def concentration_analytic(x,t): N = 30 # upper limit for the sum c = 0 d1 = 2*np.sqrt(d*t) for i in range (0,N): c += erfc((1-x+2*i) / d1) - erfc((1+x+2*i) / d1) return c
def concentration_gradient(x,t):
exact = (lam/(D*beta)) * (
np.exp(-beta*(x))
- 0.5*np.exp(-beta*(x))*erfc((2.0*beta*D*t-(x))/np.sqrt(4.0*D*t))
- 0.5*np.exp(beta*(x))*erfc((2.0*beta*D*t+(x))/np.sqrt(4.0*D*t))
)
return exact
def expected(x,t):
porosity = 0.3
alphal = 0.2
v = 1.05e-3 / porosity
D = alphal * v
return 0.5 * erfc((x - v * t)/(2 *np.sqrt(D * t))) + np.sqrt(v * v * t/(np.pi * D)) * \
np.exp(- (x - v * t)**2/(4 * D * t)) - 0.5 * (1 + v * x / D + v * v * t / D) * np.exp(v * x / D) *\
erfc((x+v*t)/(2*np.sqrt(D*t)))
def analytic(x,t):
D=1
N = 30 # upper limit for the sum
c = 0
d = 2*np.sqrt(D*t)
for i in range (0,N):
c += erfc((1-x+2*i) / d) - erfc((1+x+2*i) / d)
return c
def Csq(x,sig,a):
X=x/(np.sqrt(2.0)*sig)
A=(x+a)/(np.sqrt(2.0)*sig)
Am=(x-a)/(np.sqrt(2.0)*sig)
A=x/(np.sqrt(2.0)*sig)+a
Am=x/(np.sqrt(2.0)*sig)-a
csq=erfc(Am)-erfc(A)
return (erfc(Am)-erfc(A))/csq[0]
def endf(c,x): #error function to fit end of tubes
if c[3]<0:
c[3]=0
if c[0]>0:
return c[0]*erfc((c[1]-x)/c[2]) + c[3]
if c[0]<0:
return -2*c[0] + c[0]*erfc((c[1]-x)/c[2]) + c[3]
def I1(Smax):
A1 = (Ni/2.)*((G/alpha/V)**0.5)
A2 = Smax
C1 = erfc(upart)
C2 = 0.5*((sgi/Smax)**2)
C3a = np.exp(9.*(log_sig**2)/2.)
C3b = erfc(upart + 3.*log_sig/np.sqrt(2.))
return A1*A2*(C1 - C2*C3a*C3b)
def _probit_inplace(x):
'''
Probit of every element of the given array, calculated in-place
'''
x *= -OneOverSqrtTwo
fns.erfc(x, out=x)
x /= 2
return x
def transient_temp_1D(z,t,G,vz,kappa):
# Calculate T separately for case where t = 0 to avoid divide-by-zero warnings
if t == 0:
T = G * z
else:
T = G * (z + vz * t) + (G / 2.0) * ((z - vz * t) * np.exp(-(vz * z) / kappa) *\
erfc((z - vz * t) / (2.0 * np.sqrt(kappa * t))) - (z + vz * t) * \
erfc((z + vz * t) / (2.0 * np.sqrt(kappa * t))))
return T
def pair_fn(i, j, d, delta):
energy = 0.0
#Dipole-Dipole (only term for this test)
fac1 = erfc(alpha*d) + 2.0*alpha*d/np.sqrt(np.pi)*np.exp(-alpha**2*d**2)
fac2 = 3.0*erfc(alpha*d) + 4.0*alpha**3*d**3/np.sqrt(np.pi)*np.exp(-alpha**2*d**2) \
+ 6.0*alpha*d/np.sqrt(np.pi)*np.exp(-alpha**2*d**2)
energy += np.dot( pair_pot.dipoles[i,:] , pair_pot.dipoles[j,:] )*fac1/d**3 - \
1.0*np.dot(pair_pot.dipoles[i,:],delta)*np.dot(delta,pair_pot.dipoles[j,:])*fac2/d**5
return energy
def radford_peak(x, x0, sigma, a, hstep, htail, tau, bg0, bg1):
bg_term = bg0 + x*bg1
step = a * hstep * erfc( (x - x0)/(sigma * np.sqrt(2)) )
le_tail = a * htail * erfc( (x - x0)/(sigma * np.sqrt(2)) + sigma/(tau*np.sqrt(2)) ) * np.exp( (x-x0)/tau ) / (2 * tau * np.exp( -(sigma / (np.sqrt(2)*tau))**2 ))
gauss_term = gaussian(x, x0, sigma, a, htail)
return gauss_term+bg_term + step + le_tail
def theoreticalQAM(M,EbN0dB):
'''
this function generates the theoretical error rate for a QAM modulated
signal where the number of lattice points is M and of the form
2^k=M
'''
from scipy import special
k = 1/((2/3.0)*(M-1))**.5
return [log(2*(1-1/M**(.5))*special.erfc(k*(10**(eb/10))**.5) -
(1-2/M**.5 + 1.0/M)*(special.erfc(k*(10**(eb/10))**.5))**2)/log(10) for eb in EbN0dB]
def gaussian_pulse_shaping_filter(BT,sps,T): T = 0.05 sps = 18 Ts = T/sps t = np.arange(-2*T,2*T,Ts) B = BT/T alpha = (2*np.pi*B)/np.sqrt(np.log(2)) gauss = (1/(2*T))*(0.5*special.erfc((alpha*(2*t-(T/2)))/np.sqrt(2)) - 0.5*special.erfc((alpha*(2*t+(T/2)))/np.sqrt(2))) K = (np.pi/2)/np.sum(gauss) y = K * gauss return y
def g2c_direct(t, c1, l1, l2, d):
sig = 0.3
c2 = -1
a = c1*0.5*np.exp(-l1*(t-d) +
0.5*(sig**2.0)*(l1**2.0))*(erfc(-(t-d)/np.sqrt(2.0*sig**2.0) + l1*(sig)/np.sqrt(2.0)))
b = c2*0.5*np.exp(l2*(t-d) +
0.5*(sig**2.0)*(l2**2.0))*(erfc((t-d)/np.sqrt(2.0*sig**2.0) + l2*(sig)/np.sqrt(2.0)))
y = 1 + a + b
return y
def F(self,z): #function describing how photo-z erros soften bin edges
if self.sigz0==0: #tophat
width=(self.zmaxnom-self.zminnom)/2.
sig=width*self.sharpness
result = 0.5*(np.tanh((z-self.zminnom)/sig)+np.tanh((self.zmaxnom-z)/sig))
else: #inside full z range, finite z uncertainty
sigzroot2 = self.sigz(z)*np.sqrt(2.)
