def __init__(self, x, y, n_bg_coef, wave_err=0.05, # MAGIC NUMBER: wavelength error hack
log_sigma0=0., log_rho0=np.log(10.), # initial params for GP
x_shift=None):
self.x = np.array(x)
self.y = np.array(y)
if n_bg_coef >= 2:
a_kw = dict([('a{}'.format(i),0.) for i in range(2,n_bg_coef)])
# estimate background
a_kw['a1'] = (y[-1]-y[0])/(x[-1]-x[0]) # slope
a_kw['a0'] = y[-1] - a_kw['a1']*x[-1] # estimate constant term
else:
a_kw = dict(a0=np.mean([y[0], y[-1]]))
# initialize model
self.mean_model = MeanModel(n_bg_coef=n_bg_coef, **a_kw)
self.kernel = terms.Matern32Term(log_sigma=log_sigma0, log_rho=log_rho0)
# set up the gp
self.gp = GP(self.kernel, mean=self.mean_model, fit_mean=True)
self.gp.compute(x, yerr=wave_err)
logger.debug("Initial log-likelihood: {0}"
.format(self.gp.log_likelihood(y)))
if x_shift is None:
self.x_shift = 0.
else:
self.x_shift = x_shift
def Pred1_2D(par, x2d, y2d, y2derr, doPlot=True, x2dpred=None):
K = x2d.shape[0]
k = terms.Matern32Term(log_sigma=par[-2], log_rho=par[-1])
gp = GP(k, mean=1.0)
shifts = np.append(0, par[:K - 1])
x1d = (x2d + shifts[:, None]).flatten()
inds = np.argsort(x1d)
y1d = y2d.flatten()
y1derr = y2derr.flatten()
gp.compute(x1d[inds], yerr=y1derr[inds])
if x2dpred is None:
x2dpred = np.copy(x2d)
x2dpreds = (x2dpred + shifts[:, None])
y2dpred = np.zeros_like(x2dpred)
y2dprederr = np.zeros_like(x2dpred)
for k in range(K):
print 'Prediction for spectrum %d' % (k + 1)
x1dpred = x2dpreds[k, :].flatten()
indspred = np.argsort(x1dpred)
mu, var = gp.predict(y1d[inds], x1dpred[indspred], return_var=True)
std = np.sqrt(var)
y2dpred[k, indspred] = mu
y2dprederr[k, indspred] = std
if doPlot:
for i in range(K):
plt.errorbar(x2d[i, :],
y2d[i, :] - i,
yerr=y2derr[i, :],
fmt=".k",
capsize=0,
alpha=0.5)
plt.plot(x2dpred[i, :], y2dpred[i, :] - i, 'C0')
plt.fill_between(x2dpred[i,:], y2dpred[i,:] + y2dprederr[i,:] - i, \
y2dpred[i,:] - y2dprederr[i,:] - i, color = 'C0', alpha = 0.4, lw = 0)
return x2dpred, y2dpred, y2dprederr
def __call__(self, rho: Optional[float] = 2., niter: int = 5):
logger.info("Running Celerite Matern 3/2 detrending")
time = self.ts.time.copy()
flux = self.ts.flux.copy()
mask = ones(time.size, bool)
for i in range(niter):
time_learn = time[mask]
flux_learn = flux[mask]
wn = mad_std(diff(flux_learn)) / sqrt(2)
log_sigma = log(mad_std(flux_learn))
log_rho = log(rho)
kernel = Matern32Term(log_sigma, log_rho)
gp = GP(kernel, mean=1.0)
gp.compute(time_learn, yerr=wn)
self.prediction = gp.predict(flux_learn, time, return_cov=False)
mask &= ~sigma_clip(flux - self.prediction, sigma=3).mask
residuals = flux - self.prediction
self.mask = m = ~(sigma_clip(residuals, sigma_lower=inf,
sigma_upper=4).mask)
self.ts._data.update('celerite_m32', time[m],
(flux - self.prediction + 1)[m], self.ts.ferr[m])
def __init__(self, lpf, name: str = 'gp', noise_ids=None, fixed_hps=None):
if not with_celerite:
raise ImportError("CeleriteLogLikelihood requires celerite.")
self.name = name
self.lpf = lpf
if fixed_hps is None:
self.free = True
else:
self.hps = asarray(fixed_hps)
self.free = False
if lpf.lcids is None:
raise ValueError('The LPF data needs to be initialised before initialising CeleriteLogLikelihood.')
self.noise_ids = noise_ids if noise_ids is not None else unique(lpf.noise_ids)
self.mask = m = zeros(lpf.lcids.size, bool)
for lcid,nid in enumerate(lpf.noise_ids):
if nid in self.noise_ids:
m[lcid == lpf.lcids] = 1
if m.sum() == lpf.lcids.size:
self.times = lpf.timea
self.fluxes = lpf.ofluxa
else:
self.times = lpf.timea[m]
self.fluxes = lpf.ofluxa[m]
self.gp = GP(Matern32Term(0, 0))
if self.free:
self.init_parameters()
else:
self.compute_gp(None, force=True, hps=self.hps)
def init_gp(self,
log_sigma=None,
log_rho=None,
amp=None,
x0=None,
bg=None,
bg_buffer=None,
**kwargs):
"""
**kwargs:
"""
# Call the different get_init methods with the correct kwargs passed
sig = inspect.signature(self.get_init)
kw = OrderedDict()
for k in list(sig.parameters.keys()):
kw[k] = kwargs.pop(k, None)
p0 = self.get_init(**kw)
# Call the generic init method - all line models must have these params
p0_generic = self.get_init_generic(amp=amp,
x0=x0,
bg=bg,
bg_buffer=bg_buffer)
for k, v in p0_generic.items():
p0[k] = v
# expand bg parameters
bgs = p0.pop('bg_coef')
for i in range(self.n_bg_coef):
p0['bg{}'.format(i)] = bgs[i]
# transform
p0 = self.transform_pars(p0)
# initialize model
mean_model = self.MeanModel(n_bg_coef=self.n_bg_coef,
absorp_emiss=self.absorp_emiss,
**p0)
p0_gp = self.get_init_gp(log_sigma=log_sigma, log_rho=log_rho)
kernel = terms.Matern32Term(log_sigma=p0_gp['log_sigma'],
log_rho=p0_gp['log_rho'])
# set up the gp model
self.gp = GP(kernel, mean=mean_model, fit_mean=True)
if self._err is not None:
self.gp.compute(self.x, self._err)
else:
self.gp.compute(self.x)
init_params = self.gp.get_parameter_vector()
init_ll = self.gp.log_likelihood(self.flux)
logger.log(0, "Initial log-likelihood: {0}".format(init_ll))
return init_params
def gpSimFull(carmaTerm, SNR, duration, N, nLC=1):
"""Simulate full CARMA time series.
Args:
carmaTerm (object): celerite GP term.
SNR (float): Signal to noise ratio defined as ratio between
CARMA amplitude and the mode of the errors (simulated using
log normal).
duration (float): The duration of the simulated time series in days.
