=0, max_fixed=0)
+                       for x in stirling_series(8)[::-1]]
+    taylor_coeffs = [mpmath.nstr(x, 20, min_fixed=0, max_fixed=0)
+                     for x in taylor_series_at_1(23)[::-1]]
+    print("Stirling series coefficients")
+    print("----------------------------")
+    print("\n".join(stirling_coeffs))
+    print()
+    print("Taylor series coefficients")
+    print("--------------------------")
+    print("\n".join(taylor_coeffs))
+    print()
+
+
+if __name__ == '__main__':
+    main()

venv/Lib/site-packages/scipy/special/_precompute/setup.py

@@ -0,0 +1,11 @@
+
+def configuration(parent_name='special', top_path=None):
+    from numpy.distutils.misc_util import Configuration
+    config = Configuration('_precompute', parent_name, top_path)
+    return config
+
+
+if __name__ == '__main__':
+    from numpy.distutils.core import setup
+    setup(**configuration().todict())
+

venv/Lib/site-packages/scipy/special/_precompute/struve_convergence.py

@@ -0,0 +1,120 @@
+"""
+Convergence regions of the expansions used in ``struve.c``
+
+Note that for v >> z both functions tend rapidly to 0,
+and for v << -z, they tend to infinity.
+
+The floating-point functions over/underflow in the lower left and right
+corners of the figure.
+
+
+Figure legend
+=============
+
+Red region
+    Power series is close (1e-12) to the mpmath result
+
+Blue region
+    Asymptotic series is close to the mpmath result
+
+Green region
+    Bessel series is close to the mpmath result
+
+Dotted colored lines
+    Boundaries of the regions
+
+Solid colored lines
+    Boundaries estimated by the routine itself. These will be used
+    for determining which of the results to use.
+
+Black dashed line
+    The line z = 0.7*|v| + 12
+
+"""
+import numpy as np
+import matplotlib.pyplot as plt  # type: ignore[import]
+
+import mpmath  # type: ignore[import]
+
+
+def err_metric(a, b, atol=1e-290):
+    m = abs(a - b) / (atol + abs(b))
+    m[np.isinf(b) & (a == b)] = 0
+    return m
+
+
+def do_plot(is_h=True):
+    from scipy.special._ufuncs import (_struve_power_series,
+                                       _struve_asymp_large_z,
+                                       _struve_bessel_series)
+
+    vs = np.linspace(-1000, 1000, 91)
+    zs = np.sort(np.r_[1e-5, 1.0, np.linspace(0, 700, 91)[1:]])
+
+    rp = _struve_power_series(vs[:,None], zs[None,:], is_h)
+    ra = _struve_asymp_large_z(vs[:,None], zs[None,:], is_h)
+    rb = _struve_bessel_series(vs[:,None], zs[None,:], is_h)
+
+    mpmath.mp.dps = 50
+    if is_h:
+        sh = lambda v, z: float(mpmath.struveh(mpmath.mpf(v), mpmath.mpf(z)))
+    else:
+        sh = lambda v, z: float(mpmath.struvel(mpmath.mpf(v), mpmath.mpf(z)))
+    ex = np.vectorize(sh, otypes='d')(vs[:,None], zs[None,:])
+
+    err_a = err_metric(ra[0], ex) + 1e-300
+    err_p = err_metric(rp[0], ex) + 1e-300
+    err_b = err_metric(rb[0], ex) + 1e-300
+
+    err_est_a = abs(ra[1]/ra[0])
+    err_est_p = abs(rp[1]/rp[0])
+    err_est_b = abs(rb[1]/rb[0])
+
+    z_cutoff = 0.7*abs(vs) + 12
+
+    levels = [-1000, -12]
+
+    plt.cla()
+
+    plt.hold(1)
+    plt.contourf(vs, zs, np.log10(err_p).T, levels=levels, colors=['r', 'r'], alpha=0.1)
+    plt.contourf(vs, zs, np.log10(err_a).T, levels=levels, colors=['b', 'b'], alpha=0.1)
+    plt.contourf(vs, zs, np.log10(err_b).T, levels=levels, colors=['g', 'g'], alpha=0.1)
+
+    plt.contour(vs, zs, np.log10(err_p).T, levels=levels, colors=['r', 'r'], linestyles=[':', ':'])
+    plt.contour(vs, zs, np.log10(err_a).T, levels=levels, colors=['b', 'b'], linestyles=[':', ':'])
+    plt.contour(vs, zs, np.log10(err_b).T, levels=levels, colors=['g', 'g'], linestyles=[':', ':'])
+
+    lp = plt.contour(vs, zs, np.log10(err_est_p).T, levels=levels, colors=['r', 'r'], linestyles=['-', '-'])
+    la = plt.contour(vs, zs, np.log10(err_est_a).T, levels=levels, colors=['b', 'b'], linestyles=['-', '-'])
+    lb = plt.contour(vs, zs, np.log10(err_est_b).T, levels=levels, colors=['g', 'g'], linestyles=['-', '-'])
+
+    plt.clabel(lp, fmt={-1000: 'P', -12: 'P'})
+    plt.clabel(la, fmt={-1000: 'A', -12: 'A'})
+    plt.clabel(lb, fmt={-1000: 'B', -12: 'B'})
+
+    plt.plot(vs, z_cutoff, 'k--')
+
+    plt.xlim(vs.min(), vs.max())
+    plt.ylim(zs.min(), zs.max())
+
+    plt.xlabel('v')
+    plt.ylabel('z')
+
+
+def main():
+    plt.clf()
+    plt.subplot(121)
+    do_plot(True)
+    plt.title('Struve H')
+
+    plt.subplot(122)
+    do_plot(False)
+    plt.title('Struve L')
+
+    plt.savefig('struve_convergence.png')
+    plt.show()
+
+
+if __name__ == "__main__":
+    main()