lowerpart=.5*erfc((self.zminnom-z)/sigzroot2)
upperpart=-.5*erfc((self.zmaxnom-z)/sigzroot2)
result= lowerpart+upperpart
return result
def diffusion_analytic(t, h, V0, dy, viscosity):
y = np.arange(0, h + dy, dy)
eta1 = h / (2 * (t * viscosity)**0.5)
eta = y / (2 * (t * viscosity)**0.5)
sum1 = 0
sum2 = 0
for n in range(0, 1000):
sum1 = sum1 + erfc(2 * n * eta1 + eta)
sum2 = sum2 + erfc(2 * (n + 1) * eta1 - eta)
V_analytic = V0 * (sum1 - sum2)
return V_analytic
def erfc(x):
''' erfc(x), x is ADVar or scalar '''
if not isinstance(x, ADVar):
return special.erfc(x)
r = ADVar()
r.val = special.erfc(float(x))
tmp = - 2./math.sqrt(math.pi) * math.exp(-float(x)*float(x))
for i,dx in x.deriv:
r.deriv.append( (i, tmp*dx) )
return r
def gaussian_pulse_shaping_filter(self,BT,T,ts): #T = 0.05 sps = 18 ts = T/sps # t = np.arange(-T,T,ts) t = np.linspace(-T,T,num=27*3) B = BT/T alpha = (2*np.pi*B)/np.sqrt(np.log(2)) gauss = (1/(2*T))*(0.5*special.erfc((alpha*(2*t-(T/2)))/np.sqrt(2)) - 0.5*special.erfc((alpha*(2*t+(T/2)))/np.sqrt(2))) K = (np.pi/2)/np.sum(gauss) y = K * gauss return y
def hypermet(x, amplitude=1.0, center=0., sigma=1.0, step=0, tail=0, gamma=0.1):
"""
hypermet function to simulate XRF peaks and/or Compton Scatter Peak
Arguments
---------
x array of ordinate (energy) values
amplitude overall scale factor
center peak centroid
sigma Gaussian sigma
step step parameter for low-x erfc step [0]
tail amplitude of tail function [0]
gamma slope of tail function [0.1]
Notes
-----
The function is given by (with error checking for
small values of sigma, gamma and s2 = sqrt(2) and
s2pi = sqrt(2*pi)):
arg = (x - center)/sigma
gauss = (1.0/(s2pi*sigma)) * exp(-arg**2 /2)
sfunc = step * max(gauss) * erfc(arg/2.0) / 2.0
tfunc = tail * exp((x-center)/(gamma*sigma)) * erfc(arg/s2 + 1.0/gamma))
hypermet = amplitude * (gauss + sfunc + tfunc) / 2.0
This follows the definitions given in
ED-XRF SPECTRUM EVALUATION AND QUANTITATIVE ANALYSIS
USING MULTIVARIATE AND NONLINEAR TECHNIQUES
P. Van Espen, P. Lemberge
JCPDS-International Centre for Diffraction Data 2000,
Advances in X-ray Analysis,Vol.43 560
But is modified slightly to better preserve area with changing tail and gamma
"""
sigma = max(1.e-8, sigma)
gamma = max(1.e-8, gamma)
arg = (x - center)/sigma
arg[where(arg>100)] = 100.0
arg[where(arg<-100)] = -100.0
gscale = s2pi*sigma
gauss = (1.0/gscale) * exp(-arg**2 / 2.0)
sfunc = step * erfc(arg/2.0) / (2.0*gscale)
targ = (x-center)/(gamma*sigma)
targ[where(targ>100)] = 100.0
targ[where(targ<-100)] = -100.0
tfunc = exp(targ) * erfc(arg/2.0 + 1.0/gamma)
tfunc = tail*tfunc / (max(tfunc)*gscale)
return amplitude * (gauss + sfunc + tfunc) /2.0
def F(self,n,z): #function describing how photo-z erros soften bin edges
minedge = self.zedges[n]
maxedge = self.zedges[n+1]
if self.sigz0==0:
result= smoothedtophat(z,minedge,maxedge)#float((z>=minedge)*(z<maxedge)) #tophat fn
else: #inside full z range, finite z uncertainty
sigzroot2 = self.sigz(z)*np.sqrt(2.)
lowerpart=.5*erfc((minedge-z)/sigzroot2)
upperpart=-.5*erfc((maxedge-z)/sigzroot2)
result= lowerpart+upperpart
return result
def difffull(x1, x2, t, C0, C1, D):
"""
Simple equation for the diffusion into a semi-infinite slab, see Crank 1975
C0 is the concentration in the core
C1 is the concentration at the border
D is the diffusion coefficient in log10 unit, m^2.s^-1
x1 and x2 are the profil lengths from beginning and end respectively, in meters
t is the time in seconds
"""
x = (x2-x1)
Cx = (C1 - C0) * ( erfc(x / (2. * np.sqrt((10**D)*t))) + erfc((x) / (2. * np.sqrt((10**D)*t))))+ C0
return Cx
def hypermet(x, amplitude, center, sigma, step=0, tail=0, gamma=0.1):
"""
hypermet function to simulate XRF peaks and/or Compton Scatter Peak
Arguments
---------
x array of ordinate (energy) values
amplitude overall scale factor
center peak centroid
sigma Gaussian sigma
step step parameter for low-x erfc step [0]
tail amplitude of tail function [0]
gamma slope of tail function [0.1]
Notes
-----
The function is given by (with error checking for
small values of sigma, gamma and s2 = sqrt(2) and
s2pi = sqrt(2*pi)):
arg = (x - center)/sigma
gaus = exp(-arg**2/2.0) / (s2pi*sigma)
step = step * erfc(arg/s2) / (2*center)
tail = tail * exp(arg/gamma) * erfc(arg/s2 + 1.0/(s2*gamma))
tail = tail / (2*sigma*gamma*exp(-1.0/(2*gamma**2)))
hypermet = amplitude * (peak + step + tail)
This follows the definitions given in
ED-XRF SPECTRUM EVALUATION AND QUANTITATIVE ANALYSIS
USING MULTIVARIATE AND NONLINEAR TECHNIQUES
P. Van Espen, P. Lemberge
JCPDS-International Centre for Diffraction Data 2000,
Advances in X-ray Analysis,Vol.43 560
"""
sigma = max(1.e-8, sigma)
gamma = max(0.1, gamma)
arg = (x - center)/sigma
gaus = exp(-arg**2/2.0) / (s2pi*sigma)
step = step * erfc(arg/s2) / (2*center)
tail = tail * exp(arg/gamma) * erfc(arg/s2 + 1.0/(s2*gamma))
tail = tail / (2*sigma*gamma*exp(-1.0/(2*gamma**2)))
return amplitude * (gaus + step + tail)
def _lerfc(self,t):
''' numerically safe implementation of f(t) = log(1-erf(t)) = log(erfc(t))'''
from scipy.special import erfc
f = np.zeros_like(t)
tmin = 20; tmax = 25
ok = t<tmin # log(1-erf(t)) is safe to evaluate
bd = t>tmax # evaluate tight bound
nok = np.logical_not(ok)
interp = np.logical_and(nok,np.logical_not(bd)) # interpolate between both of them
f[nok] = np.log(old_div(2,np.sqrt(np.pi))) -t[nok]**2 -np.log(t[nok]+np.sqrt( t[nok]**2+old_div(4,np.pi) ))
lam = old_div(1,(1+np.exp( 12*(0.5-old_div((t[interp]-tmin),(tmax-tmin))) ))) # interp. weights
f[interp] = lam*f[interp] + (1-lam)*np.log(erfc( t[interp] ))
f[ok] += np.log(erfc( t[ok] )) # safe eval
return f
def stefan_parameter(λ, **kw):
""" Objective function to solve transcendental equation for the stefan
parameter λ """
Lf = kw.get("Lf", 1.0)
cpl = kw.get("cpl", 10.0)
cps = kw.get("cps", 1.0)
ρl = kw.get("rhol", 1.0)
ρs = kw.get("rhos", 1.0)
κl = kw.get("kapl", 1.0)
κs = kw.get("kaps", 1.0)
Tl = kw.get("Tl", 1.0)
Tm = kw.get("Tm", 0.0)
Ts = kw.get("Ts", -1.0)
αl = κl / (ρl * cpl)
αs = κs / (ρs * cps)
ν = sqrt(αl / αs)
Stl = cpl * (Tl-Tm) / Lf
Sts = cps * (Tm-Ts) / Lf
res = Stl / (exp(λ*λ) * erf(λ)) \
- Sts / (ν * exp(ν*ν*λ*λ) * erfc(ν*λ)) \
- sqrt(π)*λ
return res**2
def computeObservationPValue(self, siteObs):
"""
Compute a p-value on the observation of a kinetic event
"""
# p-value of detection -- FIXME needs much more thought here!
# p-value computation (slightly robustified Gaussian model)
# emf - rms fractional error of background model
# em - rms error of background model = um * emf
# um - predicted mean of unmodified ipd from model
# uo - (trimmed) observed mean ipd
# eo - (trimmed) standard error of observed mean (std / sqrt(coverage))
# Null model is ~N(um, em^2 + eo^2)
# Then compute standard gaussian p-value = erfc((uo-um) / sqrt(2 * (em^2 + eo^2))) / 2
# FIXME? -- right now we only detect the case where the ipd gets longer.
um = siteObs['modelPrediction']
# FIXME -- pipe through model error
em = 0.1 * um
# em = model.fractionalModelError * em
uo = siteObs['tMean']
eo = siteObs['tErr']
pvalue = erfc((uo - um) / sqrt(2 * (em ** 2 + eo ** 2))) / 2
return pvalue.item()
def I(t_):
y=exp(-b*(t-t_))/(a**2+8*D*(t-t_))*\
exp(-2*r**2/(a**2+8*D*(t-t_)))*\
exp(-alpha*z+alpha**2*D*(t-t_))*\
spec.erfc((2*D*alpha*(t-t_)-z)/(sqrt(4*D*(t-t_))))
return y
def _tTest(x, y, exclude=95):
"""Compute a one-sided Welsh t-statistic."""