N (int): The number of data points.
nLC (int, optional): Number of light curves to simulate. Defaults to 1.
Raises:
RuntimeError: If the input CARMA term/model is not stable, thus cannot be
solved by celerite.
Returns:
Arrays: t, y and yerr of the simulated light curves in numpy arrays.
Note that errors have been added to y.
"""
assert isinstance(
carmaTerm,
celerite.celerite.terms.Term), "carmaTerm must a celerite GP term"
if (not isinstance(carmaTerm,
DRW_term)) and (carmaTerm._arroots.real > 0).any():
raise RuntimeError(
"The covariance matrix of the provided CARMA term is not positive definite!"
)
t = np.linspace(0, duration, N)
noise = carmaTerm.get_rms_amp() / SNR
yerr = np.random.lognormal(0, 0.25, N) * noise
yerr = yerr[np.argsort(np.abs(yerr))]
# init GP and solve matrix
gp_sim = GP(carmaTerm)
gp_sim.compute(t)
# simulate, assign yerr based on y
t = np.repeat(t[None, :], nLC, axis=0)
y = gp_sim.sample(size=nLC)
# format yerr making it heteroscedastic
y_rank = y.argsort(axis=1).argsort(axis=1)
yerr = np.repeat(yerr[None, :], nLC, axis=0)
yerr = np.array(list(map(lambda x, y: x[y], yerr, y_rank)))
yerr_sign = np.random.binomial(1, 0.5, yerr.shape)
yerr_sign[yerr_sign < 1] = -1
yerr = yerr * yerr_sign
if nLC == 1:
return t[0], y[0] + yerr[0], yerr[0]
else:
return t, y + yerr, yerr
def Fit0_Jitter(x, y, yerr = None, verbose = True, doPlot = False, \
xpred = None):
k = terms.Matern32Term(log_sigma=0.0, log_rho=0.0)
if yerr is None:
wn = np.median(abs(np.diff(y)))
else:
wn = np.median(yerr)
k += terms.JitterTerm(log_sigma=np.log(wn))
gp = GP(k, mean=1.0)
gp.compute(x)
HP_init = gp.get_parameter_vector()
soln = minimize(NLL0, gp.get_parameter_vector(), jac=True, args=(gp, y))
gp.set_parameter_vector(soln.x)
if verbose:
print 'Initial pars:', HP_init
print 'Fitted pars:', soln.x
if xpred is None:
return soln.x
mu, var = gp.predict(y, xpred, return_var=True)
std = np.sqrt(var)
if doPlot:
plt.errorbar(x, y, yerr=yerr, fmt=".k", capsize=0)
plt.plot(xpred, mu, 'C0')
plt.fill_between(xpred,
mu + std,
mu - std,
color='C0',
alpha=0.4,
lw=0)
return soln.x, mu, std
def drw_fit(t, y, yerr, debug=False, user_bounds=None, n_iter=10):
"""
Fit time series to DRW.
Args:
t (array(float)): Time stamps of the input time series (the default unit is day).
y (array(float)): y values of the input time series.
yerr (array(float)): Measurement errors for y values.
debug (bool, optional): Turn on/off debug mode. Defaults to False.
user_bounds (list, optional): Parameter boundaries for the optimizer.
Defaults to None.
n_iter (int, optional): Number of iterations to run the optimizer. Defaults to 10.
Raises:
celerite.solver.LinAlgError: For non-positive definite autocovariance matrices.
Returns:
array(float): Best-fit parameters
"""
best_fit = np.empty(2)
std = np.std(y)
# init bounds for fitting
if user_bounds is not None and (len(user_bounds) == 2):
bounds = user_bounds
else:
bounds = [(-4, np.log(4 * std)), (-4, 10)]
# re-position lc
t = t - t[0]
y = y - np.median(y)
# initialize parameter and kernel
kernel = DRW_term(*drw_log_param_init(std, max_tau=np.log(t[-1] / 8)))
gp = GP(kernel, mean=0)
gp.compute(t, yerr)
best_fit_return = _min_opt(
y,
best_fit,
gp,
lambda: drw_log_param_init(std, size=n_iter, max_tau=np.log(t[-1] / 8)
),
"param",
debug,
bounds,
n_iter,
)
return best_fit_return
def test_pickle(with_general, seed=42):
solver = celerite.CholeskySolver()
np.random.seed(seed)
t = np.sort(np.random.rand(500))
diag = np.random.uniform(0.1, 0.5, len(t))
y = np.sin(t)
if with_general:
U = np.vander(t - np.mean(t), 4).T
V = U * np.random.rand(4)[:, None]
A = np.sum(U * V, axis=0) + 1e-8
else:
A = np.empty(0)
U = np.empty((0, 0))
V = np.empty((0, 0))
alpha_real = np.array([1.3, 1.5])
beta_real = np.array([0.5, 0.2])
alpha_complex_real = np.array([1.0])
alpha_complex_imag = np.array([0.1])
beta_complex_real = np.array([1.0])
beta_complex_imag = np.array([1.0])
def compare(solver1, solver2):
assert solver1.computed() == solver2.computed()
if not solver1.computed():
return
assert np.allclose(solver1.log_determinant(),
solver2.log_determinant())
assert np.allclose(solver1.dot_solve(y), solver2.dot_solve(y))
s = pickle.dumps(solver, -1)
solver2 = pickle.loads(s)
compare(solver, solver2)
solver.compute(0.0, alpha_real, beta_real, alpha_complex_real,
alpha_complex_imag, beta_complex_real, beta_complex_imag, A,
U, V, t, diag)
solver2 = pickle.loads(pickle.dumps(solver, -1))
compare(solver, solver2)
# Test that models can be pickled too.
kernel = terms.RealTerm(0.5, 0.1)
kernel += terms.ComplexTerm(0.6, 0.7, 1.0)
gp1 = GP(kernel)
gp1.compute(t, diag)
s = pickle.dumps(gp1, -1)
gp2 = pickle.loads(s)
assert np.allclose(gp1.log_likelihood(y), gp2.log_likelihood(y))
def __init__(self, lpf, name: str = 'gp', lcids=None, fixed_hps=None):
if not with_celerite:
raise ImportError("MultiCeleriteLogLikelihood requires celerite.")
self.name = name
self.lpf = lpf
if fixed_hps is None:
self.free = True
else:
self.hps = asarray(fixed_hps)
self.free = False
if lpf.lcids is None:
raise ValueError(
'The LPF data needs to be initialised before initialising CeleriteLogLikelihood.'
