def sim_lc(t, ppm, per_aper_ratio, periodic_freq, aperiodic_freq, A, planet=False, mean=1.0, planet_params=[0, 0, 0, 0, 0, 0, 0, 0, 0]):
if not planet:
mean=mean
if planet:
mean=transit_model(*planet_params, t)
aperiodic_sigma = 1/per_aper_ratio
periodic_sigma = 1
white_noise_sigma = ppm/1e6
# set up a gaussian process with two components and white noise
# non-periodic component
Q = 1. / np.sqrt(2.0) # related to the frequency of the variability
w0 = 2*np.pi*aperiodic_freq
S0 = (aperiodic_sigma**2.) / (w0 * Q)
bounds = dict(log_S0=(-15, 15), log_Q=(-15, 15), log_omega0=(-15, 15))
kernel = terms.SHOTerm(log_S0=np.log(S0), log_Q=np.log(Q), log_omega0=np.log(w0), bounds=bounds)
# periodic component
Q = 1.0
w0 = 2*np.pi*periodic_freq
S0 = (periodic_sigma**2.) / (w0 * Q)
kernel += terms.SHOTerm(log_S0=np.log(S0), log_Q=np.log(Q), log_omega0=np.log(w0), bounds=bounds)
# white noise
# kernel += terms.JitterTerm(log_sigma=np.log(white_noise_sigma), bounds=dict(log_sigma=(-15,15)))
gp = celerite.GP(kernel, mean=0, fit_mean=True, fit_white_noise=True)
gp.compute(t, white_noise_sigma)
return A * gp.sample() + mean + white_noise_sigma * np.random.randn(len(t))
def generate_solar_fluxes(duration, cadence=60 * u.s, seed=None):
"""
Generate an array of fluxes with zero mean which mimic the power spectrum of
the SOHO/VIRGO SPM observations.
Parameters
----------
duration : `~astropy.units.Quantity`
Duration of simulated observations to generate.
cadence : `~astropy.units.Quantity`
Length of time between fluxes
seed : float, optional
Random seed.
Returns
-------
times : `~astropy.units.Quantity`
Array of times at cadence ``cadence`` of length ``duration/cadence``
fluxes : `~numpy.ndarray`
Array of fluxes at cadence ``cadence`` of length ``duration/cadence``
kernel : `~celerite.terms.TermSum`
Celerite kernel used to approximate the solar power spectrum.
"""
if seed is not None:
np.random.seed(seed)
_process_inputs(duration, cadence, 5777 * u.K)
##########################
# Assemble celerite kernel
##########################
parameter_vector = np.copy(PARAM_VECTOR)
nterms = len(parameter_vector) // 3
kernel = terms.SHOTerm(log_S0=0, log_omega0=0, log_Q=0)
for term in range(nterms - 1):
kernel += terms.SHOTerm(log_S0=0, log_omega0=0, log_Q=0)
kernel.set_parameter_vector(parameter_vector)
gp = celerite.GP(kernel)
times = np.arange(0, duration.to(u.s).value, cadence.to(u.s).value) * u.s
x = times.value
gp.compute(x, check_sorted=False)
###################################
# Get samples with the kernel's PSD
###################################
y = gp.sample()
# Remove a linear trend:
y -= np.polyval(np.polyfit(x - x.mean(), y, 1), x - x.mean())
return times, y, kernel
def generate_solar_fluxes(size,
cadence=60 * u.s,
parameter_vector=parameter_vector):
"""
Generate an array of fluxes with zero mean which mimic the power spectrum of
the SOHO/VIRGO SPM observations.
Parameters
----------
size : int
Number of fluxes to generate. Note: Assumes ``size``>>500.
cadence : `~astropy.units.Quantity`
Length of time between fluxes
Returns
-------
y : `~numpy.ndarray`
Array of fluxes at cadence ``cadence`` of length ``size``.
"""
nterms = len(parameter_vector) // 3
kernel = terms.SHOTerm(log_S0=0, log_omega0=0, log_Q=0)
for term in range(nterms - 1):
kernel += terms.SHOTerm(log_S0=0, log_omega0=0, log_Q=0)
kernel.set_parameter_vector(parameter_vector)
gp = celerite.GP(kernel)
x = np.arange(0, size // 500, cadence.to(u.s).value)
gp.compute(x, check_sorted=False)
y = gp.sample(500)
y_concatenated = []
for i, yi in enumerate(y):
xi = np.arange(len(yi))
fit = np.polyval(np.polyfit(xi - xi.mean(), yi, 1), xi - xi.mean())
yi -= fit
if i == 0:
y_concatenated.append(yi)
else:
offset = yi[0] - y_concatenated[i - 1][-1]
y_concatenated.append(yi - offset)
y_concatenated = np.hstack(y_concatenated)
x_c = np.arange(len(y_concatenated))
y_concatenated -= np.polyval(
np.polyfit(x_c - x_c.mean(), y_concatenated, 1), x_c - x_c.mean())
return y_concatenated, kernel
def dSHO_maxlikelihood(lc, npeaks=1):
# Now let's do some of the GP stuff on this
# A non-periodic component
Q = 1.0 / np.sqrt(2.0)
w0 = 3.0
S0 = np.var(lc['NormFlux']) / (w0 * Q)
bounds = dict(log_S0=(-16, 16), log_Q=(-15, 15), log_omega0=(-15, 15))
kernel = terms.SHOTerm(log_S0=np.log(S0),
log_Q=np.log(Q),
log_omega0=np.log(w0),
bounds=bounds)
kernel.freeze_parameter(
"log_Q") # We don't want to fit for "Q" in this term
# A periodic component
for i in range(npeaks):
Q = 1.0
w0 = 3.0
S0 = np.var(lc['NormFlux']) / (w0 * Q)
kernel += terms.SHOTerm(log_S0=np.log(S0),
log_Q=np.log(Q),
log_omega0=np.log(w0),
bounds=bounds)
sigma = np.median(lc['NormErr'])
kernel += terms.JitterTerm(log_sigma=np.log(sigma))
gp = celerite.GP(kernel, mean=np.mean(lc['NormFlux']))
gp.compute(lc['Time'],
lc['NormErr']) # You always need to call compute once.