with np.errstate(all="ignore"):
def cappedSlog(v):
q = np.percentile(v, exclude)
v2 = v.copy()
v2 = v2[~np.isnan(v2)]
v2[v2 > q] = q
v2[v2 <= 0] = 1. / (75 + 1)
return np.log(v2)
x1 = cappedSlog(x)
x2 = cappedSlog(y)
sx1 = np.var(x1) / len(x1)
sx2 = np.var(x2) / len(x2)
totalSE = np.sqrt(sx1 + sx2)
if totalSE == 0:
stat = 0
else:
stat = (np.mean(x1) - np.mean(x2)) / totalSE
#df = (sx1 + sx2)**2 / (sx1**2/(len(x1)-1) + sx2**2/(len(x2) - 1))
#pval = 1 - scidist.t.cdf(stat, df)
# Scipy's t distribution CDF implementaton has inadequate
# precision. We have switched to the normal distribution for
# better behaved p values.
pval = 0.5 * erfc(stat / sqrt(2))
return {'testStatistic': stat, 'pvalue': pval}
def __test_ks(self, x):
x = x[np.invert(np.isnan(x))]
n = len(x)
x = np.sort(x)
# Get cumulative sums
yCDF = np.array(np.arange(n)+1) / float(n)
# Remove duplicates; only need final one with total count
notdup = np.append(np.diff(x), 1) > 0
x_expcdf = x[notdup]
y_expcdf = np.append(0, yCDF[notdup])
# The theoretical CDF (theocdf) is assumed to be normal
# with unknown mean and sigma
zScores = (x_expcdf - np.mean(x))/np.std(x, ddof=1)
mu = 0
sigma = 1
theocdf = 0.5 * erfc(-(zScores-mu)/(np.sqrt(2)*sigma))
delta1 = y_expcdf[0:len(y_expcdf)-1] - theocdf # Vertical difference at jumps approaching from the LEFT.
delta2 = y_expcdf[1:len(y_expcdf)] - theocdf # Vertical difference at jumps approaching from the RIGHT.
deltacdf = np.abs(np.concatenate((delta1, delta2), axis=0))
return np.max(deltacdf)
def get_gamma(self, a=None):
"""
Return the gamma correlation matrix, i.e. phi(i) = gamma(i, j)*q(j)
"""
if a is not None:
self.update(a)
self.timer.start('get_gamma')
nat = len(self.a)
il, jl, dl, nl = get_neighbors(self.a, self.cutoff)
if il is None:
G = None
else:
G = np.zeros([nat, nat], dtype=float)
if self.cutoff is None:
for i, j, d in zip(il, jl, dl):
G[i, j] += 1.0/d
else:
for i, j, d in zip(il, jl, dl):
G[i, j] += 1.0*erfc(d/self.cutoff)/d
self.timer.stop('get_gamma')
return G
def reflection_factor_spherical_wave(impedance, angle, distance, wavenumber):
r"""
Spherical wave reflection factor :math:`Q`.
:param impedance: Normalized impedance :math:`Z`.
:param angle: Angle of incidence :math:`\theta`.
:param distance: Path length of the reflected ray :math:`r`.
:param wavenumber: Wavenumber :math:`k`.
The spherical wave relfection factor :math:`Q` is given by
.. math:: Q = R \left(1 - R \right) F
where :math:`R` is the plane wave reflection factor as calculated in :func:`reflection_factor_plane_wave` and :math:`F` is given by
.. math:: F = 1 - j \sqrt{ \pi} w e^{-w^2} \mathrm{erfc} \left( j w \right)
where :math:`w` is the numerical distance as calculated in :func:`numerical_distance`.
"""
w = numerical_distance(impedance, angle, distance, wavenumber)
F = 1.0 - 1j * np.sqrt(np.pi) * w * np.exp(-w**2.0) * erfc(1j * w)
plane_factor = reflection_factor_plane_wave(impedance, angle)
return plane_factor * (1.0 - plane_factor) * F
def brnoisecalc(lowP, highP):
return lowP.n * cmuxnoisecalc(highP, highP.α)
# https://tches.iacr.org/index.php/TCHES/article/view/8793
def iknoisecalc(lowP, highP, funcP):
return highP.n * (2**(-2 * (funcP.basebit * funcP.t) +
1)) + funcP.t * highP.n * (lowP.α**2)
def gbnoisecalc(lowP, highP, funcP):
return brnoisecalc(lowP, highP) + iknoisecalc(lowP, highP, funcP)
# FA parameter
d = 1600
m = 9
fanoise = d * cmuxnoisecalc(lvl1param, lvl1param.α)
print(fanoise)
print(erfc(1 / (2 * 2 * 2 * np.sqrt(2 * fanoise))))
print(
np.prod([(1 - mpfr(
erfc(1 /
(2 * 2 * 2 *
np.sqrt(2 *
(i + 1) * cmuxnoisecalc(lvl1param, lvl1param.α))))))**(
m * lvl1param.n) for i in range(d)]))
def rv_posterior_predictive_distribution_credible_regions(
model_like, df, tmin, tmax, Ntimes=200, Nsample=1000, **kwargs):
r"""
Compute credible regions of the posterior predictive distribution of the RV
signal using a radvel model, a dataframe of posterior samples, and a range
of times.
Arguemnts
---------
model_like : radvel.likelihood
Likelihood used for the forward-modeling of the RV signal.
df : pandas.DataFrame
Dataframe containing posterior sample of
parameters taken by model_like
tmin : float
Minimum time of time range for computing RV signal predictive posterior
tmax : float
Maximum time of time range for computing RV signal predictive posterior
Ntimes : int, optional
Number of times to sample between tmin and tmax.
Default value is 200.
Nsample : int, optional
Number of posterior samples to generate RV signals for.
Default value is 1000.
Other Arguments
---------------
levels : array-like, optional
Credible region levels to return.
Default values correspond to 1,2, and 3$\sigma$
full_sample : bool, optional
Return the underlying sample of RV signals in addition to credible regions.
Default is False.
Returns
-------
time : ndarray (N,)
Time values of posterior predictive distribution values.
lower : ndarray (M,N)
Lower values bounding the credible regions given by 'levels' of the poserior
predictive distribution.
Shape is (M,N) where M is the number of credible regions and N is the number
of times.
upper : ndarray (M,N)
Upper values bounding the credible regions given by 'levels' of the poserior
predictive distribution.
Shape is (M,N) where M is the number of credible regions and N is the number
of times.
normalized_residual_info : ndarray (3,Nobs)
Information on the normalized residuals of the fit to the observations.
Contains the median and ±1sigma values of the normalized residuals for
each observation point.
sample : ndarray, (N,Nsample), optional
If 'full_sample' is True, contains the full sample of the posterior predictive
distribution used to compute the credible regoins.