)
self.lcids = lcids if lcids is not None else unique(lpf.lcids)
self.lcslices = [lpf.lcslices[i] for i in self.lcids]
self.nlc = len(self.lcslices)
self.times, self.fluxes, self.wns = [], [], []
for i in self.lcids:
self.times.append(lpf.times[i])
self.fluxes.append(lpf.fluxes[i])
self.wns.append(diff(lpf.fluxes[i]).std() / sqrt(2))
self.gps = [GP(Matern32Term(0, 0)) for i in range(self.nlc)]
if self.free:
self.init_parameters()
else:
self.compute_gp(None, force=True, hps=self.hps)
def Fit0(x, y, yerr, verbose = True, doPlot = False, \
xpred = None, HP_init = None):
if HP_init is None:
HP_init = np.zeros(2)
k = terms.Matern32Term(log_sigma=HP_init[0], log_rho=HP_init[1])
gp = GP(k, mean=1.0)
gp.compute(x, yerr=yerr)
soln = minimize(NLL0, HP_init, jac=True, args=(gp, y))
gp.set_parameter_vector(soln.x)
if verbose:
print 'Initial pars:', HP_init
print 'Fitted pars:', soln.x
if xpred is None:
return soln.x
mu, var = gp.predict(y, xpred, return_var=True)
std = np.sqrt(var)
if doPlot:
plt.errorbar(x, y, yerr=yerr, fmt=".k", capsize=0)
plt.plot(xpred, mu, 'C0')
plt.fill_between(xpred,
mu + std,
mu - std,
color='C0',
alpha=0.4,
lw=0)
return soln.x, mu, std
def test_build_gp(method, seed=42):
kernel = terms.RealTerm(0.5, 0.1)
kernel += terms.ComplexTerm(0.6, 0.7, 1.0)
gp = GP(kernel, method=method)
assert gp.vector_size == 5
p = gp.get_parameter_vector()
assert np.allclose(p, [0.5, 0.1, 0.6, 0.7, 1.0])
gp.set_parameter_vector([0.5, 0.8, 0.6, 0.7, 2.0])
p = gp.get_parameter_vector()
assert np.allclose(p, [0.5, 0.8, 0.6, 0.7, 2.0])
with pytest.raises(ValueError):
gp.set_parameter_vector([0.5, 0.8, -0.6])
with pytest.raises(ValueError):
gp.set_parameter_vector("face1")
def test_pickle(method, seed=42):
solver = get_solver(method)
np.random.seed(seed)
t = np.sort(np.random.rand(500))
diag = np.random.uniform(0.1, 0.5, len(t))
y = np.sin(t)
alpha_real = np.array([1.3, 1.5])
beta_real = np.array([0.5, 0.2])
alpha_complex_real = np.array([1.0])
alpha_complex_imag = np.array([0.1])
beta_complex_real = np.array([1.0])
beta_complex_imag = np.array([1.0])
def compare(solver1, solver2):
assert solver1.computed() == solver2.computed()
if not solver1.computed():
return
assert np.allclose(solver1.log_determinant(),
solver2.log_determinant())
assert np.allclose(solver1.dot_solve(y),
solver2.dot_solve(y))
s = pickle.dumps(solver, -1)
solver2 = pickle.loads(s)
compare(solver, solver2)
if method != "sparse":
solver.compute(
alpha_real, beta_real, alpha_complex_real, alpha_complex_imag,
beta_complex_real, beta_complex_imag, t, diag
)
solver2 = pickle.loads(pickle.dumps(solver, -1))
compare(solver, solver2)
# Test that models can be pickled too.
kernel = terms.RealTerm(0.5, 0.1)
kernel += terms.ComplexTerm(0.6, 0.7, 1.0)
gp1 = GP(kernel, method=method)
gp1.compute(t, diag)
s = pickle.dumps(gp1, -1)
gp2 = pickle.loads(s)
assert np.allclose(gp1.log_likelihood(y), gp2.log_likelihood(y))
def Fit1(x2d, y2d, y2derr, par_in=None, verbose=True):
K = x2d.shape[0]
if par_in is None:
par_in = np.zeros(K + 1)
k = terms.Matern32Term(log_sigma=par_in[-2], log_rho=par_in[-1])
gp = GP(k, mean=1.0)
soln = minimize(NLL1, par_in, args=(gp, x2d, y2d, y2derr))
if verbose:
print 'Initial pars:', par_in
print 'Fitted pars:', soln.x
return soln.x
def __init__(self, kernel, t, yerr=None, white_noise=1e-10):
"""
t : the independent variable where the GP is "trained"
(opt) yerr : array of observational uncertainties
(opt) white_noise : value/array added to the diagonal of the kernel matrix
"""
self.kernel = kernel
self.t = t
self.yerr = yerr
self.white_noise = white_noise
self._GP = GPcel(self.kernel)
self._GP.compute(self.t, self.yerr)
def test_nyquist_singularity(method, seed=4220):
np.random.seed(seed)
kernel = terms.ComplexTerm(1.0, np.log(1e-6), np.log(1.0))
gp = GP(kernel, method=method)
# Samples are very close to Nyquist with f = 1.0
ts = np.array([0.0, 0.5, 1.0, 1.5])
ts[1] = ts[1] + 1e-9 * np.random.randn()
ts[2] = ts[2] + 1e-8 * np.random.randn()
ts[3] = ts[3] + 1e-7 * np.random.randn()
yerr = np.random.uniform(low=0.1, high=0.2, size=len(ts))
y = np.random.randn(len(ts))
gp.compute(ts, yerr)
llgp = gp.log_likelihood(y)
K = gp.get_matrix(ts)
K[np.diag_indices_from(K)] += yerr**2.0
ll = (-0.5 * np.dot(y, np.linalg.solve(K, y)) -
0.5 * np.linalg.slogdet(K)[1] - 0.5 * len(y) * np.log(2.0 * np.pi))
assert np.allclose(ll, llgp)
def learn_gps(self):
for lcid in tqdm(self.lcids, desc='Learning GPs', leave=False):
kernel = Matern32Term(log(self.lpf.fluxes[lcid].std()), log(0.1))
gp = GP(kernel, mean=0.0)
gp.freeze_parameter('kernel:log_sigma')
def nll(x):
gp.set_parameter_vector(x)
gp.compute(self.lpf.times[lcid], self.lpf.wn[lcid])
return -gp.log_likelihood(self.lpf.fluxes[lcid] - 1.0)
res = minimize(nll, [log(1.)])
gp.set_parameter_vector(res.x)
gp.compute(self.lpf.times[lcid], self.lpf.wn[lcid])
self.gps.append(gp)
class GPModel(object):
def __init__(
self,
x,
y,
n_bg_coef,
wave_err=0.05, # MAGIC NUMBER: wavelength error hack
log_sigma0=0.,
log_rho0=np.log(10.), # initial params for GP
x_shift=None):
self.x = np.array(x)
self.y = np.array(y)
if n_bg_coef >= 2:
a_kw = dict([('a{}'.format(i), 0.) for i in range(2, n_bg_coef)])
# estimate background
a_kw['a1'] = (y[-1] - y[0]) / (x[-1] - x[0]) # slope
a_kw['a0'] = y[-1] - a_kw['a1'] * x[-1] # estimate constant term
else:
a_kw = dict(a0=np.mean([y[0], y[-1]]))
# initialize model
self.mean_model = MeanModel(n_bg_coef=n_bg_coef, **a_kw)
self.kernel = terms.Matern32Term(log_sigma=log_sigma0,
log_rho=log_rho0)
# set up the gp
self.gp = GP(self.kernel, mean=self.mean_model, fit_mean=True)
self.gp.compute(x, yerr=wave_err)
logger.log(
0, "Initial log-likelihood: {0}".format(self.gp.log_likelihood(y)))
if x_shift is None:
self.x_shift = 0.