venv/Lib/site-packages/scipy/special/_precompute/utils.py

@@ -0,0 +1,44 @@
+from numpy.testing import suppress_warnings
+
+try:
+    import mpmath as mp  # type: ignore[import]
+except ImportError:
+    pass
+
+try:
+    # Can remove when sympy #11255 is resolved; see
+    # https://github.com/sympy/sympy/issues/11255
+    with suppress_warnings() as sup:
+        sup.filter(DeprecationWarning, "inspect.getargspec.. is deprecated")
+        from sympy.abc import x  # type: ignore[import]
+except ImportError:
+    pass
+
+
+def lagrange_inversion(a):
+    """Given a series
+
+    f(x) = a[1]*x + a[2]*x**2 + ... + a[n-1]*x**(n - 1),
+
+    use the Lagrange inversion formula to compute a series
+
+    g(x) = b[1]*x + b[2]*x**2 + ... + b[n-1]*x**(n - 1)
+
+    so that f(g(x)) = g(f(x)) = x mod x**n. We must have a[0] = 0, so
+    necessarily b[0] = 0 too.
+
+    The algorithm is naive and could be improved, but speed isn't an
+    issue here and it's easy to read.
+
+    """
+    n = len(a)
+    f = sum(a[i]*x**i for i in range(len(a)))
+    h = (x/f).series(x, 0, n).removeO()
+    hpower = [h**0]
+    for k in range(n):
+        hpower.append((hpower[-1]*h).expand())
+    b = [mp.mpf(0)]
+    for k in range(1, n):
+        b.append(hpower[k].coeff(x, k - 1)/k)
+    b = map(lambda x: mp.mpf(x), b)
+    return b

venv/Lib/site-packages/scipy/special/_precompute/wrightomega.py

@@ -0,0 +1,41 @@
+import numpy as np
+
+try:
+    import mpmath  # type: ignore[import]
+except ImportError:
+    pass
+
+
+def mpmath_wrightomega(x):
+    return mpmath.lambertw(mpmath.exp(x), mpmath.mpf('-0.5'))
+
+
+def wrightomega_series_error(x):
+    series = x
+    desired = mpmath_wrightomega(x)
+    return abs(series - desired) / desired
+
+
+def wrightomega_exp_error(x):
+    exponential_approx = mpmath.exp(x)
+    desired = mpmath_wrightomega(x)
+    return abs(exponential_approx - desired) / desired
+
+
+def main():
+    desired_error = 2 * np.finfo(float).eps
+    print('Series Error')
+    for x in [1e5, 1e10, 1e15, 1e20]:
+        with mpmath.workdps(100):
+            error = wrightomega_series_error(x)
+        print(x, error, error < desired_error)
+
+    print('Exp error')
+    for x in [-10, -25, -50, -100, -200, -400, -700, -740]:
+        with mpmath.workdps(100):
+            error = wrightomega_exp_error(x)
+        print(x, error, error < desired_error)
+
+
+if __name__ == '__main__':
+    main()

venv/Lib/site-packages/scipy/special/_precompute/zetac.py

@@ -0,0 +1,27 @@
+"""Compute the Taylor series for zeta(x) - 1 around x = 0."""
+try:
+    import mpmath  # type: ignore[import]
+except ImportError:
+    pass
+
+
+def zetac_series(N):
+    coeffs = []
+    with mpmath.workdps(100):
+        coeffs.append(-1.5)
+        for n in range(1, N):
+            coeff = mpmath.diff(mpmath.zeta, 0, n)/mpmath.factorial(n)
+            coeffs.append(coeff)
+    return coeffs
+
+
+def main():
+    print(__doc__)
+    coeffs = zetac_series(10)
+    coeffs = [mpmath.nstr(x, 20, min_fixed=0, max_fixed=0)
+              for x in coeffs]
+    print("\n".join(coeffs[::-1]))
+
+
+if __name__ == '__main__':
+    main()