"""
levels = kwargs.pop('levels',
np.array([erf(n / np.sqrt(2)) for n in range(1, 4)]))
full_sample = kwargs.pop('full_sample', False)
rv_out = np.zeros((Nsample, Ntimes))
Nobs = len(model_like.x)
normalized_resids = np.zeros((Nobs, Nsample))
times = np.linspace(tmin, tmax, Ntimes)
for k in range(Nsample):
rv_out[k] = np.infty
# Avoid unstable posterior points
while np.any(np.isinf(rv_out[k])):
i = np.random.randint(0, len(df) - 1)
pars = df.iloc[i]
vpars = pars[model_like.list_vary_params()]
model_like.set_vary_params(vpars)
rv_out[k] = model_like.model(times)
ebs = model_like.errorbars()
normalized_resids[:, k] = model_like.residuals() / ebs
lo, hi = [], []
for lvl in levels:
hi.append(np.quantile(rv_out.T, 0.5 + lvl / 2, axis=1))
lo.append(np.quantile(rv_out.T, 0.5 - lvl / 2, axis=1))
normalized_residual_quantiles = np.quantile(normalized_resids,
(0.5, erf(1), erfc(1)),
axis=1)
if full_sample:
return times, lo, hi, normalized_residual_quantiles, rv_out
return times, lo, hi, normalized_residual_quantiles
def emg(mu, amp, sigma, tau):
return amp \
* np.exp(0.5 * (2.0 * mu + sigma**2.0 / tau - 2.0 * x) / tau) \
* erfc((mu + sigma**2.0 / tau - x) / (2.0**0.5 * sigma))
def acetylAnalytical(x, telapsed):
z = x / np.sqrt(4 * D0 * telapsed)
acetyl = 1 - np.exp(-p0 * arate * telapsed * (
(1 + 2 *
(z**2)) * scis.erfc(z) - 2 * z * np.exp(-(z**2)) / np.sqrt(np.pi)))
return acetyl
def df(x):
return _k * _special.erfc(x) - _np.exp(-x**2)
def mback(energy,
mu,
group=None,
order=3,
z=None,
edge='K',
e0=None,
emin=None,
emax=None,
whiteline=None,
leexiang=False,
tables='chantler',
fit_erfc=False,
return_f1=False,
_larch=None):
"""
Match mu(E) data for tabulated f''(E) using the MBACK algorithm and,
optionally, the Lee & Xiang extension
Arguments:
energy, mu: arrays of energy and mu(E)
order: order of polynomial [3]
group: output group (and input group for e0)
z: Z number of absorber
edge: absorption edge (K, L3)
e0: edge energy
emin: beginning energy for fit
emax: ending energy for fit
whiteline: exclusion zone around white lines
leexiang: flag to use the Lee & Xiang extension
tables: 'chantler' (default) or 'cl'
fit_erfc: True to float parameters of error function
return_f1: True to put the f1 array in the group
Returns:
group.f2: tabulated f2(E)
group.f1: tabulated f1(E) (if return_f1 is True)
group.fpp: matched data
group.mback_params: Group of parameters for the minimization
References:
* MBACK (Weng, Waldo, Penner-Hahn): http://dx.doi.org/10.1086/303711
* Lee and Xiang: http://dx.doi.org/10.1088/0004-637X/702/2/970
* Cromer-Liberman: http://dx.doi.org/10.1063/1.1674266
* Chantler: http://dx.doi.org/10.1063/1.555974
"""
order = int(order)
if order < 1: order = 1 # set order of polynomial
if order > MAXORDER: order = MAXORDER
### implement the First Argument Group convention
energy, mu, group = parse_group_args(energy,
members=('energy', 'mu'),
defaults=(mu, ),
group=group,
fcn_name='mback')
if len(energy.shape) > 1:
energy = energy.squeeze()
if len(mu.shape) > 1:
mu = mu.squeeze()
group = set_xafsGroup(group, _larch=_larch)
if e0 is None: # need to run find_e0:
e0 = xray_edge(z, edge, _larch=_larch)[0]
if e0 is None:
e0 = group.e0
if e0 is None:
find_e0(energy, mu, group=group)
### theta is an array used to exclude the regions <emin, >emax, and
### around white lines, theta=0.0 in excluded regions, theta=1.0 elsewhere
(i1, i2) = (0, len(energy) - 1)
if emin is not None: i1 = index_of(energy, emin)
if emax is not None: i2 = index_of(energy, emax)
theta = np.ones(len(energy)) # default: 1 throughout
theta[0:i1] = 0
theta[i2:-1] = 0
if whiteline:
pre = 1.0 * (energy < e0)
post = 1.0 * (energy > e0 + float(whiteline))
theta = theta * (pre + post)
if edge.lower().startswith('l'):
l2 = xray_edge(z, 'L2', _larch=_larch)[0]
l2_pre = 1.0 * (energy < l2)
l2_post = 1.0 * (energy > l2 + float(whiteline))
theta = theta * (l2_pre + l2_post)
## this is used to weight the pre- and post-edge differently as
## defined in the MBACK paper
weight1 = 1 * (energy < e0)
weight2 = 1 * (energy > e0)
weight = np.sqrt(sum(weight1)) * weight1 + np.sqrt(sum(weight2)) * weight2
## get the f'' function from CL or Chantler
if tables.lower() == 'chantler':
f1 = f1_chantler(z, energy, _larch=_larch)
f2 = f2_chantler(z, energy, _larch=_larch)
else:
(f1, f2) = f1f2(z, energy, edge=edge, _larch=_larch)
group.f2 = f2
if return_f1: group.f1 = f1
n = edge
if edge.lower().startswith('l'): n = 'L'
params = Group(
s=Parameter(1, vary=True, _larch=_larch), # scale of data
xi=Parameter(50, vary=fit_erfc, min=0, _larch=_larch), # width of erfc
em=Parameter(xray_line(z, n, _larch=_larch)[0],
vary=False,
_larch=_larch), # erfc centroid
e0=Parameter(e0, vary=False, _larch=_larch), # abs. edge energy
## various arrays need by the objective function
en=energy,
mu=mu,
f2=group.f2,
weight=weight,
theta=theta,
leexiang=leexiang,
_larch=_larch)
if fit_erfc:
params.a = Parameter(1, vary=True, _larch=_larch) # amplitude of erfc
else:
params.a = Parameter(0, vary=False, _larch=_larch) # amplitude of erfc
for i in range(order): # polynomial coefficients
setattr(params, 'c%d' % i, Parameter(0, vary=True, _larch=_larch))
fit = Minimizer(match_f2, params, _larch=_larch, toler=1.e-5)
fit.leastsq()
eoff = energy - params.e0.value
normalization_function = params.a.value * erfc(
(energy - params.em.value) / params.xi.value) + params.c0.value
for i in range(MAXORDER):
j = i + 1
attr = 'c%d' % j
if hasattr(params, attr):
normalization_function = normalization_function + getattr(
getattr(params, attr), 'value') * eoff**j
group.fpp = params.s * mu - normalization_function
group.mback_params = params
def approxLognormal(N,
mu=0.0,
sigma=1.0,
tail_N=0,
tail_bound=[0.02, 0.98],
tail_order=np.e):
'''
Construct a discrete approximation to a lognormal distribution with underlying
normal distribution N(mu,sigma). Makes an equiprobable distribution by
default, but user can optionally request augmented tails with exponentially
sized point masses. This can improve solution accuracy in some models.
Parameters
----------
N: int
Number of discrete points in the "main part" of the approximation.
mu: float
Mean of underlying normal distribution.
sigma: float
Standard deviation of underlying normal distribution.
tail_N: int
Number of points in each "tail part" of the approximation; 0 = no tail.
tail_bound: [float]
CDF boundaries of the tails vs main portion; tail_bound[0] is the lower
tail bound, tail_bound[1] is the upper tail bound. Inoperative when
tail_N = 0. Can make "one tailed" approximations with 0.0 or 1.0.
tail_order: float
Factor by which consecutive point masses in a "tail part" differ in
probability. Should be >= 1 for sensible spacing.
Returns
-------
pmf: np.ndarray
Probabilities for discrete probability mass function.
X: np.ndarray
Discrete values in probability mass function.
Written by Luca Gerotto
Based on Matab function "setup_workspace.m," from Chris Carroll's
[Solution Methods for Microeconomic Dynamic Optimization Problems]
(http://www.econ2.jhu.edu/people/ccarroll/solvingmicrodsops/) toolkit.
