def Rlam(x, lam=None):
v1 = kve(lam + 1, x) * np.exp(x)
v0 = kve(lam, x) * np.exp(x)
val = v1 / v0
if np.isinf(v1) or np.isinf(v0) or v0 == 0 or v1 == 0:
lv1 = np.log(sympy.Float(mpmath.besselk(np.abs(lam + 1), x)))
lv0 = np.log(sympy.Float(mpmath.besselk(np.abs(lam), x)))
val = np.exp(lv1 - lv0)
return val
def R0(kpsigr, b, N):
h = float(b) / N
k = 0.0
for i in range(1, N):
x = i * h
if i % 2 == 1:
k += 4.0 * x * mp.besselk(0, x) * rho_perp(x, kpsigr)
else:
k += 2.0 * x * mp.besselk(0, x) * rho_perp(x, kpsigr)
return (h /
3.0) * (b * mp.besselk(0, b) * rho_perp(b, kpsigr) + k)
def f(th_e, A):
tmp = A - chi*(m_e/m_i)*th_e
if (tmp <= 0.0):
tmp = 1.0e-20
th_e = mp.mpf(th_e)
th_i = mp.mpf(tmp)
tmp = (1/(mp.besselk(2, 1/th_e) * mp.besselk(2, 1/th_i))) * (((2*(th_e + th_i)**2 + 1)/(th_e + th_i)) * mp.besselk(1, (th_e + th_i)/(th_e*th_i)) + 2*mp.besselk(0, (th_e + th_i)/(th_e*th_i)))
# print repr(tmp)
return float(tmp)
def mpbesselk(v, x):
r = float(mpmath.besselk(v, x, **HYPERKW))
if abs(r) > 1e305:
# overflowing to inf a bit earlier is OK
r = np.inf * np.sign(r)
if abs(v) == abs(x) and abs(r) == np.inf and abs(x) > 1:
# wrong result (kv(x,x) -> 0 for x > 1),
# try with higher dps
old_dps = mpmath.mp.dps
mpmath.mp.dps = 200
try:
r = float(mpmath.besselk(v, x, **HYPERKW))
finally:
mpmath.mp.dps = old_dps
return r
def mpbesselk(v, x):
r = float(mpmath.besselk(v, x, **HYPERKW))
if abs(r) > 1e305:
# overflowing to inf a bit earlier is OK
r = np.inf * np.sign(r)
if abs(v) == abs(x) and abs(r) == np.inf and abs(x) > 1:
# wrong result (kv(x,x) -> 0 for x > 1),
# try with higher dps
old_dps = mpmath.mp.dps
mpmath.mp.dps = 200
try:
r = float(mpmath.besselk(v, x, **HYPERKW))
finally:
mpmath.mp.dps = old_dps
return r
def roughness_spectrum(sp, xx, wvnb, sig, L, Ts):
wn = np.zeros(Ts)
# -- math.exponential correl func
if sp.lower() == 'math.exponential':
for n in np.arange(1,Ts+1):
wn[n-1] = L ** 2 / n ** 2 * (1 + (wvnb * L / n) ** 2) ** (-1.5)
rss = sig / L
# -- gaussian correl func
if sp.lower() == 'gaussian':
for n in np.arange(1,Ts+1):
wn[n-1] = L ** 2 / (2 * n) * math.exp(-(wvnb * L) ** 2 / (4 * n))
rss = sqrt(2) * sig / L
# -- x - power correl func
if sp.lower() == 'power_spec':
for n in np.arange(1,Ts+1):
if wvnb == 0:
wn[n-1] = L ** 2 / (3 * n - 2)
else:
wn[n-1] = L ** 2 * (wvnb * L) ** (-1 + xx * n) * besselk(1 - xx * n, wvnb * L) \
/ (2 ** (xx * n - 1) * math.gamma(xx * n))
return(wn, rss)
def pdf(x, p, b, loc=0, scale=1):
"""
Probability density function of the generalized inverse Gaussian
distribution.
The PDF for x > loc is:
z**(p - 1) * exp(-b*(z + 1/z)/2))
---------------------------------
s * K_p(b)
where s is the scale, z = (x - loc)/s, and K_p(b) is the modified Bessel
function of the second kind. For x <= loc, the PDF is zero.