else:
self.x_shift = x_shift
def neg_ln_like(self, params):
# minimize -log(likelihood)
self.gp.set_parameter_vector(params)
ll = self.gp.log_likelihood(self.y)
if np.isnan(ll):
return np.inf
return -ll
def __call__(self, params):
return self.neg_ln_like(params)
def test_predict(method, seed=42):
np.random.seed(seed)
x = np.sort(np.random.rand(10))
yerr = np.random.uniform(0.1, 0.5, len(x))
y = np.sin(x)
kernel = terms.RealTerm(0.1, 0.5)
for term in [(0.6, 0.7, 1.0)]:
kernel += terms.ComplexTerm(*term)
gp = GP(kernel, method=method)
gp.compute(x, yerr)
mu0, cov0 = gp.predict(y, x)
mu, cov = gp.predict(y)
assert np.allclose(mu0, mu)
assert np.allclose(cov0, cov)
def __call__(self, period: Optional[float] = None):
logger.info("Running Celerite")
assert hasattr(
self.ts, 'ls'), "Celerite step requires a prior Lomb-Scargle step"
self.period = period if period is not None else self.ts.ls.period
gp = GP(SHOTerm(log(self.ts.flux.var()), log(10),
log(2 * pi / self.period)),
mean=1.)
gp.freeze_parameter('kernel:log_omega0')
gp.compute(self.ts.time, yerr=self.ts.ferr)
def minfun(pv):
gp.set_parameter_vector(pv)
gp.compute(self.ts.time, yerr=self.ts.ferr)
return -gp.log_likelihood(self.ts.flux)
res = minimize(minfun,
gp.get_parameter_vector(),
jac=False,
method='powell')
self.result = res
self.parameters = res.x
self.prediction = gp.predict(self.ts.flux, return_cov=False)
def carma_fit(t,
y,
yerr,
p,
q,
de=True,
debug=False,
mode="coeff",
user_bounds=None,
n_iter=10):
"""Fit time series to any CARMA model.
Args:
t (object): An array of time stamps in days.
y (object): An array of y values.
yerr (object): An array of the errors in y values.
p (int): P order of a CARMA(p, q) model.
q (int): Q order of a CARMA(p, q) model.
de (bool, optional): Whether to use differential_evolution as the
optimizer. Defaults to True.
debug (bool, optional): Turn on/off debug mode. Defaults to False.
mode (str, optional): Specify which space to sample, 'param' or 'coeff'.
Defaults to 'coeff'.
user_bounds (list, optional): Factorized polynomial coefficient boundaries
for the optimizer. Defaults to None.
n_iter (int, optional): Number of iterations to run the optimizer if de==False.
Defaults to 10.
Raises:
celerite.solver.LinAlgError: For non-positive definite matrices.
Returns:
object: An array of best-fit CARMA parameters
"""
dim = int(p + q + 1)
best_fit = np.empty(dim)
# init bounds for fitting
if user_bounds is not None and (len(user_bounds) == dim):
bounds = user_bounds
elif p == 2 and q == 1:
bounds = [(-10, 13), (-14, 7), (-10, 6), (-12, 3)]
elif p == 2 and q == 0:
bounds = [(-10, 16), (-14, 16), (-13, 15)]
else:
ARbounds = [(-6, 3)] * p
MAbounds = [(-6, 1)] * (q + 1)
bounds = ARbounds + MAbounds
# re-position lc
t = t - t[0]
y = y - np.median(y)
# initialize parameter and kernel
ARpars, MApars = sample_carma(p, q)
kernel = CARMA_term(np.log(ARpars), np.log(MApars))
gp = GP(kernel, mean=np.median(y))
gp.compute(t, yerr)
if p > 2:
mode = "coeff"
if mode == "coeff":
init_func = lambda: carma_log_fcoeff_init(dim)
else:
init_func = lambda: carma_log_param_init(dim)
if de:
best_fit_return = _de_opt(
y,
best_fit,
gp,
init_func,
mode,
debug,
bounds,
)
else:
best_fit_return = _min_opt(
y,
best_fit,
gp,
init_func,
mode,
debug,
bounds,
n_iter,
)
return best_fit_return
def gpSimFull(carmaTerm, SNR, duration, N, nLC=1, log_flux=True):
"""
Simulate CARMA time series using uniform sampling.
Args:
carmaTerm (object): An EzTao CARMA kernel.
SNR (float): Signal-to-noise defined as ratio between CARMA RMS amplitude and
the median of the measurement errors (simulated using log normal).
duration (float): The duration of the simulated time series (default in days).
N (int): The number of data points in the simulated time series.
nLC (int, optional): Number of time series to simulate. Defaults to 1.
log_flux (bool): Whether the flux/y values are in astronomical magnitude.
This argument affects how errors are assigned. Defaults to True.
Raises:
RuntimeError: If the input CARMA term/model is not stable, thus cannot be
solved by celerite.
Returns:
(array(float), array(float), array(float)): Time stamps (default in day), y
values and measurement errors of the simulated time series.
"""
assert isinstance(
carmaTerm,
celerite.celerite.terms.Term), "carmaTerm must a celerite GP term"
if (not isinstance(carmaTerm,
DRW_term)) and (carmaTerm._arroots.real > 0).any():
raise RuntimeError(
"The covariance matrix of the provided CARMA term is not positive definite!"
)
t = np.linspace(0, duration, N)
noise = carmaTerm.get_rms_amp() / SNR
yerr = np.random.lognormal(0, 0.25, N) * noise
yerr = yerr[np.argsort(np.abs(yerr))] # small->large
# init GP and solve matrix
gp_sim = GP(carmaTerm)
gp_sim.compute(t)
# simulate, assign yerr based on y
t = np.repeat(t[None, :], nLC, axis=0)
y = gp_sim.sample(size=nLC)
# format yerr making it heteroscedastic
yerr = np.repeat(yerr[None, :], nLC, axis=0)
# if in mag, large value with large error; in flux, the opposite
if log_flux:
# ascending sort
y_rank = y.argsort(axis=1).argsort(axis=1)
yerr = np.array(list(map(lambda x, y: x[y], yerr, y_rank)))
else:
# descending sort
y_rank = (-y).argsort(axis=1).argsort(axis=1)
yerr = np.array(list(map(lambda x, y: x[y], yerr, y_rank)))
if nLC == 1:
return t[0], y[0], yerr[0]
else:
return t, y, yerr
def mcmc(t,
y,
yerr,
p,
q,
n_walkers=32,
burn_in=500,
n_samples=2000,
init_param=None):
"""
A simple wrapper to run quick MCMC using emcee.