print("Initial log likelihood: {0}".format(
gp.log_likelihood(lc['NormFlux'])))
initial_params = gp.get_parameter_vector()
bounds = gp.get_parameter_bounds()
r = minimize(neg_log_like,
initial_params,
method="L-BFGS-B",
bounds=bounds,
args=(lc['NormFlux'], gp))
gp.set_parameter_vector(r.x)
print("Final log likelihood: {0}".format(gp.log_likelihood(
lc['NormFlux'])))
print("Maximum Likelihood Soln: {}".format(gp.get_parameter_dict()))
return gp
def gp_sho(idx, data, resid, params):
logQ, logw, logS, log_jit = params
err = data.err / data.flux
#logerr = np.log(np.median(err))
logerr = log_jit
#print "logQ, logw, logS", logQ, logw, logS
mean_resid = np.mean(resid)
#resid -= mean_resid
t = data.t_vis[idx]
kernel = (terms.SHOTerm(log_S0=logS, log_Q=logQ, log_omega0=logw) +
terms.JitterTerm(log_sigma=logerr))
gp = celerite.GP(kernel, fit_mean=True)
gp.compute(t, err, check_sorted=False) #t must be ascending!
mu = gp.predict(resid, t, return_cov=False)
gp_lnlike = gp.log_likelihood(resid)
"""plt.errorbar(t, resid, err, fmt = '.k')
plt.plot(t, mu)
x = np.linspace(np.min(t), np.max(t), 1000)
pred_mean, pred_var = gp.predict(resid, x, return_var = True)
pred_std = np.sqrt(pred_var)
plt.fill_between(x, pred_mean+pred_std, pred_mean-pred_std, color='blue', alpha=0.3)
plt.show()"""
#np.save("resid", [t, resid, err])
#return [1.0 + np.array(mu) + mean_resid, gp_lnlike]
return [1.0 + np.array(mu), gp_lnlike]
def GPfit(times, fluxes, errors, nonmask_idxs):
#t, y, yerr = times[np.isfinite(times)], fluxes[np.isfinite(fluxes)], errors[np.isfinite(errors)]
t, y, yerr = times[nonmask_idxs], fluxes[nonmask_idxs], errors[nonmask_idxs]
t, y, yerr = t[np.isfinite(t)], y[np.isfinite(y)], yerr[np.isfinite(yerr)]
Q = 1.0
w0 = 3.0
S0 = np.var(y) / (w0 * Q)
bounds = dict(log_S0=(-15, 15), log_Q=(-15, 15), log_omega0=(-15, 15))
kernel = terms.SHOTerm(log_S0=np.log(S0), log_Q=np.log(Q), log_omega0=np.log(w0), bounds=bounds)
#kernel += terms.SHOTerm(log_S0=np.log(S0), log_Q=np.log(Q), log_omega0=np.log(w0) bounds=bounds)
gp = celerite.GP(kernel, mean=np.mean(y))
gp.compute(t, yerr) # You always need to call compute once.
print("Initial log likelihood: {0}".format(gp.log_likelihood(y)))
initial_params = gp.get_parameter_vector()
bounds = gp.get_parameter_bounds()
r = minimize(neg_log_like, initial_params, method="L-BFGS-B", bounds=bounds, args=(y, gp))
gp.set_parameter_vector(r.x)
print(r)
#interpolated_times = np.linspace(np.nanmin(t), np.nanmax(t), 1000)
#interpolated_epochs = np.linspace(np.nanmin(t), np.nanmax(t), 1000)
#pred_mean, pred_var = gp.predict(y, interpolated_times, return_var=True)
pred_mean, pred_var = gp.predict(y, times, return_var=True)
pred_std = np.sqrt(pred_var)
return pred_mean, pred_std ### should be the same dimension as times.
