def prepare_characterization(light_curve_file, basis_file,
periods, time0s, rors, impacts,
es=None):
# Download and process the light curves.
pipe = K2Data(cache=False)
pipe = K2Likelihood(pipe, cache=False)
query = dict(
basis_file=os.path.abspath(basis_file),
light_curve_file=os.path.abspath(light_curve_file)
)
r = pipe.query(**query)
lc = r.model_light_curves[0]
# Set up the initial system model.
star = transit.Central()
s = transit.System(star)
for i in range(len(periods)):
planet = transit.Body(r=rors[i],
period=periods[i],
t0=time0s[i] % periods[i],
b=impacts[i],
e=0.0 if es is None else es[i])
s.add_body(planet)
return ProbabilisticModel(lc, s)
def get_result(self, query, parent_response):
# Parse the arguments.
injections = query["injections"]
if not len(injections):
return dict(target_datasets=parent_response.target_datasets)
# Build the system.
q1 = query["q1"]
q2 = query["q2"]
mstar = query["mstar"]
rstar = query["rstar"]
s = transit.System(transit.Central(q1=q1, q2=q2, mass=mstar,
radius=rstar))
# Loop over injected bodies and add them to the system.
for inj in injections:
body = transit.Body(r=inj["radius"], period=inj["period"],
t0=inj["t0"], b=inj.get("b", 0.0),
e=inj.get("e", 0.0),
pomega=inj.get("pomega", 0.0))
s.add_body(body)
try:
body.ix
except ValueError:
logging.warn("Removing planet with invalid impact parameter")
s.bodies.pop()
# Inject the transit into each dataset.
results = []
for _ in parent_response.target_datasets:
lc = InjectedLightCurve(_)
lc.flux[lc.m] *= s.light_curve(lc.time[lc.m])
results.append(lc)
return dict(target_datasets=results, injected_system=s)
def __init__(self, fn, median=True):
self.t, self.f, self.fe, self.truth = load_data(fn, median)
self.ivar = 1.0 / self.fe**2
self.central = transit.Central(q1=self.truth["q1"],
q2=self.truth["q2"])
self.system = transit.System(self.central)
self.body = transit.Body(period=self.truth["period"],
r=self.truth["r"],
b=self.truth["b"],
t0=self.truth["t0"])
self.system.add_body(self.body)
def lightcurve(star, planet, t):
"""
Generate a light curve for the given values.
Star - [mass,radius] in stellar radii
Planet - [radius,mass,period,t0]
t - the array for time points from the data
"""
# Build the transiting system.
star = transit.Central(star[0], star[1])
s = transit.System(star)
#body = transit.Body(r=0.11/0.74,m=0.002375/0.74,period=1.337,to=0.99,b=0.2,e=0.0167)
body = transit.Body(r=planet[0],
mass=planet[1],
period=planet[2],
t0=planet[3],
b=0.029)
s.add_body(body)
# Compute the light curve for given values of t
f = s.light_curve(t)
return f
#lightcurve([0.74,0.713],[0.1,0.001,1.33712,1], np.arange(0,2,1e-4))
#
#star = transit.Central(0.74,0.713)
#s = transit.System(star)
#body = transit.Body(r=0.1,mass=0.01, period=1.33712, t0=0, b=0)
#s.add_body(body)
#
#t = np.arange(0, 2, 1e-4)
#
#f = s.light_curve(t)
rstar = float(cand.radius)
rstar_err1 = float(cand.radius_err1)
rstar_err2 = float(cand.radius_err2)
ln_rstar = np.log(rstar)
ln_rstar_err = np.mean([np.log(rstar + rstar_err1) - np.log(rstar),
np.log(rstar) - np.log(rstar + rstar_err2)])
mstar = float(cand.mass)
mstar_err1 = float(cand.mass_err1)
mstar_err2 = float(cand.mass_err2)
ln_mstar = np.log(mstar)
ln_mstar_err = np.mean([np.log(mstar + mstar_err1) - np.log(mstar),
np.log(mstar) - np.log(mstar + mstar_err2)])
# Set up the model.
system = transit.System(transit.Central(mass=mstar, radius=rstar))
system.add_body(transit.Body(
period=period, r=1.3*rp, b=b, t0=t0,
))