Args:
t (array(float)): Time stamps of the input time series (the default unit is day).
y (array(float)): y values of the input time series.
yerr (array(float)): Measurement errors for y values.
p (int): The p order of a CARMA(p, q) model.
q (int): The q order of a CARMA(p, q) model.
n_walkers (int, optional): Number of MCMC walkers. Defaults to 32.
burn_in (int, optional): Number of burn in steps. Defaults to 500.
n_samples (int, optional): Number of MCMC steps to run. Defaults to 2000.
init_param (array(float), optional): The initial position for the MCMC walker.
Defaults to None.
Returns:
(object, array(float), array(float)): The emcee sampler object. The MCMC
flatchain (n_walkers*n_samplers, dim) and chain (n_walkers, n_samplers, dim)
in CARMA space if p > 2, otherwise empty.
"""
assert p > q, "p order must be greater than q order."
if init_param is not None and p <= 2:
assert len(init_param) == int(
p + q + 1
), "The initial parameters doesn't match the dimension of the CARMA model!"
else:
print("Searching for best-fit CARMA parameters...")
init_param = carma_fit(t, y, yerr, p, q, n_iter=200)
# set on param or fcoeff
if p > 2:
ll = lambda *args: -neg_fcoeff_ll(*args)
init_sample = CARMA_term.carma2fcoeffs(np.log(init_param[:p]),
np.log(init_param[p:]))
else:
ll = lambda *args: -neg_param_ll(*args)
init_sample = init_param
# create vectorized functions
vec_fcoeff2carma = np.vectorize(
CARMA_term.fcoeffs2carma,
excluded=[
1,
],
signature="(n)->(m),(k)",
)
# reposition ts
t = t - t[0]
y = y - np.median(y)
# init celerite kernel/GP
kernel = CARMA_term(np.log(init_param[:p]), np.log(init_param[p:]))
gp = GP(kernel, mean=0)
gp.compute(t, yerr)
# init sampler
ndim = len(init_sample)
sampler = emcee.EnsembleSampler(n_walkers, ndim, ll, args=[y, gp])
print("Running burn-in...")
p0 = np.log(init_sample) + 1e-8 * np.random.randn(n_walkers, ndim)
p0, lp, _ = sampler.run_mcmc(p0, burn_in)
print("Running production...")
sampler.reset()
sampler.run_mcmc(p0, n_samples)
carma_flatchain = np.array([])
carma_chain = np.array([])
if p > 2:
ar, ma = vec_fcoeff2carma(sampler.flatchain, p)
carma = np.log(np.hstack((ar, ma)))
carma_flatchain = carma
carma_chain = carma.reshape((n_walkers, n_samples, ndim), order="F")
return sampler, carma_flatchain, carma_chain
def carma_fit(
t,
y,
yerr,
p,
q,
init_func=None,
neg_lp_func=None,
optimizer_func=None,
n_opt=20,
user_bounds=None,
scipy_opt_kwargs={},
scipy_opt_options={},
debug=False,
):
"""
Fit an arbitrary CARMA model
The default settings are optimized for normalized LCs.
Args:
t (array(float)): Time stamps of the input time series (the default unit is day).
y (array(float)): y values of the input time series.
yerr (array(float)): Measurement errors for y values.
p (int): The p order of a CARMA(p, q) model.
q (int): The q order of a CARMA(p, q) model.
init_func (object, optional): A user-provided function to generate initial
guesses for the optimizer. Defaults to None.
neg_lp_func (object, optional): A user-provided function to compute negative
probability given an array of parameters, an array of time series values and
a celerite GP instance. Defaults to None.
optimizer_func (object, optional): A user-provided optimizer function.
Defaults to None.
n_opt (int, optional): Number of optimizers to run.
Defaults to 20.
user_bounds (array(float), optional): Parameter boundaries for the default
optimizer. If p > 2, these are boundaries for the coefficients of the
factored polynomial. Defaults to None.
scipy_opt_kwargs (dict, optional): Keyword arguments for scipy.optimize.minimize.
Defaults to {}.
scipy_opt_options (dict, optional): "options" argument for scipy.optimize.minimize.
Defaults to {}.
debug (bool, optional): Turn on/off debug mode. Defaults to False.
Raises:
celerite.solver.LinAlgError: For non-positive definite autocovariance matrices.
Returns:
array(float): Best-fit parameters
"""
# set core config
dim = int(p + q + 1)
mode = "fcoeff" if p > 2 else "param"
# init bounds for fitting
if user_bounds is not None and (len(user_bounds) == dim):
bounds = user_bounds
else:
bounds = [(-15, 15)] * dim
bounds[p:-1] = [(a[0] - 5, a[1] - 5) for a in bounds[p:-1]]
bounds[-1] = (-15, 5)
# re-position lc
t = t - t[0]
y = y - np.median(y)
y_std = mad(y) * 1.4826
y = y / y_std
yerr = yerr / y_std
# initialize parameter and kernel
ARpars, MApars = sample_carma(p, q)
kernel = CARMA_term(np.log(ARpars), np.log(MApars))
gp = GP(kernel, mean=0)
gp.compute(t, yerr)
# determine/set init func
if init_func is not None:
init = init_func
else:
init = partial(carma_log_fcoeff_init, p, q)
# determine/set negative log probability function
if neg_lp_func is None:
neg_lp = partial(neg_lp_flat, bounds=np.array(bounds), mode=mode)
else:
neg_lp = neg_lp_func
# determine/set optimizer function
if optimizer_func is None:
scipy_opt_kwargs.update({"method": "L-BFGS-B", "bounds": bounds})
opt = partial(
scipy_opt,
mode=mode,
opt_kwargs=scipy_opt_kwargs,
opt_options=scipy_opt_options,
debug=debug,
)
else:
opt = optimizer_func
# get best-fit solution & adjust MA params (multiply by y_std)
best_fit_return = opt(y, gp, init, neg_lp, n_opt)
best_fit_return[p:] = best_fit_return[p:] * y_std
return best_fit_return
def dho_fit(
t,
y,
yerr,
init_func=None,
neg_lp_func=None,
optimizer_func=None,
n_opt=20,
user_bounds=None,
scipy_opt_kwargs={},
scipy_opt_options={},
debug=False,
):
"""
Fit DHO to time series
The default settings are optimized for normalized LCs.
Args:
t (array(float)): Time stamps of the input time series (the default unit is day).
y (array(float)): y values of the input time series.
yerr (array(float)): Measurement errors for y values.
init_func (object, optional): A user-provided function to generate initial
guesses for the optimizer. Defaults to None.
neg_lp_func (object, optional): A user-provided function to compute negative
probability given an array of parameters, an array of time series values and
a celerite GP instance. Defaults to None.
optimizer_func (object, optional): A user-provided optimizer function.
Defaults to None.
n_opt (int, optional): Number of optimizers to run.. Defaults to 20.
user_bounds (list, optional): Parameter boundaries for the default optimizer and
the default flat prior. Defaults to None.
scipy_opt_kwargs (dict, optional): Keyword arguments for scipy.optimize.minimize.
Defaults to {}.
scipy_opt_options (dict, optional): "options" argument for scipy.optimize.minimize.
Defaults to {}.
debug (bool, optional): Turn on/off debug mode. Defaults to False.
Raises:
celerite.solver.LinAlgError: For non-positive definite autocovariance matrices.