"""
x = mpmath.mpf(x)
p = mpmath.mpf(p)
b = mpmath.mpf(b)
loc = mpmath.mpf(loc)
scale = mpmath.mpf(scale)
if x <= loc:
return mpmath.mp.zero
z = (x - loc) / scale
return (mpmath.power(z, p - 1) * mpmath.exp(-b * (z + 1 / z) / 2) /
(2 * mpmath.besselk(p, b)) / scale)
def matern_function(Xi, Xj, *args):
r"""Matern covariance function of arbitrary dimension, for use with :py:class:`ArbitraryKernel`.
The Matern kernel has the following hyperparameters, always referenced in
the order listed:
= ===== ====================================
0 sigma prefactor
1 nu order of kernel
2 l1 length scale for the first dimension
3 l2 ...and so on for all dimensions
= ===== ====================================
The kernel is defined as:
.. math::
k_M = \sigma^2 \frac{2^{1-\nu}}{\Gamma(\nu)}
\left (\sqrt{2\nu \sum_i\left (\frac{\tau_i^2}{l_i^2}\right )}\right )^\nu
K_\nu\left(\sqrt{2\nu \sum_i\left(\frac{\tau_i^2}{l_i^2}\right)}\right)
Parameters
----------
Xi, Xj : :py:class:`Array`, :py:class:`mpf`, tuple or scalar float
Points to evaluate the covariance between. If they are :py:class:`Array`,
:py:mod:`scipy` functions are used, otherwise :py:mod:`mpmath`
functions are used.
*args
Remaining arguments are the 2+num_dim hyperparameters as defined above.
"""
num_dim = len(args) - 2
nu = args[1]
if isinstance(Xi, scipy.ndarray):
if isinstance(Xi, scipy.matrix):
Xi = scipy.asarray(Xi, dtype=float)
Xj = scipy.asarray(Xj, dtype=float)
tau = scipy.asarray(Xi - Xj, dtype=float)
l_mat = scipy.tile(args[-num_dim:], (tau.shape[0], 1))
r2l2 = scipy.sum((tau / l_mat)**2, axis=1)
y = scipy.sqrt(2.0 * nu * r2l2)
k = 2.0**(1 - nu) / scipy.special.gamma(nu) * y**nu * scipy.special.kv(
nu, y)
k[r2l2 == 0] = 1
else:
try:
tau = [xi - xj for xi, xj in zip(Xi, Xj)]
except TypeError:
tau = Xi - Xj
try:
r2l2 = sum([(t / l)**2 for t, l in zip(tau, args[2:])])
except TypeError:
r2l2 = (tau / args[2])**2
y = mpmath.sqrt(2.0 * nu * r2l2)
k = 2.0**(1 - nu) / mpmath.gamma(nu) * y**nu * mpmath.besselk(nu, y)
k *= args[0]**2.0
return k
def matern_function(Xi, Xj, *args):
r"""Matern covariance function of arbitrary dimension, for use with :py:class:`ArbitraryKernel`.
The Matern kernel has the following hyperparameters, always referenced in
the order listed:
= ===== ====================================
0 sigma prefactor
1 nu order of kernel
2 l1 length scale for the first dimension
3 l2 ...and so on for all dimensions
= ===== ====================================
The kernel is defined as:
.. math::
k_M = \sigma^2 \frac{2^{1-\nu}}{\Gamma(\nu)}
\left (\sqrt{2\nu \sum_i\left (\frac{\tau_i^2}{l_i^2}\right )}\right )^\nu
K_\nu\left(\sqrt{2\nu \sum_i\left(\frac{\tau_i^2}{l_i^2}\right)}\right)
Parameters
----------
Xi, Xj : :py:class:`Array`, :py:class:`mpf`, tuple or scalar float
Points to evaluate the covariance between. If they are :py:class:`Array`,
:py:mod:`scipy` functions are used, otherwise :py:mod:`mpmath`
functions are used.
*args
Remaining arguments are the 2+num_dim hyperparameters as defined above.