def __init__(self, params):
Ampl, Plife, Prot = params
S0 = (Ampl*(Prot**2))/(2. * (np.pi**2) * Plife)
w0 = (2.0*np.pi)/Prot
Q = Plife(np.pi)/Prot
kernel = terms.SHOTerm(log_S0=np.log(S0), log_Q=np.log(Q), log_omega0=np.log(w0) )
def simple_kernel():
# A non-periodic component
Q = 1.0 / np.sqrt(2.0)
w0 = 3.0
S0 = np.var(y) / (w0 * Q)
kernel = terms.SHOTerm(log_S0=np.log(S0), log_Q=np.log(Q),
log_omega0=np.log(w0),
bounds=[(-15, 15), (-15, 15), (-15, 15)])
kernel.freeze_parameter("log_Q") # We don't want to fit for "Q" in this term
# A periodic component
Q = 1.0
w0 = 3.0
S0 = np.var(y) / (w0 * Q)
kernel += terms.SHOTerm(log_S0=np.log(S0), log_Q=np.log(Q),
log_omega0=np.log(w0),
bounds=[(-15, 15), (-15, 15), (-15, 15)])
return kernel
def log_probability(self, params, band):
"""
Calculate log of posterior probability
Parameters
-----------
params : iterable
list of parameter values
band : string
SDSS/HiPERCAM band
"""
self.set_parameter_vector(params)
t, _, y, ye, _, _ = np.loadtxt(self.lightcurves[band]).T
# check model params are valid - checks against bounds
lp = self.log_prior()
if not np.isfinite(lp):
return -np.inf
# make a GP for this band, if it doesnt already exist
if not hasattr(self, 'gpdict'):
self.gpdict = dict()
# Oscillation params
pulsation_amp = self.pulse_amp * scale_pulsation(band, self.pulse_temp)
if band not in self.gpdict:
kernel = terms.SHOTerm(np.log(pulsation_amp), np.log(self.pulse_q),
np.log(self.pulse_omega))
gp = celerite.GP(kernel)
gp.compute(t, ye)
self.gpdict[band] = gp
else:
gp = self.gpdict[band]
gp.set_parameter_vector(
(np.log(pulsation_amp), np.log(self.pulse_q),
np.log(self.pulse_omega)))
gp.compute(t, ye)
# now add prior of Gaussian process
lp += gp.log_prior()
if not np.isfinite(lp):
return -np.inf
try:
ym = self.get_value(band)
except ValueError as err:
# invalid lcurve params
print('warning: model failed ', err)
return -np.inf
else:
return gp.log_likelihood(y - ym) + lp
def get_basic_kernel(t, y, yerr):
kernel = terms.SHOTerm(
log_S0=np.log(np.var(y)),
log_Q=-np.log(4.0),
log_omega0=np.log(2*np.pi/10.),
bounds=dict(
log_S0=(-20.0, 10.0),
log_omega0=(np.log(2*np.pi/80.0), np.log(2*np.pi/2.0)),
),
)
kernel.freeze_parameter('log_Q')
# Finally some jitter
kernel += terms.JitterTerm(log_sigma=np.log(yerr),
bounds=[(-20.0, 5.0)])
return kernel
def compute_background(t, amp, freqs, white=0):
"""
Compute granulation background using Gaussian process
"""
if white == 0:
white = 1e-6
kernel = terms.JitterTerm(log_sigma=np.log(white))
S0 = calculateS0(amp, freqs)
print(f"S0: {S0}")
for i in range(len(amp)):
kernel += terms.SHOTerm(log_S0=np.log(S0[i]),
log_Q=np.log(1 / np.sqrt(2)),
log_omega0=np.log(2 * np.pi * freqs[i]))
gp = celerite.GP(kernel)
return kernel, gp, S0
def get_basic_kernel(t, y, yerr, period=False):
if not period:
period = 0.5
kernel = terms.SHOTerm(
log_S0=np.log(np.var(y)),
log_Q=-np.log(4.0),
log_omega0=np.log(2 * np.pi / 20.),
bounds=dict(
log_S0=(-20.0, 10.0),
log_omega0=(np.log(2 * np.pi / 100.), np.log(2 * np.pi / (10))),
),
)
kernel.freeze_parameter('log_Q')
## tau = 2*np.exp(-1*np.log(4.0))/np.exp(log_omega0)
# Finally some jitter
ls = np.log(np.median(yerr))
kernel += terms.JitterTerm(log_sigma=ls, bounds=[(ls - 5.0, ls + 5.0)])
return kernel
def test_product(seed=42):
np.random.seed(seed)
t = np.sort(np.random.uniform(0, 5, 100))
tau = t[:, None] - t[None, :]
k1 = terms.RealTerm(log_a=0.1, log_c=0.5)
k2 = terms.ComplexTerm(0.2, -3.0, 0.5, 0.01)
k3 = terms.SHOTerm(1.0, 0.2, 3.0)
K1 = k1.get_value(tau)
K2 = k2.get_value(tau)
K3 = k3.get_value(tau)
assert np.allclose((k1 + k2).get_value(tau), K1 + K2)
assert np.allclose((k3 + k2).get_value(tau), K3 + K2)
assert np.allclose((k1 + k2 + k3).get_value(tau), K1 + K2 + K3)
for (a, b), (A, B) in zip(
product((k1, k2, k3, k1 + k2, k1 + k3, k2 + k3), (k1, k2, k3)),
product((K1, K2, K3, K1 + K2, K1 + K3, K2 + K3), (K1, K2, K3))):
assert np.allclose((a * b).get_value(tau), A * B)
mean_model = Model(**kwargs)