Returns:
array(float): Best-fit DHO parameters
"""
# determine user defined boundaries if any
if user_bounds is not None and (len(user_bounds) == 4):
bounds = user_bounds
else:
bounds = [(-15, 15)] * 4
bounds[2:] = [(a[0] - 8, a[1] - 8) for a in bounds[2:]]
# re-position/normalize lc
t = t - t[0]
y = y - np.median(y)
y_std = mad(y) * 1.4826
y = y / y_std
yerr = yerr / y_std
# determine negative log probability function
if neg_lp_func is None:
neg_lp = partial(neg_lp_flat, bounds=np.array(bounds), mode="param")
else:
neg_lp = neg_lp_func
# initialize parameter, kernel and GP
kernel = DHO_term(*dho_log_param_init())
gp = GP(kernel, mean=0)
gp.compute(t, yerr)
# determine initialize function
if init_func is None:
init = partial(dho_log_param_init)
else:
init = init_func
# determine the optimizer function
if optimizer_func is None:
scipy_opt_kwargs.update({"method": "L-BFGS-B", "bounds": bounds})
opt = partial(
scipy_opt,
mode="param",
opt_kwargs=scipy_opt_kwargs,
opt_options=scipy_opt_options,
debug=debug,
)
else:
opt = optimizer_func
# get best-fit solution & adjust MA params (multiply by y_std)
best_fit_return = opt(y, gp, init, neg_lp, n_opt)
best_fit_return[2:] = best_fit_return[2:] * y_std
return best_fit_return
def dho_fit(t, y, yerr, de=True, debug=False, user_bounds=None, n_iter=10):
"""Fix time series to a DHO model.
Args:
t (object): An array of time stamps in days.
y (object): An array of y values.
yerr (object): An array of the errors in y values.
de (bool, optional): Whether to use differential_evolution as the
optimizer. Defaults to True.
debug (bool, optional): Turn on/off debug mode. Defaults to False.
user_bounds (list, optional): Parameter boundaries for the optimizer.
Defaults to None.
n_iter (int, optional): Number of iterations to run the optimizer if de==False.
Defaults to 10.
Raises:
celerite.solver.LinAlgError: For non-positive definite matrices.
Returns:
object: An array of best-fit parameters
"""
best_fit = np.zeros(4)
if user_bounds is not None and (len(user_bounds) == 4):
bounds = user_bounds
else:
bounds = [(-10, 13), (-14, 7), (-10, 6), (-12, 3)]
# re-position lc
t = t - t[0]
y = y - np.median(y)
# initialize parameter, kernel and GP
kernel = DHO_term(*dho_log_param_init())
gp = GP(kernel, mean=np.mean(y))
gp.compute(t, yerr)
if de:
best_fit_return = _de_opt(
y,
best_fit,
gp,
lambda: dho_log_param_init(),
"param",
debug,
bounds,
)
else:
best_fit_return = _min_opt(
y,
best_fit,
gp,
lambda: dho_log_param_init(),
"param",
debug,
bounds,
n_iter,
)
return best_fit_return
plot_setup.setup()
# Set up the dimensions of the problem
N = 2**np.arange(6, 20)
times = np.empty((len(N), 3))
times[:] = np.nan
# Simulate a "dataset"
np.random.seed(42)
t = np.sort(np.random.rand(np.max(N)))
yerr = np.random.uniform(0.1, 0.2, len(t))
y = np.sin(t)
# Set up the GP model
kernel = terms.RealTerm(1.0, 0.1) + terms.ComplexTerm(0.1, 2.0, 1.6)
gp = GP(kernel)
for i, n in enumerate(N):
times[i, 0] = benchmark("gp.compute(t[:{0}], yerr[:{0}])".format(n),
"from __main__ import gp, t, yerr")
gp.compute(t[:n], yerr[:n])
times[i, 1] = benchmark("gp.log_likelihood(y[:{0}])".format(n),
"from __main__ import gp, y")
if n <= 4096:
times[i, 2] = benchmark(
"""
C = gp.get_matrix(t[:{0}])
C[np.diag_indices_from(C)] += yerr[:{0}]**2
cho_factor(C)
class CeleriteLogLikelihood:
def __init__(self, lpf, name: str = 'gp', noise_ids=None, fixed_hps=None):
if not with_celerite:
raise ImportError("CeleriteLogLikelihood requires celerite.")
self.name = name
self.lpf = lpf
if fixed_hps is None:
self.free = True
else:
self.hps = asarray(fixed_hps)
self.free = False
if lpf.lcids is None:
raise ValueError(
'The LPF data needs to be initialised before initialising CeleriteLogLikelihood.'
)
self.noise_ids = noise_ids if noise_ids is not None else unique(
lpf.noise_ids)
self.mask = m = zeros(lpf.lcids.size, bool)
for lcid, nid in enumerate(lpf.noise_ids):
if nid in self.noise_ids:
m[lcid == lpf.lcids] = 1
if m.sum() == lpf.lcids.size:
self.times = lpf.timea
self.fluxes = lpf.ofluxa
else:
self.times = lpf.timea[m]
self.fluxes = lpf.ofluxa[m]
self.gp = GP(Matern32Term(0, 0))
if self.free:
self.init_parameters()
else:
self.compute_gp(None, force=True, hps=self.hps)
def init_parameters(self):
name = self.name
wns = log10(mad_std(diff(self.fluxes)) / sqrt(2))
pgp = [
LParameter(f'{name}_ln_out',
f'{name} ln output scale',
'',
NP(-6, 1.5),
bounds=(-inf, inf)),
LParameter(f'{name}_ln_in',
f'{name} ln input scale',
'',
UP(-8, 8),
bounds=(-inf, inf)),
LParameter(f'{name}_log10_wn',
f'{name} log10 white noise sigma',
'',
NP(wns, 0.025),
bounds=(-inf, inf))
]
self.lpf.ps.thaw()
self.lpf.ps.add_global_block(self.name, pgp)
self.lpf.ps.freeze()
self.pv_slice = self.lpf.ps.blocks[-1].slice
self.pv_start = self.lpf.ps.blocks[-1].start
setattr(self.lpf, f"_sl_{name}", self.pv_slice)
setattr(self.lpf, f"_start_{name}", self.pv_start)
def compute_gp(self, pv, force: bool = False, hps=None):
if self.free or force:
parameters = pv[self.pv_slice] if hps is None else hps
self.gp.set_parameter_vector(parameters[:-1])
self.gp.compute(self.times, yerr=10**parameters[-1])
def compute_gp_lnlikelihood(self, pv, model):
self.compute_gp(pv)
return self.gp.log_likelihood(self.fluxes - model[self.mask])
def predict_baseline(self, pv):
self.compute_gp(pv)
residuals = self.fluxes - squeeze(
self.lpf.transit_model(pv))[self.mask]
bl = zeros_like(self.lpf.timea)
bl[self.mask] = self.gp.predict(residuals,
self.times,
return_cov=False)
return 1. + bl
def __call__(self, pvp, model):
if pvp.ndim == 1:
lnlike = self.compute_gp_lnlikelihood(pvp, model)
else:
lnlike = zeros(pvp.shape[0])
for ipv, pv in enumerate(pvp):
if all(isfinite(model[ipv])):
lnlike[ipv] = self.compute_gp_lnlikelihood(pv, model[ipv])
else:
lnlike[ipv] = -inf
return lnlike
class LineFitter(object):
def __init__(self,
x,
flux,
ivar=None,
absorp_emiss=None,
n_bg_coef=2,
target_x=None):
"""
Parameters
----------
x : array_like
Dispersion parameter. Usually either pixel or wavelength.
flux : array_like
Flux array.
ivar : array_like
Inverse-variance array.
absorp_emiss : numeric [-1, 1] (optional)
If -1, absorption line. If +1, emission line. If not specified,
will try to guess.
n_bg_coef : int (optional)
The order of the polynomial used to fit the background.