Latest update: 11 February 2017 by Matthew N. White
'''
# Find the CDF boundaries of each segment
if sigma > 0.0:
if tail_N > 0:
lo_cut = tail_bound[0]
hi_cut = tail_bound[1]
else:
lo_cut = 0.0
hi_cut = 1.0
inner_size = hi_cut - lo_cut
inner_CDF_vals = [
lo_cut + x * N**(-1.0) * inner_size for x in range(1, N)
]
if inner_size < 1.0:
scale = 1.0 / tail_order
mag = (1.0 - scale**tail_N) / (1.0 - scale)
lower_CDF_vals = [0.0]
if lo_cut > 0.0:
for x in range(tail_N - 1, -1, -1):
lower_CDF_vals.append(lower_CDF_vals[-1] +
lo_cut * scale**x / mag)
upper_CDF_vals = [hi_cut]
if hi_cut < 1.0:
for x in range(tail_N):
upper_CDF_vals.append(upper_CDF_vals[-1] +
(1.0 - hi_cut) * scale**x / mag)
CDF_vals = lower_CDF_vals + inner_CDF_vals + upper_CDF_vals
temp_cutoffs = list(
stats.lognorm.ppf(CDF_vals[1:-1], s=sigma, loc=0,
scale=np.exp(mu)))
cutoffs = [0] + temp_cutoffs + [np.inf]
CDF_vals = np.array(CDF_vals)
# Construct the discrete approximation by finding the average value within each segment
K = CDF_vals.size - 1 # number of points in approximation
pmf = CDF_vals[1:(K + 1)] - CDF_vals[0:K]
X = np.zeros(K)
for i in range(K):
zBot = cutoffs[i]
zTop = cutoffs[i + 1]
tempBot = (mu + sigma**2 - np.log(zBot)) / (np.sqrt(2) * sigma)
tempTop = (mu + sigma**2 - np.log(zTop)) / (np.sqrt(2) * sigma)
if tempBot <= 4:
X[i] = -0.5 * np.exp(mu + (sigma**2) * 0.5) * (
erf(tempTop) - erf(tempBot)) / pmf[i]
else:
X[i] = -0.5 * np.exp(mu + (sigma**2) * 0.5) * (
erfc(tempBot) - erfc(tempTop)) / pmf[i]
else:
pmf = np.ones(N) / N
X = np.exp(mu) * np.ones(N)
return [pmf, X]
l = np.exp(-alp * xauxf**pf)
L = np.diag(l)
elif OP == 2:
l = (np.pi * xauxf)**(-1) * np.sin(np.pi * xauxf)
l[0] = 1.0
L = np.diag(l)
elif OP == 3:
l = (1 + np.cos(np.pi * xauxf)) / 2
L = np.diag(l)
elif OP == 4:
l = (1 + np.cos(np.pi * xauxf)) / 2
l = l**4 * (35 - 84 * l + 70 * l**2 - 20 * l**3)
L = np.diag(l)
elif OP == 5:
#l = 0.5*scs.erfc(2*pf**0.5*(np.abs(xauxf) - 0.5)*np.sqrt((-np.log(1-4*(xauxf-0.5)**2))/(4*(xauxf-0.5)**2)))
l = 0.5 * scs.erfc(2 * pf**0.5 * (np.abs(xauxf) - 0.5))
L = np.diag(l)
elif OP == 6:
l = 0.5 * scs.erfc(2 * pf**0.5 * (np.abs(xauxf) - 0.5) * np.sqrt(
(-np.log(1 - 4 * (xauxf - 0.5)**2)) / (4 * (xauxf - 0.5)**2)))
if (np.mod(i, 2) == 0):
l[int(i / 2)] = 0.5
L = np.diag(l)
#what if L is the identity
# L = np.identity(j)
# L[i,i] = 0
# L[i-1,i-1] = 0
#M = inv[B.T@W@B]@B.T@W, lets proof the properties of B@W in all the possibilities
WP1 = W1 @ P1
def logpdf(x, mu, sigma):
xmin = np.min(x)
F = lambda x: (sp.erfc((np.log(x) - mu) / (np.sqrt(2) * sigma))) / 2
g = lambda x: F(x) - F(x + 1)
h = -np.log(F(xmin)) + np.log(g(x))
return h
def pVIC(x, A, alpha, beta, R, gamma, sigma, k, xp):
"""
The pseudo-Voigt-Ikeda-Carpenter function is the convolution of a pseudo-Voigt
profile with the Ikeda-Carpenter function.
See Nuclear Instruments and Methods in Physics Research, A239 (1985) 536-544
and the "time of flight" computation behind the FullProf software, explained
at http://www.ccp14.ac.uk/ccp/web-mirrors/plotr/Tutorials&Documents/TOF_FullProf.pdf
Parameters
----------
x : Numerical value (float)
Independent 'pVIC' function variable, which is the position on the x axis.
A : Float
Amplitude of the pseudo-Voigt-Ikeda-Carpenter function.
alpha : Float
IC (Ikeda-Carpenter) fast decay parameter.
beta : Float
IC slow decay parameter.
R : Float
IC weight ratio between fast and slow decays.
gamma : Float
pV (pseudo-Voigt) Lorentzian width.
sigma : Float
pV Gaussian width.
k : Float
IC "approximation" parameter (not sure what it does).
Its value is set to 0.05 by default.
xp : Float
Position of peak (not max).
Returns
-------
TYPE: float
Value of the pseudo-Voigt-Ikeda-Carpenter function at position x in reciprocal space.
"""
xr = x - xp # position of current datapoint in reciprocal space, relative to peak position
def pseudoVoigtFWHM(gamma, sigma):
fL = 2 * gamma
fG = 2 * sigma * np.sqrt(2 * np.log(2))
Gamma = (fG**5 + 2.69269*fG**4*fL + 2.42843*fG**3*fL**2 \
+ 4.47163*fG**2*fL**3 + 0.07842*fG*fL**4 + fL**5)**(1/5)
return Gamma
def pseudoVoigtEta(gamma, sigma):
fL = 2 * gamma
f = pseudoVoigtFWHM(gamma, sigma)
fLbyf = fL / f # note: because of this line, one cannot have both gamma and sigma set to zero
eta = 1.36603 * fLbyf - 0.47719 * fLbyf**2 + 0.11116 * fLbyf**3
return eta
Gamma = pseudoVoigtFWHM(gamma, sigma)
eta = pseudoVoigtEta(gamma, sigma)
sigmaSq = Gamma**2 / (8 * np.log(2))
gWidth = np.sqrt(2 * sigmaSq)
am = alpha * (1 - k)
ap = alpha * (1 + k)
x = am - beta
y = alpha - beta
z = ap - beta
zs = -alpha * xr + 1j * alpha * Gamma / 2
zu = (1 - k) * zs
zv = (1 + k) * zs
zr = -beta * xr + 1j * beta * Gamma / 2
u = am * (am * sigmaSq - 2 * xr) / 2
v = ap * (ap * sigmaSq - 2 * xr) / 2
s = alpha * (alpha * sigmaSq - 2 * xr) / 2
r = beta * (beta * sigmaSq - 2 * xr) / 2
n = (1 / 4) * alpha * (1 - k**2) / k**2
nu = 1 - R * am / x
nv = 1 - R * ap / z
ns = -2 * (1 - R * alpha / y)
nr = 2 * R * alpha**2 * beta * k**2 / (x * y * z)
yu = (am * sigmaSq - xr) / gWidth
yv = (ap * sigmaSq - xr) / gWidth
ys = (alpha * sigmaSq - xr) / gWidth
yr = (beta * sigmaSq - xr) / gWidth
val = A*n*((1 - eta)*(nu*np.exp(u)*erfc(yu) + nv*np.exp(v)*erfc(yv)\
+ ns*np.exp(s)*erfc(ys) + nr*np.exp(r)*erfc(yr))\
- eta*2/np.pi*(np.imag(nu*exp1(zu)*np.exp(zu)) + np.imag(nv*exp1(zv)*np.exp(zv))\
+ np.imag(ns*exp1(zs)*np.exp(zs)) + np.imag(nr*exp1(zr)*np.exp(zr))))
return val
plt.title("Constellation diagram, with SNR = 6")
plt.scatter(real1, Y1, c='b')
plt.axis([-10, 10, -20, 20])
plt.show()
plt.grid(True)
plt.title("Constellation diagram, with SNR = 6")
plt.scatter(real2, Y2, color='g')
plt.axis([-10, 10, -20, 20])
plt.show()
# erwthhma 1e
Yax = [0] * 16
Xax = [0] * 16
SNR = np.linspace(0, 16, 16 * 5000)
Ber_th = [0] * 80000
for i in range(80000):
Ber_th[i] = 0.5 * special.erfc(math.sqrt(10**(i * 0.0002 / 10)))
for snr in range(16):
Xax[snr] = snr
n0 = (A**2) * Tb / 10**(snr / 10)
correct = 0
for i in range(100000):
bit = random.randint(0, 1)
if bit == 1:
bit_tr = A
else:
bit_tr = -A
x1 = np.random.normal(0, math.sqrt(n0 / 2))
x2 = np.random.normal(0, math.sqrt(n0 / 2))
bit_noise = bit_tr * math.sqrt(Tb) + x1
if bit_tr * bit_noise >= 0: # αν το αρχικό σήμα και το σήμα μετά τον θόρυβο έχουν ενέργειες με ίδιο πρόσημο
correct += 1 # τότε ο θόρυβος δεν επηρεάζει την αποδιαμόρφωση
def EfficInROIOpt(b):
roiedge = ROIOpt(b) / 2.