# mean_model = Model(P1=8., tau1=1., k1=np.std(y)/100, w1=0., e1=0.4,
# P2=100, tau2=1., k2=np.std(y)/100, w2=0., e2=0.4, offset1=0., offset2=0.)
#==============================================================================
# The fit
#==============================================================================
from scipy.optimize import minimize
import celerite
from celerite import terms
# Set up the GP model
# kernel = terms.RealTerm(log_a=np.log(np.var(y)), log_c=-np.log(10.0))
kernel = terms.SHOTerm(log_S0=np.log(2), log_Q=np.log(2), log_omega0=np.log(5))
gp = celerite.GP(kernel, mean=mean_model, fit_mean=True)
gp.compute(x, yerr)
print("Initial log-likelihood: {0}".format(gp.log_likelihood(y)))
# Define a cost function
def neg_log_like(params, y, gp):
gp.set_parameter_vector(params)
return -gp.log_likelihood(y)
# def grad_neg_log_like(params, y, gp):
# gp.set_parameter_vector(params)
# return -gp.grad_log_likelihood(y)[1]
# Fit for the maximum likelihood parameters
initial_params = gp.get_parameter_vector()
#ecosw=self.ecosw, # eccentricity vector
#esinw=self.esinw,
occ=0.0) # a secondary eclipse depth in ppm
M.add_data(time=t)
return M.transitmodel
#set the GP parameters
Q = 1.0 / np.sqrt(2.0)
w0 = muhz2omega(13)
S0 = np.var(y) / (w0 * Q)
kernel = terms.SHOTerm(log_S0=np.log(S0),
log_Q=np.log(Q),
log_omega0=np.log(w0),
bounds=[(-25, 0), (-15, 15),
(np.log(muhz2omega(3)), np.log(muhz2omega(50)))
]) #omega upper bound: 275 muhz
kernel.freeze_parameter("log_Q") #to make it a Harvey model
Q = 1.0 / np.sqrt(2.0)
w0 = muhz2omega(61.0)
S0 = np.var(y) / (w0 * Q)
kernel += terms.SHOTerm(log_S0=np.log(S0),
log_Q=np.log(Q),
log_omega0=np.log(w0),
bounds=[(-25, 0), (-15, 15),
(np.log(muhz2omega(30)),
np.log(muhz2omega(1000)))])
kernel.freeze_parameter("terms[1]:log_Q") #to make it a Harvey model
plt.errorbar(t, y, yerr=yerr, fmt=".k", capsize=0)
plt.ylabel('RV' r"$[m/s]$")
plt.xlabel(r"$t$")
plt.title("Simulated data -- planet and jitter")
plt.show()
#==============================================================================
# Modelling
#==============================================================================
import celerite
celerite.__version__
from celerite import terms
kernel = terms.SHOTerm(np.log(2), np.log(2), np.log(25))
#gp = celerite.GP(kernel, mean=Model(amp=np.var(y)/2, P=7, phase=0))
gp = celerite.GP(kernel, mean=Model(**truth), fit_mean=True)
#gp = celerite.GP(kernel, mean=Model(**truth ), log_white_noise=np.log(1), fit_white_noise=True)
gp.compute(t, yerr)
def lnprob2(p):
gp.set_parameter_vector(p) # Set the parameter values to the given vector
return gp.log_likelihood(y, quiet=True) + gp.log_prior(
) # Compute the logarithm of the marginalized likelihood of a set of observations under the Gaussian process model.
#==============================================================================
# run MCMC on this model
#==============================================================================
log_period=0.0,
bounds=[(-10, 10), (-10, 10), (-10, 10)])
# Simulate a dataset from the true model
np.random.seed(42)
N = 100
t = np.sort(np.random.uniform(0, 20, N))
yerr = 0.5
K = true_model.get_K(t)
K[np.diag_indices_from(K)] += yerr**2
y = np.random.multivariate_normal(np.zeros(N), K)
# Set up the celerite model that we will use to fit - product of two SHOs
log_Q = 1.0
kernel = terms.SHOTerm(log_S0=np.log(np.var(y)) - 2 * log_Q,
log_Q=log_Q,
log_omega0=np.log(2 * np.pi))
kernel *= terms.SHOTerm(log_S0=0.0, log_omega0=0.0, log_Q=-0.5 * np.log(2))
kernel.freeze_parameter("k2:log_S0")
kernel.freeze_parameter("k2:log_Q")
gp = celerite.GP(kernel)
gp.compute(t, yerr)
# Fit for the maximum likelihood
def nll(params, gp, y):
gp.set_parameter_vector(params)
if not np.isfinite(gp.log_prior()):
return 1e10
ll = gp.log_likelihood(y)
truth = dict(P1=8., tau1=1., k1=np.std(y)/100, w1=0., e1=0.4, offset1=0., offset2=0.)