1 = constant, 2 = linear, 3 = quadratic, etc.
target_x : numeric (optional)
Where we think the line of interest is.
"""
self.x = np.array(x)
self.flux = np.array(flux)
if ivar is not None:
self.ivar = np.array(ivar)
self._err = 1 / np.sqrt(self.ivar)
else:
self.ivar = np.ones_like(flux)
self._err = None
if absorp_emiss is None:
raise NotImplementedError(
"You must supply an absorp_emiss for now")
self.absorp_emiss = float(absorp_emiss)
self.target_x = target_x
self.n_bg_coef = int(n_bg_coef)
# result of fitting
self.gp = None
self.success = None
@classmethod
def transform_pars(cls, p):
new_p = OrderedDict()
for k, v in p.items():
if k in cls._log_pars:
k = 'ln_{0}'.format(k)
v = np.log(v)
new_p[k] = v
return new_p
def get_init_generic(self, amp=None, x0=None, bg=None, bg_buffer=None):
"""
Get initial guesses for the generic line parameters.
Parameters
----------
amp : numeric
Line amplitude. This should be strictly positive; the
``absorp_emiss`` parameter controls whether it is absorption or
emission.
x0 : numeric
Line centroid.
bg : iterable
Background polynomial coefficients.
bg_buffer : int
The number of values to use on either end of the flux array to
estimate the background parameters.
"""
target_x = self.target_x
if x0 is None: # estimate the initial guess for the centroid
relmins = argrelmin(-self.absorp_emiss * self.flux)[0]
if len(relmins) > 1 and target_x is None:
logger.log(
0, "no target_x specified - taking largest line "
"in spec region")
target_x = self.x[np.argmin(-self.absorp_emiss * self.flux)]
if len(relmins) == 1:
x0 = self.x[relmins[0]]
else:
x0_idx = relmins[np.abs(self.x[relmins] - target_x).argmin()]
x0 = self.x[x0_idx]
# shift x array so that line is approximately at 0
_x = np.array(self.x, copy=True)
x = np.array(_x) - x0
# background polynomial parameters
if bg_buffer is None:
bg_buffer = len(x) // 4
bg_buffer = max(1, bg_buffer)
if bg is None:
if self.n_bg_coef < 2:
bg = np.array([0.])
bg[0] = np.median(self.flux)
else:
# estimate linear background model
bg = np.array([0.] * self.n_bg_coef)
i1 = argmedian(self.flux[:bg_buffer])
i2 = argmedian(self.flux[-bg_buffer:])
f1 = self.flux[:bg_buffer][i1]
f2 = self.flux[-bg_buffer:][i2]
x1 = x[:bg_buffer][i1]
x2 = x[-bg_buffer:][i2]
bg[1] = (f2 - f1) / (x2 - x1) # slope
bg[0] = f2 - bg[1] * x2 # estimate constant term
else:
if len(bg) != self.n_bg_coef:
raise ValueError('Number of bg polynomial coefficients does '
'not match n_bg_coef specified when fitter '
'was created.')
if amp is None: # then estimate the initial guess for amplitude
_i = np.argmin(np.abs(x))
if len(bg) > 1:
_bg = bg[0] + bg[1] * x[_i]
else:
_bg = bg[0]
# amp = np.sqrt(2*np.pi) * (flux[_i] - bg)
amp = self.flux[_i] - _bg
p0 = OrderedDict()
p0['amp'] = np.abs(amp)
p0['x0'] = x0
p0['bg_coef'] = bg
return p0
def get_init_gp(self, log_sigma=None, log_rho=None, x0=None):
"""
Set up the GP kernel parameters.
"""
if log_sigma is None:
if x0 is not None:
# better initial guess for GP parameters
mask = np.abs(self.x - x0) > 2.
y_MAD = np.median(
np.abs(self.flux[mask] - np.median(self.flux[mask])))
else:
y_MAD = np.median(np.abs(self.flux - np.median(self.flux)))
log_sigma = np.log(3 * y_MAD)
if log_rho is None:
log_rho = np.log(5.)
p0 = OrderedDict()
p0['log_sigma'] = log_sigma
p0['log_rho'] = log_rho
return p0
def init_gp(self,
log_sigma=None,
log_rho=None,
amp=None,
x0=None,
bg=None,
bg_buffer=None,
**kwargs):
"""
**kwargs:
"""
# Call the different get_init methods with the correct kwargs passed
sig = inspect.signature(self.get_init)
kw = OrderedDict()
for k in list(sig.parameters.keys()):
kw[k] = kwargs.pop(k, None)
p0 = self.get_init(**kw)
# Call the generic init method - all line models must have these params
p0_generic = self.get_init_generic(amp=amp,
x0=x0,
bg=bg,
bg_buffer=bg_buffer)
for k, v in p0_generic.items():
p0[k] = v
# expand bg parameters
bgs = p0.pop('bg_coef')
for i in range(self.n_bg_coef):
p0['bg{}'.format(i)] = bgs[i]
# transform
p0 = self.transform_pars(p0)
# initialize model
mean_model = self.MeanModel(n_bg_coef=self.n_bg_coef,
absorp_emiss=self.absorp_emiss,
**p0)
p0_gp = self.get_init_gp(log_sigma=log_sigma, log_rho=log_rho)
kernel = terms.Matern32Term(log_sigma=p0_gp['log_sigma'],
log_rho=p0_gp['log_rho'])
# set up the gp model
self.gp = GP(kernel, mean=mean_model, fit_mean=True)
if self._err is not None:
self.gp.compute(self.x, self._err)
else:
self.gp.compute(self.x)
init_params = self.gp.get_parameter_vector()
init_ll = self.gp.log_likelihood(self.flux)
logger.log(0, "Initial log-likelihood: {0}".format(init_ll))
return init_params
def get_gp_mean_pars(self):
"""
Return a parameter dictionary for the mean model parameters only.