"""
num_dim = len(args) - 2
nu = args[1]
if isinstance(Xi, scipy.ndarray):
if isinstance(Xi, scipy.matrix):
Xi = scipy.asarray(Xi, dtype=float)
Xj = scipy.asarray(Xj, dtype=float)
tau = scipy.asarray(Xi - Xj, dtype=float)
l_mat = scipy.tile(args[-num_dim:], (tau.shape[0], 1))
r2l2 = scipy.sum((tau / l_mat)**2, axis=1)
y = scipy.sqrt(2.0 * nu * r2l2)
k = 2.0**(1 - nu) / scipy.special.gamma(nu) * y**nu * scipy.special.kv(nu, y)
k[r2l2 == 0] = 1
else:
try:
tau = [xi - xj for xi, xj in zip(Xi, Xj)]
except TypeError:
tau = Xi - Xj
try:
r2l2 = sum([(t / l)**2 for t, l in zip(tau, args[2:])])
except TypeError:
r2l2 = (tau / args[2])**2
y = mpmath.sqrt(2.0 * nu * r2l2)
k = 2.0**(1 - nu) / mpmath.gamma(nu) * y**nu * mpmath.besselk(nu, y)
k *= args[0]**2.0
return k
def mean(p, b, loc=0, scale=1):
"""
Mean of the generalized inverse Gaussian distribution.
The mean is:
K_{p + 1}(b)
loc + scale --------------
K_p(b)
where K_n(x) is the modified Bessel function of the second kind
(implemented in mpmath as besselk(n, x)).
"""
p = mpmath.mpf(p)
b = mpmath.mpf(b)
loc = mpmath.mpf(loc)
scale = mpmath.mpf(scale)
return loc + scale * mpmath.besselk(p + 1, b) / mpmath.besselk(p, b)
def cramerVonMises(x, y):
try:
x = sorted(x)
y = sorted(y)
pool = x + y
ps, pr = _sortRank(pool)
rx = array([pr[ind] for ind in [ps.index(element) for element in x]])
ry = array([pr[ind] for ind in [ps.index(element) for element in y]])
n = len(x)
m = len(y)
i = array(range(1, n + 1))
j = array(range(1, m + 1))
u = n * sum(power((rx - i), 2)) + m * sum(power((ry - j), 2))
t = u / (n * m * (n + m)) - (4 * n * m - 1) / (6 * (n + m))
Tmu = 1 / 6 + 1 / (6 * (n + m))
Tvar = 1 / 45 * ((m + n + 1) / power(
(m + n), 2)) * (4 * m * n * (m + n) - 3 *
(power(m, 2) + power(n, 2)) - 2 * m * n) / (4 * m *
n)
t = (t - Tmu) / power(45 * Tvar, 0.5) + 1 / 6
if t < 0:
return -1
elif t <= 12:
a = 1 - mp.nsum(
lambda x:
(mp.gamma(x + 0.5) /
(mp.gamma(0.5) * mp.fac(x))) * mp.power(4 * x + 1, 0.5) * mp.
exp(-mp.power(4 * x + 1, 2) /
(16 * t)) * mp.besselk(0.25,
mp.power(4 * x + 1, 2) / (16 * t)),
[0, 100]) / (mp.pi * mp.sqrt(t))
return float(mp.nstr(a, 3))
else:
return 0
except Exception as e:
print e
return -1
def shot_noise_laplace_A_norm(X, g):
"""
Returns the normalized pdf of a shot noise process with laplace distributed amplitudes, A~Laplace(0,a)
Input:
X: Variable values, 1d numpy array.
g: shape parameter
Output:
F: The pdf
"""
F = np.zeros(len(X))
assert (g > 0)
g = mm.mpf(g)
for i in range(len(X)):
x = abs(X[i])
F[i] = (np.sqrt(g) * x / 2)**((g - 1) / 2) * mm.besselk((1 - g) / \
2, np.sqrt(g) * x) * np.sqrt(g / np.pi) / mm.gamma(g / 2)
return F
def spectrm1(sp, xx, kl2, L, rx, ry, s, np_, nr):
wn = np.zeros((np_,nr))
if sp.lower() == 'exponential': # exponential
for n in np.arange(1,np_+1):
wn[n-1, :] = n* kl2/(n**2 + kl2 *((rx-s)**2+ry**2))**1.5
if sp.lower() == 'gaussian': # gaussian
for n in np.arange(1,np_+1):
wn[n-1,:] = 0.5 * kl2/n* math.exp(-kl2*((rx-s)**2 + ry**2)/(4*n))
if sp.lower() == 'power_spec': # x-power
for n in np.arange(1,np_+1):
wn[n-1,:] = kl2/(2.**(xx*n-1)*math.gamma(xx*n))* ( ( (rx-s)**2.