return 1. - spec.erfc(2.35 * roiedge / (numpy.sqrt(2.)))
def gauss(x, s, mu):
result = 1 - special.erfc((x - mu) / s) / 2
return result
def P_model_erf(T, P0, B, T0, sigma):
return B*(T-T0)*0.5*erfc((T-T0)/(np.sqrt(2)*sigma)) \
- B*sigma/(np.sqrt(2*np.pi))*np.exp(-(T-T0)**2/(2*sigma**2))+P0
def W(r,Ion=0.1): k = np.sqrt(Ion) / 3.08 #Ion=0.1 is default x = k * r / np.sqrt(6) return 332.286 * np.sqrt(6 / np.pi) * (1 - np.sqrt(np.pi) * x * np.exp(x ** 2) * erfc(x)) / (e * r)
def approx(self, N, tail_N=0, tail_bound=[0.02, 0.98], tail_order=np.e):
'''
Construct a discrete approximation to a lognormal distribution with underlying
normal distribution N(mu,sigma). Makes an equiprobable distribution by
default, but user can optionally request augmented tails with exponentially
sized point masses. This can improve solution accuracy in some models.
Parameters
----------
N: int
Number of discrete points in the "main part" of the approximation.
tail_N: int
Number of points in each "tail part" of the approximation; 0 = no tail.
tail_bound: [float]
CDF boundaries of the tails vs main portion; tail_bound[0] is the lower
tail bound, tail_bound[1] is the upper tail bound. Inoperative when
tail_N = 0. Can make "one tailed" approximations with 0.0 or 1.0.
tail_order: float
Factor by which consecutive point masses in a "tail part" differ in
probability. Should be >= 1 for sensible spacing.
Returns
-------
d : DiscreteDistribution
Probability associated with each point in array of discrete
points for discrete probability mass function.
'''
# Find the CDF boundaries of each segment
if self.sigma > 0.0:
if tail_N > 0:
lo_cut = tail_bound[0]
hi_cut = tail_bound[1]
else:
lo_cut = 0.0
hi_cut = 1.0
inner_size = hi_cut - lo_cut
inner_CDF_vals = [
lo_cut + x * N**(-1.0) * inner_size for x in range(1, N)
]
if inner_size < 1.0:
scale = 1.0 / tail_order
mag = (1.0 - scale**tail_N) / (1.0 - scale)
lower_CDF_vals = [0.0]
if lo_cut > 0.0:
for x in range(tail_N - 1, -1, -1):
lower_CDF_vals.append(lower_CDF_vals[-1] +
lo_cut * scale**x / mag)
upper_CDF_vals = [hi_cut]
if hi_cut < 1.0:
for x in range(tail_N):
upper_CDF_vals.append(upper_CDF_vals[-1] +
(1.0 - hi_cut) * scale**x / mag)
CDF_vals = lower_CDF_vals + inner_CDF_vals + upper_CDF_vals
temp_cutoffs = list(
stats.lognorm.ppf(CDF_vals[1:-1],
s=self.sigma,
loc=0,
scale=np.exp(self.mu)))
cutoffs = [0] + temp_cutoffs + [np.inf]
CDF_vals = np.array(CDF_vals)
# Construct the discrete approximation by finding the average value within each segment.
# This codeblock ignores warnings because it throws a "divide by zero encountered in log"
# warning due to computing erf(infty) at the tail boundary. This is irrelevant and
# apparently freaks new users out.
with warnings.catch_warnings():
warnings.simplefilter("ignore")
K = CDF_vals.size - 1 # number of points in approximation
pmf = CDF_vals[1:(K + 1)] - CDF_vals[0:K]
X = np.zeros(K)
for i in range(K):
zBot = cutoffs[i]
zTop = cutoffs[i + 1]
tempBot = (self.mu + self.sigma**2 -
np.log(zBot)) / (np.sqrt(2) * self.sigma)
tempTop = (self.mu + self.sigma**2 -
np.log(zTop)) / (np.sqrt(2) * self.sigma)
if tempBot <= 4:
X[i] = -0.5 * np.exp(self.mu +
(self.sigma**2) * 0.5) * (
erf(tempTop) -
erf(tempBot)) / pmf[i]
else:
X[i] = -0.5 * np.exp(self.mu +
(self.sigma**2) * 0.5) * (
erfc(tempBot) -
erfc(tempTop)) / pmf[i]
else:
pmf = np.ones(N) / N
X = np.exp(self.mu) * np.ones(N)
return DiscreteDistribution(pmf, X)
print(("Potential softened below", H, "kpc and truncated above", r_s, "kpc"))
####################################################################
r = np.logspace(np.log10(e_plummer) - 1.2, np.log10(box_size) + 0.2, 10000)
# Newtonian gravity
f_newton = 1 / r**2
# Simulated gravity
u = r / H
u = u[u <= 1]
W_swift = 21.0 * u**6 - 90.0 * u**5 + 140.0 * u**4 - 84.0 * u**3 + 14.0 * u
f_swift = f_newton * (special.erfc(0.5 * r / r_s) +
(1.0 / math.sqrt(math.pi)) *
(r / r_s) * np.exp(-0.25 * (r / r_s)**2))
f_swift[r <= H] = W_swift / H**2
f_swift[r > r_cut] = 0
W_gadget = u * (21.333333 - 48 * u + 38.4 * u**2 - 10.6666667 * u**3 -
0.06666667 * u**-3)
W_gadget[u < 0.5] = u[u < 0.5] * (10.666667 + u[u < 0.5]**2 *
(32.0 * u[u < 0.5] - 38.4))
f_gadget = f_newton * (special.erfc(0.5 * r / r_s) +
(1.0 / math.sqrt(math.pi)) *
(r / r_s) * np.exp(-0.25 * (r / r_s)**2))
f_gadget[r <= H] = W_gadget / H**2
f_gadget[r > r_cut] = 0
def gaussianCDF(x, amplitude, mu, sigma):
"""
CDF of gaussian is P(X<=x) = .5 erfc((mu-x)/(sqrt(2)sig))
"""
return 0.5 * amplitude * erfc((mu - x) / (np.sqrt(2) * sigma))
def err_psk_8(EsN0, coh):
q = np.sqrt(2. * EsN0)
ps = sp.erfc(q / np.sqrt(2.) * np.sin(np.pi / 8.))
pb = 1. / 3. * ps # Код Грея
return ps, pb
def densityAnalytical(x, telapsed):
z = x / np.sqrt(4 * D0 * (telapsed))
return scis.erfc(z)
def approx(self, N, tail_N=0, tail_bound=None, tail_order=np.e):
'''
Construct a discrete approximation to a lognormal distribution with underlying
normal distribution N(mu,sigma). Makes an equiprobable distribution by
default, but user can optionally request augmented tails with exponentially
sized point masses. This can improve solution accuracy in some models.