"""
fit_pars = OrderedDict()
for k, v in self.gp.get_parameter_dict().items():
if 'mean' not in k:
continue
k = k[5:] # remove 'mean:'
if k.startswith('ln'):
if 'amp' in k:
fit_pars[k[3:]] = self.absorp_emiss * np.exp(v)
else:
fit_pars[k[3:]] = np.exp(v)
elif k.startswith('bg'):
if 'bg_coef' not in fit_pars:
fit_pars['bg_coef'] = []
fit_pars['bg_coef'].append(v)
else:
fit_pars[k] = v
return fit_pars
def _neg_log_like(self, params):
self.gp.set_parameter_vector(params)
try:
ll = self.gp.log_likelihood(self.flux)
except (RuntimeError, LinAlgError):
return np.inf
if np.isnan(ll):
return np.inf
return -ll
def fit(self, bounds=None, **kwargs):
"""
"""
init_params = self.init_gp()
if bounds is None:
bounds = OrderedDict()
# default bounds for mean model parameters
i = self.gp.models['mean'].get_parameter_names().index('x0')
mean_bounds = self.gp.models['mean'].get_parameter_bounds()
if mean_bounds[i][0] is None and mean_bounds[i][1] is None:
mean_bounds[i] = (min(self.x), max(self.x))
for i, k in enumerate(self.gp.models['mean'].parameter_names):
mean_bounds[i] = bounds.get(k, mean_bounds[i])
self.gp.models['mean'].parameter_bounds = mean_bounds
# HACK: default bounds for kernel parameters
self.gp.models['kernel'].parameter_bounds = [(None, None),
(np.log(0.5), None)]
soln = minimize(self._neg_log_like,
init_params,
method="L-BFGS-B",
bounds=self.gp.get_parameter_bounds())
self.success = soln.success
self.gp.set_parameter_vector(soln.x)
if self.success:
logger.debug("Success: {0}, Final log-likelihood: {1}".format(
soln.success, -soln.fun))
else:
logger.warning("Fit failed! Final log-likelihood: {0}, "
"Final parameters: {1}".format(-soln.fun, soln.x))
return self
def plot_fit(self, axes=None, fit_alpha=0.5):
unbinned_color = '#3182bd'
binned_color = '#2ca25f'
gp_color = '#ff7f0e'
noise_color = '#de2d26'
if self.gp is None:
raise ValueError("You must run .fit() first!")
# ----------------------------------------------------------------------
# Plot the maximum likelihood model
if axes is None:
fig, axes = plt.subplots(1, 3, figsize=(15, 5), sharex=True)
# data
for ax in axes[:2]:
ax.plot(self.x,
self.flux,
drawstyle='steps-mid',
marker='',
color='#777777',
zorder=2)
if self._err is not None:
ax.errorbar(self.x,
self.flux,
self._err,
marker='',
ls='none',
ecolor='#666666',
zorder=1)
# mean model
wave_grid = np.linspace(self.x.min(), self.x.max(), 256)
mu, var = self.gp.predict(self.flux, self.x, return_var=True)
std = np.sqrt(var)
axes[0].plot(wave_grid,
self.gp.mean.get_unbinned_value(wave_grid),
marker='',
alpha=fit_alpha,
zorder=10,
color=unbinned_color)
axes[0].plot(self.x,
self.gp.mean.get_value(self.x),
marker='',
alpha=fit_alpha,
zorder=10,
drawstyle='steps-mid',
color=binned_color)
# full GP model
axes[1].plot(self.x,
mu,
color=gp_color,
drawstyle='steps-mid',
marker='',
alpha=fit_alpha,
zorder=10)
axes[1].fill_between(self.x,
mu + std,
mu - std,
color=gp_color,
alpha=fit_alpha / 10.,
edgecolor="none",
step='mid')
# just GP noise component
mean_model = self.gp.mean.get_value(self.x)
axes[2].plot(self.x,
mu - mean_model,
marker='',
alpha=fit_alpha,
zorder=10,
drawstyle='steps-mid',
color=noise_color)
axes[2].plot(self.x,
self.flux - mean_model,
drawstyle='steps-mid',
marker='',
color='#777777',
zorder=2)
if self._err is not None:
axes[2].errorbar(self.x,
self.flux - mean_model,
self._err,
marker='',
ls='none',
ecolor='#666666',
zorder=1)
axes[0].set_ylabel('flux')
axes[0].set_xlim(self.x.min(), self.x.max())
axes[0].set_title('Line model')
axes[1].set_title('Line + noise model')
axes[2].set_title('Noise model')
fig = axes[0].figure
fig.tight_layout()
return fig
def drw_fit(
t,
y,
yerr,
init_func=None,
neg_lp_func=None,
optimizer_func=None,
n_opt=10,
user_bounds=None,
scipy_opt_kwargs={},
scipy_opt_options={},
debug=False,
):
"""
Fit DRW.
Args:
t (array(float)): Time stamps of the input time series (the default unit is day).
y (array(float)): y values of the input time series.
yerr (array(float)): Measurement errors for y values.
init_func (object, optional): A user-provided function to generate initial
guesses for the optimizer. Defaults to None.
neg_lp_func (object, optional): A user-provided function to compute negative
probability given an array of parameters, an array of time series values and
a celerite GP instance. Defaults to None.
optimizer_func (object, optional): A user-provided optimizer function.
Defaults to None.
n_opt (int, optional): Number of optimizers to run. Defaults to 10.
user_bounds (list, optional): Parameter boundaries for the default optimizer.
Defaults to None.
scipy_opt_kwargs (dict, optional): Keyword arguments for scipy.optimize.minimize.
Defaults to {}.
scipy_opt_options (dict, optional): "options" argument for scipy.optimize.minimize.
Defaults to {}.
debug (bool, optional): Turn on/off debug mode. Defaults to False.
Returns:
array(float): Best-fit DRW parameters
"""
# re-position lc; compute some stat
t = t - t[0]
y = y - np.median(y)
std = np.std(y)
min_dt = np.min(t[1:] - t[:-1])
# init bounds for fitting
if user_bounds is not None and (len(user_bounds) == 2):
bounds = user_bounds
else:
bounds = [
(np.log(std / 50), np.log(3 * std)),
(np.log(min_dt / 5), np.log(t[-1])),
]
# determine negative log probability function
if neg_lp_func is None:
neg_lp = partial(neg_lp_flat, bounds=np.array(bounds), mode="param")
else:
neg_lp = neg_lp_func
# define ranges to generate initial guesses
amp_range = [std / 50, 3 * std]
log_tau_range = [np.log(min_dt / 5), np.log(t[-1] / 10)]
# initialize parameter and kernel
kernel = DRW_term(*drw_log_param_init(amp_range, log_tau_range))
gp = GP(kernel, mean=0)
gp.compute(t, yerr)
# determine initialize function
if init_func is None:
init = partial(drw_log_param_init, amp_range, log_tau_range)
else:
init = init_func
# determine optimizer function
if optimizer_func is None:
scipy_opt_kwargs.update({"method": "L-BFGS-B", "bounds": bounds})
opt = partial(
scipy_opt,
mode="param",
opt_kwargs=scipy_opt_kwargs,
opt_options=scipy_opt_options,
debug=debug,
)
else:
opt = optimizer_func
best_fit_return = opt(y, gp, init, neg_lp, n_opt)
return best_fit_return