+ ry**2)*L)**(xx*n-1)* besselk(-xx*n+1, L*((rx-s)**2 + ry**2))
return wn
def norm_sym_dsn_dist(X, g):
"""
Returns the normalized pdf of the derivative of a symmetric shot noise process, (td/2)*dS(t)/dt, lambda = 1/2.
Input:
X: The normalized variable X = (x-<x>)/x_rms, 1d numpy array
g: shape parameter
Output:
F: The pdf of X.
"""
F = np.zeros(len(X))
assert (g > 0)
g = mm.mpf(g)
for i in range(len(X)):
x = mm.mpf(np.abs(X[i]))
F[i] = mm.sqrt(2. * g / mm.pi) * 2.**(-g / 2.) * (mm.sqrt(g) * x)**(
(g - 1.) / 2.) * mm.besselk((1. - g) / 2.,
mm.sqrt(g) * x) / mm.gamma(g / 2.)
return F
def shot_noise_laplace_A(X, g, a):
"""
Returns the pdf of a shot noise process with laplace distributed amplitudes, A~Laplace(0,a)
Input:
X: Variable values, 1d numpy array.
g: shape parameter
a: scale parameter
Output:
F: The pdf
"""
F = np.zeros(len(X))
assert (g > 0)
assert (a > 0)
g = mm.mpf(g)
a = mm.mpf(a)
for i in range(len(X)):
x = abs(X[i])
F[i] = (x / (2 * a))**((g - 1) / 2) * mm.besselk(
(1 - g) / 2, x / a) / (a * np.sqrt(np.pi) * mm.gamma(g / 2))
return F
def psi(index, variable, b, a, beta):
if b == 0:
if beta < 0:
return math.pow(variable, 0.5) * besseli(
v(beta), sqrt(2 * index * q(a, beta, variable)))
else:
return math.pow(variable, 0.5) * besselk(
v(beta), sqrt(2 * index * q(a, beta, variable)))
else:
if beta < 0:
return math.pow(variable, beta + 0.5) * math.exp(
0.5 * eps(b, beta) * h(b, a, beta, variable)) * whitm(
k(b, beta, index), m(beta), h(b, a, beta, variable))
else:
return math.pow(variable, beta + 0.5) * math.exp(
0.5 * eps(b, beta) * h(b, a, beta, variable)) * whitw(
k(b, beta, index), m(beta), h(b, a, beta, variable))
def logpdf(x, p, b, loc=0, scale=1):
"""
Log of the PDF of the generalized inverse Gaussian distribution.
The PDF for x > loc is:
z**(p - 1) * exp(-b*(z + 1/z)/2))
---------------------------------
scale * K_p(b)
where s is the scale, z = (x - loc)/s, and K_p(b) is the modified Bessel
function of the second kind. For x <= loc, the PDF is zero.
"""
x = mpmath.mpf(x)
p = mpmath.mpf(p)
b = mpmath.mpf(b)
loc = mpmath.mpf(loc)
scale = mpmath.mpf(scale)
if x <= loc:
return -mpmath.mp.inf
z = (x - loc) / scale
return ((p - 1) * mpmath.log(z) - b * (z + 1 / z) / 2 -
mpmath.log(2 * mpmath.besselk(p, b)) - mpmath.log(scale))
def f94(x):
# synchrotron_2
x = mpmath.mpf(x)
coff = mpmath.mpf(2) / mpmath.mpf(3)
return x * mpmath.besselk(coff, x)
def f31(x):
# bessel_k2_scaled
x = mpmath.mpf(x)
y = mpmath.sqrt(mpmath.pi /
(2 * x)) * mpmath.besselk(2.5, x) * mpmath.exp(x)
return y
def f29(x):
# bessel_k0_scaled
x = mpmath.mpf(x)
y = mpmath.sqrt(mpmath.pi /
(2 * x)) * mpmath.besselk(0.5, x) * mpmath.exp(x)
return y
def f19(x):
# bessel_K1_scaled
scale = mpmath.exp(x)
return mpmath.besselk(1, x) * scale
def _compute_dk_dy(self, y, n):
r"""Evaluate the derivative of the outer form of the Matern kernel.