Parameters
----------
N: int
Number of discrete points in the "main part" of the approximation.
tail_N: int
Number of points in each "tail part" of the approximation; 0 = no tail.
tail_bound: [float]
CDF boundaries of the tails vs main portion; tail_bound[0] is the lower
tail bound, tail_bound[1] is the upper tail bound. Inoperative when
tail_N = 0. Can make "one tailed" approximations with 0.0 or 1.0.
tail_order: float
Factor by which consecutive point masses in a "tail part" differ in
probability. Should be >= 1 for sensible spacing.
Returns
-------
d : DiscreteDistribution
Probability associated with each point in array of discrete
points for discrete probability mass function.
'''
tail_bound = tail_bound if tail_bound is not None else [0.02, 0.98]
# Find the CDF boundaries of each segment
if self.sigma > 0.0:
if tail_N > 0:
lo_cut = tail_bound[0]
hi_cut = tail_bound[1]
else:
lo_cut = 0.0
hi_cut = 1.0
inner_size = hi_cut - lo_cut
inner_CDF_vals = [
lo_cut + x * N**(-1.0) * inner_size for x in range(1, N)
]
if inner_size < 1.0:
scale = 1.0 / tail_order
mag = (1.0 - scale**tail_N) / (1.0 - scale)
lower_CDF_vals = [0.0]
if lo_cut > 0.0:
for x in range(tail_N - 1, -1, -1):
lower_CDF_vals.append(lower_CDF_vals[-1] +
lo_cut * scale**x / mag)
upper_CDF_vals = [hi_cut]
if hi_cut < 1.0:
for x in range(tail_N):
upper_CDF_vals.append(upper_CDF_vals[-1] +
(1.0 - hi_cut) * scale**x / mag)
CDF_vals = lower_CDF_vals + inner_CDF_vals + upper_CDF_vals
temp_cutoffs = list(
stats.lognorm.ppf(CDF_vals[1:-1],
s=self.sigma,
loc=0,
scale=np.exp(self.mu)))
cutoffs = [0] + temp_cutoffs + [np.inf]
CDF_vals = np.array(CDF_vals)
K = CDF_vals.size - 1 # number of points in approximation
pmf = CDF_vals[1:(K + 1)] - CDF_vals[0:K]
X = np.zeros(K)
for i in range(K):
zBot = cutoffs[i]
zTop = cutoffs[i + 1]
# Manual check to avoid the RuntimeWarning generated by "divide by zero"
# with np.log(zBot).
if zBot == 0:
tempBot = np.inf
else:
tempBot = (self.mu + self.sigma**2 -
np.log(zBot)) / (np.sqrt(2) * self.sigma)
tempTop = (self.mu + self.sigma**2 -
np.log(zTop)) / (np.sqrt(2) * self.sigma)
if tempBot <= 4:
X[i] = -0.5 * np.exp(self.mu + (self.sigma**2) * 0.5) * (
erf(tempTop) - erf(tempBot)) / pmf[i]
else:
X[i] = -0.5 * np.exp(self.mu + (self.sigma**2) * 0.5) * (
erfc(tempBot) - erfc(tempTop)) / pmf[i]
else:
pmf = np.ones(N) / N
X = np.exp(self.mu) * np.ones(N)
return DiscreteDistribution(pmf,
X,
seed=self.RNG.randint(0,
2**31 - 1,
dtype='int32'))
def build_fcoll_tab(self):
"""
Build a lookup table for the halo mass function / collapsed fraction.
Can be run in parallel.
"""
self.logMmin_tab = self.pf['hmf_logMmin']
self.logMmax_tab = self.pf['hmf_logMmax']
self.zmin = self.pf['hmf_zmin']
self.zmax = self.pf['hmf_zmax']
self.dlogM = self.pf['hmf_dlogM']
self.dz = self.pf['hmf_dz']
self.Nz = int((self.zmax - self.zmin) / self.dz + 1)
self.z = np.linspace(self.zmin, self.zmax, self.Nz)
self.Nm = np.logspace(self.logMmin_tab, self.logMmax_tab,
self.dlogM).size
if rank == 0:
print "\nComputing %s mass function..." % self.hmf_func
# Masses in hmf are in units of Msun * h
self.M = self.MF.M / self.cosm.h70
self.logM = np.log10(self.M)
self.lnM = np.log(self.M)
self.Nm = self.M.size
self.dndm = np.zeros([self.Nz, self.Nm])
self.mgtm = np.zeros_like(self.dndm)
self.ngtm = np.zeros_like(self.dndm)
fcoll_tab = np.zeros_like(self.dndm)
pb = ProgressBar(len(self.z), 'fcoll')
pb.start()
for i, z in enumerate(self.z):
if i > 0:
self.MF.update(z=z)
if i % size != rank:
continue
# Compute collapsed fraction
if self.hmf_func == 'PS' and self.hmf_analytic:
delta_c = self.MF.delta_c / self.MF.growth.growth_factor(z)
fcoll_tab[i] = erfc(delta_c / sqrt2 / self.MF._sigma_0)
else:
# Has units of h**4 / cMpc**3 / Msun
self.dndm[i] = self.MF.dndm.copy() * self.cosm.h70**4
self.mgtm[i] = self.MF.rho_gtm.copy()
self.ngtm[i] = self.MF.ngtm.copy() * self.cosm.h70**3
# Remember that mgtm and mean_density have factors of h**2
# so we're OK here dimensionally
fcoll_tab[i] = self.mgtm[i] / self.cosm.mean_density0
pb.update(i)
pb.finish()
# Collect results!
if size > 1:
tmp = np.zeros_like(fcoll_tab)
nothing = MPI.COMM_WORLD.Allreduce(fcoll_tab, tmp)
_fcoll_tab = tmp
tmp2 = np.zeros_like(self.dndm)
nothing = MPI.COMM_WORLD.Allreduce(self.dndm, tmp2)
self.dndm = tmp2
tmp3 = np.zeros_like(self.ngtm)
nothing = MPI.COMM_WORLD.Allreduce(self.ngtm, tmp3)
self.ngtm = tmp3
#
#tmp4 = np.zeros_like(self.mgtm)
#nothing = MPI.COMM_WORLD.Allreduce(self.mgtm, tmp4)
#self.mgtm = tmp4
else:
_fcoll_tab = fcoll_tab
# Fix NaN elements
_fcoll_tab[np.isnan(_fcoll_tab)] = 0.0
self._fcoll_tab = _fcoll_tab
def energy(hi, hj):
X = hi * np.sign(hj) / Gamma / Q
Ep = -Gamma * Gamma * np.log(eps +
(1 - 2 * eps) * erfc(-X / sqrt2) / 2)
return Ep
if (counter2 % 10 == 0):
pTot[int(counter2 / 10)] = sum(p[1:width]) * dx
aTot[int(counter2 / 10)] = sum(acetyl) * dx
pTot2[int(counter2 / 10)] = sum(p_2[1:width] * dx)
aTot2[int(counter2 / 10)] = sum(acetyl2) * dx
z = x / np.sqrt(4 * D0 * (telapsed))
pTotAnalytical[int(counter2 /
10)] = p0 * np.sqrt(4 * D0 * (telapsed) / np.pi)
#pTotAnalytical[int(counter2/10)-1] = scii.quad(densityAnalytical,0,x[width-1],args=(telapsed),epsabs=1e-15)[0]
#pTotAnalytical[int(counter2/10)-1] = sum(scis.erfc(z))
acetylationfit = 1 - np.exp(-p0 * arate * telapsed * (
(1 + 2 * (z**2)) * scis.erfc(z) -
2 * z * np.exp(-(z**2)) / np.sqrt(np.pi)))
aTotAnalytical[int(counter2 / 10)] = scii.quad(acetylAnalytical,
0,
x[width - 1],
args=(telapsed),
epsabs=1e-15)[0]
counter2 += 1
#p_2[0] = p0
print((time.time() - tstart) / 60)
pTotResidual[counter] = max(abs(pTot2 - pTotAnalytical))
pTotSFDResidual[counter] = max(abs(pTot - pTotAnalytical))
counter += 1
print(telapsed)
tend = time.time()
trun = (tend - tstart) / 60 #In minutes
def H(x):
return erfc(x/sqrt2)/2
def berawgn(EbN0):
""" Calculates theoretical bit error rate in AWGN (for BPSK and given Eb/N0) """
return 0.5 * erfc(math.sqrt(10**(float(EbN0) / 10)))
def antenna_pattern_gauss(d_az,
d_el,
antenna_3dB,
db=False,
twoway=True,
az_conv=None,
el_conv=None,
units='rad'):
"""
Get the antenna weighting factor due to the azimuth and elevation offsets
from the main antenna direction. The weighting factor is meant in terms of
power (not amplitude of the E-field). An Gaussian antenna pattern is
assumed.