kwargs = dict(**truth)
kwargs["bounds"] = dict(P1=(7.5,8.5), k1=(0,0.1), w1=(-2*np.pi,2*np.pi), e1=(0,0.8))
mean_model = Model(**kwargs)
#==============================================================================
# The fit
#==============================================================================
from scipy.optimize import minimize
import celerite
from celerite import terms
# Set up the GP model
# kernel = terms.RealTerm(log_a=np.log(np.var(y)), log_c=-np.log(10.0))
kernel = terms.SHOTerm(log_S0=np.log(2), log_Q=np.log(20), log_omega0=np.log(1/3000))
gp = celerite.GP(kernel, mean=mean_model, fit_mean=True)
gp.compute(x, yerr)
print("Initial log-likelihood: {0}".format(gp.log_likelihood(y)))
# Define a cost function
def neg_log_like(params, y, gp):
gp.set_parameter_vector(params)
return -gp.log_likelihood(y)
# def grad_neg_log_like(params, y, gp):
# gp.set_parameter_vector(params)
# return -gp.grad_log_likelihood(y)[1]
# Fit for the maximum likelihood parameters
initial_params = gp.get_parameter_vector()
test_terms = [
cterms.RealTerm(log_a=np.log(2.5), log_c=np.log(1.1123)),
cterms.RealTerm(log_a=np.log(12.345), log_c=np.log(1.5)) +
cterms.RealTerm(log_a=np.log(0.5), log_c=np.log(1.1234)),
cterms.ComplexTerm(log_a=np.log(10.0),
log_c=np.log(5.6),
log_d=np.log(2.1)),
cterms.ComplexTerm(
log_a=np.log(7.435),
log_b=np.log(0.5),
log_c=np.log(1.102),
log_d=np.log(1.05),
),
cterms.SHOTerm(log_S0=np.log(1.1),
log_Q=np.log(0.1),
log_omega0=np.log(1.2)),
cterms.SHOTerm(log_S0=np.log(1.1),
log_Q=np.log(2.5),
log_omega0=np.log(1.2)),
cterms.SHOTerm(
log_S0=np.log(1.1), log_Q=np.log(2.5), log_omega0=np.log(1.2)) +
cterms.RealTerm(log_a=np.log(1.345), log_c=np.log(2.4)),
cterms.SHOTerm(
log_S0=np.log(1.1), log_Q=np.log(2.5), log_omega0=np.log(1.2)) *
cterms.RealTerm(log_a=np.log(1.345), log_c=np.log(2.4)),
cterms.Matern32Term(log_sigma=0.1, log_rho=0.4),
]
@pytest.mark.parametrize("oterm", test_terms)
return k_out
def neo_update_kernel(theta, params):
gp = george.GP(mean=0.0, fit_mean=False, white_noise=jitt)
pass
from celerite import terms as cterms
# 2 or sp.log(10.) ?
T = {
'Constant': 1.**2,
'RealTerm': cterms.RealTerm(log_a=2., log_c=2.),
'ComplexTerm': cterms.ComplexTerm(log_a=2., log_b=2., log_c=2., log_d=2.),
'SHOTerm': cterms.SHOTerm(log_S0=2., log_Q=2., log_omega0=2.),
'Matern32Term': cterms.Matern32Term(log_sigma=2., log_rho=2.0),
'JitterTerm': cterms.JitterTerm(log_sigma=2.0)
}
def neo_term(terms):
t_out = T[terms[0][0]]
for f in range(len(terms[0])):
if f == 0:
pass
else:
t_out *= T[terms[0][f]]
for i in range(len(terms)):
if i == 0:
pad = 25
log_omega0 = 3
log_S0 = 0
log_Q = 2.0
amp = 10
nrot = 30
npts = 1000
nsamples = 5
x = np.linspace(0, nrot * 2 * np.pi / np.exp(log_omega0), npts, endpoint=False)
yerr = 1e2 * np.ones_like(x)
yerr[:pad] = np.logspace(-12, 2, pad)
yerr[-pad:] = np.logspace(2, -12, pad)
kernel = terms.SHOTerm(log_S0=log_S0, log_Q=log_Q, log_omega0=log_omega0)
gp = celerite.GP(kernel)
gp.compute(x, yerr)
y = [None for k in range(nsamples)]
for k in range(nsamples):
y0 = (amp * np.exp(log_S0) *
np.sin(np.exp(log_omega0) * x + 2 * np.pi * np.random.random()))
y[k] = gp.sample_conditional(y0)
y[k] = (y[k] - y[k].min()) / (y[k].max() - y[k].min())
fig, ax = plt.subplots(figsize=(3.6, 3.2), facecolor="#e5e5e5")
fig.patch.set_facecolor("#e5e5e5")
fig.subplots_adjust(left=0, bottom=0, top=1, right=1)
ax.set(xlim=(0, 1), ylim=(-0.1, 1.1))
ax.axis("off")