Uses the general Leibniz rule to compute the n-th derivative of:
.. math::
f(y) = \frac{2^{1-\nu}}{\Gamma(\nu)} y^\nu K_\nu(y)
Notice that this is very poorly-behaved at :math:`x=0`. There, the
value is approximated using :py:func:`mpmath.diff` with the `singular`
keyword. This is rather slow, so if you require a fixed value of `nu`
you may wish to consider implementing the appropriate kernel separately.
Parameters
----------
y : :py:class:`Array`, (`M`,)
`M` inputs to evaluate at.
n : non-negative scalar int.
Order of derivative to compute.
Returns
-------
dk_dy : :py:class:`Array`, (`M`,)
Specified derivative at specified locations.
"""
warnings.warn(
"The Matern kernel has not been verified for derivatives. Consider using MaternKernelArb."
)
dk_dy = scipy.zeros_like(y, dtype=float)
non_zero_idxs = (y != 0)
for k in xrange(0, n + 1):
dk_dy[non_zero_idxs] += (
scipy.special.binom(n, k) *
scipy.special.poch(1 - k + self.nu, k) *
(y[non_zero_idxs])**(-k + self.nu) *
scipy.special.kvp(self.nu, y[non_zero_idxs], n=n - k))
# Handle the cases at y=0.
# Compute the appropriate value using mpmath's arbitrary precision
# arithmetic. This is potentially slow, but seems to behave pretty
# well. In cases where the value should be infinite, very large
# (but still finite) floats are returned with the appropriate sign.
if n >= 2 * self.nu:
warnings.warn("n >= 2*nu can yield inaccurate results.",
RuntimeWarning)
# Use John Wright's expression for n < 2 * nu:
if n < 2.0 * self.nu:
if n % 2 == 1:
dk_dy[~non_zero_idxs] = 0.0
else:
m = n / 2.0
dk_dy[~non_zero_idxs] = ((-1.0)**m * 2.0**(self.nu - 1.0 - n) *
scipy.special.gamma(self.nu - m) *
scipy.misc.factorial(n) /
scipy.misc.factorial(m))
else:
# Fall back to mpmath to handle n >= 2 * nu:
core_expr = lambda x: x**self.nu * mpmath.besselk(self.nu, x)
deriv = mpmath.chop(
mpmath.diff(core_expr, 0, n=n, singular=True, direction=1))
dk_dy[~non_zero_idxs] = deriv
dk_dy *= 2.0**(1 - self.nu) / (scipy.special.gamma(self.nu))
return dk_dy
def test_besselk_int(self):
assert_mpmath_equal(sc.kn,
_exception_to_nan(lambda v, z: mpmath.besselk(v, z, **HYPERKW)),
[IntArg(), Arg()],
n=1000)
def test_besselk_complex(self):
assert_mpmath_equal(
lambda v, z: sc.kv(v.real, z),
_exception_to_nan(lambda v, z: mpmath.besselk(v, z, **HYPERKW)),
[Arg(-1e100, 1e100), ComplexArg()])
def sph_kn_bessel(n, z):
out = besselk(n + mpf(1)/2, z)*sqrt(pi/(2*z))
return out
def log_besselKvFA(nu, y):
val = np.log(kv(nu, y))
if np.isinf(val):
val = np.log(sympy.Float(mpmath.besselk(nu, y)))
return val
def test_besselk(self):
assert_mpmath_equal(
sc.kv,
_exception_to_nan(lambda v, z: mpmath.besselk(v, z, **HYPERKW)),
[Arg(-1e100, 1e100), Arg()],
n=1000)
def f16(x):
# bessel_K0
return mpmath.besselk(0, x)
def test_besselk_int(self):
assert_mpmath_equal(
sc.kn,
_exception_to_nan(lambda v, z: mpmath.besselk(v, z, **HYPERKW)),
[IntArg(), Arg()],
n=1000)
def f17(x):
# bessel_K1
return mpmath.besselk(1, x)
def test_besselk(self):
assert_mpmath_equal(sc.kv,
_exception_to_nan(lambda v, z: mpmath.besselk(v, z, **HYPERKW)),
[Arg(-1e100, 1e100), Arg()],
n=1000)
def f18(x):
# bessel_K0_scaled
scale = mpmath.exp(x)
return mpmath.besselk(0, x) * scale
def test_besselk_complex(self):
assert_mpmath_equal(lambda v, z: sc.kv(v.real, z),
_exception_to_nan(lambda v, z: mpmath.besselk(v, z, **HYPERKW)),
[Arg(-1e100, 1e100), ComplexArg()])
def _compute_dk_dy(self, y, n):
r"""Evaluate the derivative of the outer form of the Matern kernel.