Parameters
----------
d_az : array
Azimuth offsets to the main antenna axis [deg]
d_el : array
Elevation offset to the main antenna axis [deg]
antenna_3dB : float
Half power beam width [deg]
db : bool (optional)
If true return the result in dB instead linear.
twoway: bool (optional)
If true, return the two-way weighting factor instead of
the one-way factor.
az_conv (optional): float
If set, assumes that the antenna moves in azimuth direction (PPI) and
averages over the angle given by this keyword [deg].
el_conv (optional): float
If set, assumes that the antenna moves in elevation direction (RHI)
and averages over the angle given by this keyword [deg].
units (optional) : str
Specify if inputs quantities are given in radians ( "rad" ) or
degrees ( "deg" )
Returns
-------
fa : array
Weighting factor.
"""
if az_conv != None:
if az_conv <= 0:
az_conv = None
if el_conv != None:
if el_conv <= 0:
el_conv = None
if units not in ['deg', 'rad']:
print('Invalid units, must be either "rad" or "deg", ' +
'assuming they are "rad"')
if units == 'deg':
# Convert all quantities to rad
if az_conv is not None:
az_conv *= np.pi / 180.
if el_conv is not None:
el_conv *= np.pi / 180.
d_az = d_az.copy() * np.pi / 180.
d_el = d_el.copy() * np.pi / 180.
antenna_3dB *= np.pi / 180.
if az_conv == None or el_conv == None:
if az_conv is not None:
""" The antenna is moving in azimuth direction and is averaging the
received pulses over 'az_conv' deg. The norm azimuth position
of the antenna is reached, when the antenna moved half of the
azimuth distance (az_offset = 'az_conv'/2 deg).
The weighting factor at the azimuth position 'daz' from the norm
position is given by the following integral:
1 / daz+az_offset
fa(daz, del) = ---- * | f(daz) d(daz) * f(del)
Norm / daz-az_offset
where
daz : Is the azimuth deviation from the norm antenna position
del : Is the elevation deviation from the norm antenna position
fa(daz,del) : Weighting factor at point (daz,del) f(0,0) must be 1.
Norm : Normalization such that f(0,0)=1
f(x) : Weighting factor of the non moving antenna (Gaussian
function, see below)
Solving the integral above leads to:
K1
fa(daz, del) = ---- * ( erf(K*(daz+az_offset)) -erf(K*(daz-az_offset)) ) * f(del)
Norm
where
2 * sqrt(ln(2))
K = ---------------
phi3db
sqrt(!PI)
K1 = ---------
2 * K
erf : the error function
phi3db : the half power beam width
"""
az_offset = az_conv / 2.
K = 2. * np.sqrt(np.log(2)) / antenna_3dB
K1 = np.sqrt(np.pi) / 2. / K
Norm = 2.0 * K1 * erfc(K * az_offset)
faz = (K1 / Norm *
(erfc(K * (d_az + az_offset)) - erfc(K *
(d_az - az_offset))))
else:
da = (2. * d_az / antenna_3dB)**2
ind = da > 20.
da[ind] = 20
faz = np.exp(-da * np.log10(2))
if el_conv is not None:
# see explanation for az_conv above
el_offset = el_conv / 2. * np.pi / 180.
K = 2. * np.sqrt(np.log(2)) / antenna_3dB
K1 = np.sqrt(np.pi) / 2. / K
Norm = 2.0 * K1 * erfc(K * el_offset)
fel = (K1 / Norm *
(erfc(K * (d_el + el_offset)) - erfc(K *
(d_el - el_offset))))
else:
de = (2. * d_el / antenna_3dB)**2
ind = de > 20.
de[ind] = 20
fel = np.exp(-de * np.log10(2))
fa = faz * fel
else:
# Gaussian antenna pattern:
#
# f(daz,del) = e^(-( (2*daz/phi3db_el)^2 + (2*del/phi3db_az)^2 ) * ln(2))
#
# from Gauss normal distribution N(x) with N(x)=1 and N(x=X0/2)=1/2 :
#
# N(x) = e^(-(2*x/X0)^2 * ln(2))
da = 2. * d_az / antenna_3dB
de = 2. * d_el / antenna_3dB
dr = (da**2 + de**2)
ind = dr > 20.
dr[ind] = 20
fa = np.exp(-dr * np.log(2))
if twoway:
fa = fa**2
if db:
fa = 10. * np.log10(fa)
return fa
def fun(x):
return 1-0.5*special.erfc(x/2)
def ber_qpsk(Eb, n_0):
'''
Calcula o BER de acordo com a modulação QPSK
'''
if (n_0 == 0): return 0
return 1 / 2 * erfc(np.sqrt(Eb / n_0))
def _calc(self, x, y):
"""
List based implementation of binary tree algorithm for concordance
measure after Christensen (2005).
"""
x = np.array(x)
y = np.array(y)
n = len(y)
perm = range(n)
perm.sort(key=lambda a: (x[a], y[a]))
vals = y[perm]
ExtraY = 0
ExtraX = 0
ACount = 0
BCount = 0
CCount = 0
DCount = 0
ECount = 0
DCount = 0
Concordant = 0
Discordant = 0
# ids for left child
li = [None] * (n - 1)
# ids for right child
ri = [None] * (n - 1)
# number of left descendants for a node
ld = np.zeros(n)
# number of values equal to value i
nequal = np.zeros(n)
for i in range(1, n):
NumBefore = 0
NumEqual = 1
root = 0
x0 = x[perm[i - 1]]
y0 = y[perm[i - 1]]
x1 = x[perm[i]]
y1 = y[perm[i]]
if x0 != x1:
DCount = 0
ECount = 1
else:
if y0 == y1:
ECount += 1
else:
DCount += ECount
ECount = 1
root = 0
inserting = True
while inserting:
current = y[perm[i]]
if current > y[perm[root]]:
# right branch
NumBefore += 1 + ld[root] + nequal[root]
if ri[root] is None:
# insert as right child to root
ri[root] = i
inserting = False
else:
root = ri[root]
elif current < y[perm[root]]:
# increment number of left descendants
ld[root] += 1
if li[root] is None:
# insert as left child to root
li[root] = i
inserting = False
else:
root = li[root]
elif current == y[perm[root]]:
NumBefore += ld[root]
NumEqual += nequal[root] + 1
nequal[root] += 1
inserting = False
ACount = NumBefore - DCount
BCount = NumEqual - ECount
CCount = i - (ACount + BCount + DCount + ECount - 1)
ExtraY += DCount
ExtraX += BCount
Concordant += ACount
Discordant += CCount
cd = Concordant + Discordant
num = Concordant - Discordant
tau = num / np.sqrt((cd + ExtraX) * (cd + ExtraY))
v = (4. * n + 10) / (9. * n * (n - 1))
z = tau / np.sqrt(v)
pval = erfc(np.abs(z) / 1.4142136) # follow scipy
return tau, pval, Concordant, Discordant, ExtraX, ExtraY