class CustomTerm(terms.Term):
parameter_names = ("amp", "P", "phase", "offset")
bounds = dict(amp=(0, 0.002),
P=(3, 10),
phase=(0, 2 * np.pi),
offset=(min(y), max(y)))
kernel1 = CustomTerm(amp=0.001,
P=8,
phase=1,
offset=np.median(y),
bounds=bounds)
kernel2 = terms.SHOTerm(np.log(1.1), np.log(1.1), np.log(25))
kernel = kernel1 * kernel2
#gp = celerite.GP(kernel, mean=Model(amp=np.var(y)/2, P=7, phase=0))
gp = celerite.GP(kernel,
mean=Model(amp=(np.var(y))**0.5,
P=7.75,
phase=1,
offset=np.median(y)),
fit_mean=True)
#gp = celerite.GP(kernel, mean=Model(**truth ), fit_mean = True)
#gp = celerite.GP(kernel, mean=Model(**truth ), fit_mean = True, log_white_noise=np.log(1), fit_white_noise=True)
gp.compute(t, yerr)
def lnprob2(p):
gp.set_parameter_vector(p) # Set the parameter values to the given vector
assert all(np.allclose(a, b) for a, b in zip(b0, bounds))
kernel = terms.RealTerm(log_a=0.1,
log_c=0.5,
bounds=dict(zip(["log_a", "log_c"], bounds)))
assert all(
np.allclose(a, b) for a, b in zip(b0, kernel.get_parameter_bounds()))
@pytest.mark.parametrize("k", [
terms.RealTerm(log_a=0.1, log_c=0.5),
terms.RealTerm(log_a=0.1, log_c=0.5) +
terms.RealTerm(log_a=-0.1, log_c=0.7),
terms.ComplexTerm(log_a=0.1, log_c=0.5, log_d=0.1),
terms.ComplexTerm(log_a=0.1, log_b=-0.2, log_c=0.5, log_d=0.1),
terms.SHOTerm(log_S0=0.1, log_Q=-1, log_omega0=0.5),
terms.SHOTerm(log_S0=0.1, log_Q=1.0, log_omega0=0.5),
terms.SHOTerm(log_S0=0.1, log_Q=1.0, log_omega0=0.5) +
terms.RealTerm(log_a=0.1, log_c=0.4),
terms.SHOTerm(log_S0=0.1, log_Q=1.0, log_omega0=0.5) *
terms.RealTerm(log_a=0.1, log_c=0.4),
])
def test_jacobian(k, eps=1.34e-7):
if not terms.HAS_AUTOGRAD:
with pytest.raises(ImportError):
jac = k.get_coeffs_jacobian()
return
v = k.get_parameter_vector()
c = np.concatenate(k.coefficients)
jac = k.get_coeffs_jacobian()
def generate_stellar_fluxes(size,
M,
T_eff,
L,
cadence=60 * u.s,
parameter_vector=parameter_vector):
"""
Generate an array of fluxes with zero mean which mimic the power spectrum of
the SOHO/VIRGO SPM observations.
Parameters
----------
size : int
Number of fluxes to generate. Note: Assumes ``size``>>500.
M : `~astropy.units.Quantity`
Stellar mass
T_eff : `~astropy.units.Quantity`
Stellar effective temperature
L : `~astropy.units.Quantity`
Steller luminosity
cadence : `~astropy.units.Quantity`
Length of time between fluxes
Returns
-------
y : `~numpy.ndarray`
Array of fluxes at cadence ``cadence`` of length ``size``.
"""
parameter_vector = parameter_vector.copy()
tunable_amps = np.exp(parameter_vector[::3][2:])
tunable_freqs = np.exp(parameter_vector[2::3][2:]) * 1e6 / 2 / np.pi
peak_ind = np.argmax(tunable_amps)
peak_freq = tunable_freqs[peak_ind]
delta_freqs = tunable_freqs - peak_freq
T_eff_solar = 5777 * u.K
nu_max_sun = peak_freq * u.uHz
delta_nu_sun = 135.1 * u.uHz
# Huber 2011 Eqn 1
nu_factor = ((M / M_sun) * (T_eff / T_eff_solar)**3.5 / (L / L_sun))
# Huber 2011 Eqn 2
delta_nu_factor = ((M / M_sun)**0.5 * (T_eff / T_eff_solar)**3 /
(L / L_sun)**0.75)
new_peak_freq = nu_factor * peak_freq
new_delta_freqs = delta_freqs * delta_nu_factor
new_peak_freq, new_delta_freqs
new_freqs = new_peak_freq + new_delta_freqs
new_log_omegas = np.log(2 * np.pi * new_freqs * 1e-6).value
parameter_vector[2::3][2:] = new_log_omegas
nterms = len(parameter_vector) // 3
kernel = terms.SHOTerm(log_S0=0, log_omega0=0, log_Q=0)
for term in range(nterms - 1):
kernel += terms.SHOTerm(log_S0=0, log_omega0=0, log_Q=0)