Uses the general Leibniz rule to compute the n-th derivative of:
.. math::
f(y) = \frac{2^{1-\nu}}{\Gamma(\nu)} y^\nu K_\nu(y)
Notice that this is very poorly-behaved at :math:`x=0`. There, the
value is approximated using :py:func:`mpmath.diff` with the `singular`
keyword. This is rather slow, so if you require a fixed value of `nu`
you may wish to consider implementing the appropriate kernel separately.
Parameters
----------
y : :py:class:`Array`, (`M`,)
`M` inputs to evaluate at.
n : non-negative scalar int.
Order of derivative to compute.
Returns
-------
dk_dy : :py:class:`Array`, (`M`,)
Specified derivative at specified locations.
"""
warnings.warn("The Matern kernel has not been verified for derivatives. Consider using MaternKernelArb.")
dk_dy = scipy.zeros_like(y, dtype=float)
non_zero_idxs = (y != 0)
for k in xrange(0, n + 1):
dk_dy[non_zero_idxs] += (scipy.special.binom(n, k) *
scipy.special.poch(1 - k + self.nu, k) *
(y[non_zero_idxs])**(-k + self.nu) *
scipy.special.kvp(self.nu, y[non_zero_idxs], n=n-k))
# Handle the cases at y=0.
# Compute the appropriate value using mpmath's arbitrary precision
# arithmetic. This is potentially slow, but seems to behave pretty
# well. In cases where the value should be infinite, very large
# (but still finite) floats are returned with the appropriate sign.
if n >= 2 * self.nu:
warnings.warn("n >= 2*nu can yield inaccurate results.", RuntimeWarning)
# Use John Wright's expression for n < 2 * nu:
if n < 2.0 * self.nu:
if n % 2 == 1:
dk_dy[~non_zero_idxs] = 0.0
else:
m = n / 2.0
dk_dy[~non_zero_idxs] = (
(-1.0)**m *
2.0**(self.nu - 1.0 - n) *
scipy.special.gamma(self.nu - m) *
scipy.misc.factorial(n) / scipy.misc.factorial(m)
)
else:
# Fall back to mpmath to handle n >= 2 * nu:
core_expr = lambda x: x**self.nu * mpmath.besselk(self.nu, x)
deriv = mpmath.chop(mpmath.diff(core_expr, 0, n=n, singular=True, direction=1))
dk_dy[~non_zero_idxs] = deriv
dk_dy *= 2.0**(1 - self.nu) / (scipy.special.gamma(self.nu))
return dk_dy
from math import gamma # Been there a while
#has_scipy = False
if has_scipy:
from scipy.special import lambertw, ellipe, gammaincc, gamma # fluids
from scipy.special import i1, i0, k1, k0, iv # ht
from scipy.special import hyp2f1
if erf is None:
from scipy.special import erf
else:
import mpmath
# scipy is not available... fall back to mpmath as a Pure-Python implementation
from mpmath import lambertw # Same branches as scipy, supports .real
from mpmath import ellipe # seems the same so far
# Figured out this definition from test_precompute_gammainc.py in scipy
gammaincc = lambda a, x: mpmath.gammainc(a, a=x, regularized=True)
iv = mpmath.besseli
i1 = lambda x: mpmath.besseli(1, x)
i0 = lambda x: mpmath.besseli(0, x)
k1 = lambda x: mpmath.besselk(1, x)
k0 = lambda x: mpmath.besselk(0, x)
if erf is None:
from mpmath import erf
def sph_kn_bessel(n, z):
out = besselk(n + mpf(1) / 2, z) * sqrt(pi / (2 * z))
return out