kernel.set_parameter_vector(parameter_vector)
gp = celerite.GP(kernel)
x = np.arange(0, size // 500, cadence.to(u.s).value)
gp.compute(x, check_sorted=False)
y = gp.sample(500)
y_concatenated = []
for i, yi in enumerate(y):
xi = np.arange(len(yi))
fit = np.polyval(np.polyfit(xi - xi.mean(), yi, 1), xi - xi.mean())
yi -= fit
if i == 0:
y_concatenated.append(yi)
else:
offset = yi[0] - y_concatenated[i - 1][-1]
y_concatenated.append(yi - offset)
y_concatenated = np.hstack(y_concatenated)
x_c = np.arange(len(y_concatenated))
y_concatenated -= np.polyval(
np.polyfit(x_c - x_c.mean(), y_concatenated, 1), x_c - x_c.mean())
return y_concatenated, kernel
print('guess v: ',np.median(slit['rv']))
print('guess K: ',np.std(slit['rv']-np.median(slit['rv'])))
print('guess v: ',np.median(fiber['rv']))
print('guess K: ',np.std(fiber['rv']-np.median(fiber['rv'])))
#print(len(slit))67
#print(len(fiber))82
#set the GP parameters-FROM SAM RAW
#First granulation
Q = 1.0 / np.sqrt(2.0)
w0 = muhz2omega(20)
S0 = np.var(fiber['rv']) / (w0*Q)
kernel = terms.SHOTerm(log_S0=np.log(S0), log_Q=np.log(Q), log_omega0=np.log(w0),
bounds=[(-20, 20), (-15, 15), (np.log(muhz2omega(0.1)), np.log(muhz2omega(100)))]) #omega upper bound: 10 muhz
kernel.freeze_parameter("log_Q") #to make it a Harvey model
#numax
Q = np.exp(3.0)
w0 = muhz2omega(25) #peak of oscillations
S0 = np.var(fiber['rv']) / (w0*Q)
kernel += terms.SHOTerm(log_S0=np.log(S0), log_Q=np.log(Q), log_omega0=np.log(w0),
bounds=[(-20, 20), (0.1, 5), (np.log(muhz2omega(5)), np.log(muhz2omega(50)))])
kernel += terms.JitterTerm(log_sigma=1, bounds=[(-20,40)])
#initial guess of RV model
def iterGP(x,
y,
yerr,
period_guess,
acf_1pk,
num_iter=20,
ax=None,
n_samp=4000):
# Here is the kernel we will use for the GP regression
# It consists of a sum of two stochastically driven damped harmonic
# oscillators. One of the terms has Q fixed at 1/sqrt(2), which
# forces it to be non-periodic. There is also a white noise term
# included.
# Do some aggressive sigma clipping
m = np.ones(len(x), dtype=bool)
while True:
mu = np.mean(y[m])
sig = np.std(y[m])
m0 = y - mu < 3 * sig
if np.all(m0 == m):
break
m = m0
x_clip, y_clip, yerr_clip = x[m], y[m], yerr[m]
if len(x_clip) < n_samp:
n_samp = len(x_clip)
# Randomly select n points from the light curve for the GP fit
x_ind_rand = np.random.choice(len(x_clip), n_samp, replace=False)
x_ind = x_ind_rand[np.argsort(x_clip[x_ind_rand])]
x_gp = x_clip[x_ind]
y_gp = y_clip[x_ind]
yerr_gp = yerr_clip[x_ind]
# A non-periodic component
Q = 1.0 / np.sqrt(2.0)
w0 = 3.0
S0 = np.var(y_gp) / (w0 * Q)
bounds = dict(log_S0=(-20, 15), log_Q=(-15, 15), log_omega0=(-15, 15))
kernel = terms.SHOTerm(log_S0=np.log(S0),
log_Q=np.log(Q),
log_omega0=np.log(w0),
bounds=bounds)
kernel.freeze_parameter('log_Q')
# A periodic component
Q = 1.0
w0 = 2 * np.pi / period_guess
S0 = np.var(y_gp) / (w0 * Q)
kernel += terms.SHOTerm(log_S0=np.log(S0),
log_Q=np.log(Q),
log_omega0=np.log(w0),
bounds=bounds)
# Now calculate the covariance matrix using the initial
# kernel parameters
gp = celerite.GP(kernel, mean=np.mean(y_gp))
gp.compute(x_gp, yerr_gp)
def neg_log_like(params, y, gp, m):
gp.set_parameter_vector(params)
return -gp.log_likelihood(y[m])
def grad_neg_log_like(params, y, gp, m):
gp.set_parameter_vector(params)
return -gp.grad_log_likelihood(y[m])[1]
bounds = gp.get_parameter_bounds()
initial_params = gp.get_parameter_vector()
if ax:
ax.plot(x_gp, y_gp)
# Find the best fit kernel parameters. We want to try to ignore the flares
# when we do the fit. To do this, we will repeatedly find the best fit
# solution to the kernel model, calculate the covariance matrix, predict
# the flux and then mask out points based on how far they deviate from
# the model. After a few passes, this should cause the model to fit mostly
# to periodic features.
m = np.ones(len(x_gp), dtype=bool)
for i in range(num_iter):
n_pts_prev = np.sum(m)
gp.compute(x_gp[m], yerr_gp[m])
soln = minimize(neg_log_like,
initial_params,
jac=grad_neg_log_like,
method='L-BFGS-B',
bounds=bounds,
args=(y_gp, gp, m))
gp.set_parameter_vector(soln.x)
initial_params = soln.x
mu, var = gp.predict(y_gp[m], x_gp, return_var=True)
sig = np.sqrt(var + yerr_gp**2)
if ax:
ax.plot(x_gp, mu)
m0 = y_gp - mu < sig
m[m == 1] = m0[m == 1]
n_pts = np.sum(m)
print(n_pts, n_pts_prev)
if n_pts <= 10:
raise ValueError('GP iteration threw out too many points')
break
if (n_pts == n_pts_prev):
break
gp.compute(x_gp[m], yerr_gp[m])
mu, var = gp.predict(y_gp[m], x_gp, return_var=True)
return x_gp, mu, var, gp.get_parameter_vector()
final = TransitModel(lgror, lgroa, lgt0, lgper, lgb).get_value(time)
phase = fold(time, np.exp(lgper), np.exp(lgt0), 0.5) - 0.5
idx = np.argsort(phase)
################################################
###############~FUNKY PLOTS~####################
################################################
#set the GP parameters-FROM SAM RAW
#First granulation
Q = 1.0 / np.sqrt(2.0)
w0 = muhz2omega(13)
S0 = np.var(lc) / (w0 * Q)
kernel = terms.SHOTerm(log_S0=np.log(S0),
log_Q=np.log(Q),
log_omega0=np.log(w0))
kernel.freeze_parameter("log_Q") #to make it a Harvey model
#Second granulation
Q = 1.0 / np.sqrt(2.0)
w0 = muhz2omega(35)
S0 = np.var(lc) / (w0 * Q)
kernel += terms.SHOTerm(log_S0=np.log(S0),
log_Q=np.log(Q),
log_omega0=np.log(w0))
kernel.freeze_parameter("terms[1]:log_Q") #to make it a Harvey model
#numax
Q = np.exp(3.0)
w0 = muhz2omega(135) #peak of oscillations at 133 uhz
"N": terms.Term(),
"B": terms.RealTerm(log_a=-6., log_c=-np.inf,
bounds={"log_a": [-30, 30],
"log_c": [-np.inf, np.inf]}),
"W": terms.JitterTerm(log_sigma=-25,
bounds={"log_sigma": [-30, 30]}),
"Mat32": terms.Matern32Term(
log_sigma=1.,
log_rho=1.,
bounds={"log_sigma": [-30, 30],
# The `celerite` version of the Matern-3/2
# kernel has problems with very large `log_rho`
# values. -7.4 is empirical.
"log_rho": [-7.4, 16]}),
"SHO0": terms.SHOTerm(log_S0=-6, log_Q=1.0 / np.sqrt(2.), log_omega0=0.,
bounds={"log_S0": [-30, 30],
"log_Q": [-30, 30],
"log_omega0": [-30, 30]}),
"SHO1": terms.SHOTerm(log_S0=-6, log_Q=-2., log_omega0=0.,
bounds={"log_S0": [-10, 10],
"log_omega0": [-10, 10]}),
"SHO2": terms.SHOTerm(log_S0=-6, log_Q=0.5, log_omega0=0.,
bounds={"log_S0": [-10, 10],
"log_Q": [-10, 10],
"log_omega0": [-10, 10]}),
# see Foreman-Mackey et al. 2017, AJ 154, 6, pp. 220
# doi: 10.3847/1538-3881/aa9332
# Eq. (53)
"SHO3": terms.SHOTerm(log_S0=-6, log_Q=0., log_omega0=0.,
bounds={"log_S0": [-15, 5],
"log_Q": [-10, 10],
"log_omega0": [-10, 10]}) *
x = obs_time
y = obs_flux
yerr = obs_err
Q = 1.0 / np.sqrt(2.0)
log_w0 = 5 #3.0
log_S0 = 10
log_cadence_min = None # np.log(2*np.pi/(2./24))
log_cadence_max = np.log(2*np.pi/(0.25/24))
bounds = dict(log_S0=(-15, 30), log_Q=(-15, 15),
log_omega0=(log_cadence_min, log_cadence_max))
kernel = terms.SHOTerm(log_S0=log_S0, log_Q=np.log(Q),
log_omega0=log_w0, bounds=bounds)
kernel.freeze_parameter("log_Q") # We don't want to fit for "Q" in this term
gp = celerite.GP(kernel, mean=mean_model, fit_mean=True)
gp.compute(x, yerr)
print("Initial log-likelihood: {0}".format(gp.log_likelihood(y)))
# Define a cost function
def neg_log_like(params, y, gp):
gp.set_parameter_vector(params)
return -gp.log_likelihood(y)
def grad_neg_log_like(params, y, gp):
gp.set_parameter_vector(params)
return -gp.grad_log_likelihood(y)[1]
from __future__ import division, print_function
import emcee
import numpy as np
import matplotlib.pyplot as plt
from scipy.optimize import minimize
import celerite
from celerite import terms
from celerite import plot_setup
np.random.seed(42)
plot_setup.setup(auto=True)
# Simulate some data
kernel = terms.SHOTerm(log_S0=0.0, log_omega0=2.0, log_Q=2.0,
bounds=[(-10, 10), (-10, 10), (-10, 10)])
gp = celerite.GP(kernel)
true_params = np.array(gp.get_parameter_vector())
omega = 2*np.pi*np.exp(np.linspace(-np.log(10.0), -np.log(0.1), 5000))
true_psd = gp.kernel.get_psd(omega)
N = 200
t = np.sort(np.random.uniform(0, 10, N))
yerr = 2.5
y = gp.sample(t, diag=yerr**2)
# Find the maximum likelihood model
gp.compute(t, yerr)
def nll(params, gp, y):
gp.set_parameter_vector(params)
if not np.isfinite(gp.log_prior()):