# Load the light curves.
lcs = peerless.data.load_light_curves_for_kic(kicid, min_break=10)
texp = lcs[0].texp
t = np.concatenate([lc.time for lc in lcs])
f = np.concatenate([lc.flux for lc in lcs])
fe = np.concatenate([lc.yerr + np.zeros_like(lc.time) for lc in lcs])
# Find the minimum allowed period.
m = np.isfinite(t)
min_period = max(t[m].max() - t0, t0 - t[m].min())
def prepare_characterization(kicid, periods, time0s, rors, impacts,
es=None,
data_window_hw=3.0, min_data_window_hw=0.5):
# Download and process the light curves.
pipe = Download()
pipe = Prepare(pipe)
pipe = Discontinuity(pipe)
r = pipe.query(kicid=kicid)
# Find the data chunks that hit a transit.
lcs = []
for lc in r.light_curves:
# Build the mask of times that hit transits.
m = np.zeros_like(lc.time, dtype=bool)
mmin = np.zeros_like(lc.time, dtype=bool)
for p, t0 in zip(periods, time0s):
hp = 0.5 * p
t0 = t0 % p
dt = np.abs((lc.time - t0 + hp) % p - hp)
m += dt < data_window_hw
mmin += dt < min_data_window_hw
# Trim the dataset and set up the Gaussian Process model.
if np.any(mmin) and np.sum(m) > 10:
# Re-normalize the trimmed light curve.
mu = np.median(lc.flux[m])
lc.time = np.ascontiguousarray(lc.time[m])
lc.flux = np.ascontiguousarray(lc.flux[m] / mu)
lc.ferr = np.ascontiguousarray(lc.ferr[m] / mu)
# Make sure that the light curve knows its integration time.
lc.texp = kplr.EXPOSURE_TIMES[1] / 86400.0
# Heuristically guess the Gaussian Process parameters.
lc.factor = 1000.0
amp = np.median((lc.factor * (lc.flux-1.0))**2)
kernel = amp*kernels.Matern32Kernel(4.0)
lc.gp = george.GP(kernel)
# Run an initial computation of the GP.
lc.gp.compute(lc.time, lc.ferr * lc.factor)
# Save this light curve.
lcs.append(lc)
# Set up the initial system model.
spars = r.star.huber
star = transit.Central(mass=spars.M, radius=spars.R)
s = transit.System(star)
for i in range(len(periods)):
planet = transit.Body(r=rors[i] * star.radius,
period=periods[i],
t0=time0s[i] % periods[i],
b=impacts[i],
e=0.0 if es is None else es[i])
s.add_body(planet)
# Approximate the stellar mass and radius measurements as log-normal.
q = np.array(spars[["R", "E_R", "e_R"]], dtype=float)
lnsr = (np.log(q[0]),
1.0 / np.mean([np.log(q[0] + q[1]) - np.log(q[0]),
np.log(q[0]) - np.log(q[0] - q[2])]) ** 2)
q = np.array(spars[["M", "E_M", "e_M"]], dtype=float)
lnsm = (np.log(q[0]),
1.0 / np.mean([np.log(q[0] + q[1]) - np.log(q[0]),
np.log(q[0]) - np.log(q[0] - q[2])]) ** 2)
return ProbabilisticModel(lcs, s, lnsr, lnsm)
def simulation_system(smass, srad, q1, q2, period, t0, rp, b, e, pomega):
s = transit.System(transit.Central(mass=smass, radius=srad, q1=q1, q2=q2))
s.add_body(
transit.Body(period=period, t0=t0, r=rp, b=b, e=e, pomega=pomega))
return s
def setup_fit(args, fit_kois=False, max_points=300):
kicid = args["kicid"]
# Initialize the system.
system = transit.System(
transit.Central(
flux=1.0,
radius=args["srad"],
mass=args["smass"],
q1=0.5,
q2=0.5,
))
system.add_body(
transit.Body(
radius=args["radius"],
period=args["period"],
t0=args["t0"],
b=args["impact"],
e=1.123e-7,
omega=0.0,
))
if fit_kois:
kois = KOICatalog().df
kois = kois[kois.kepid == kicid]
for _, row in kois.iterrows():
system.add_body(
transit.Body(
radius=float(row.koi_ror) * args["srad"],
period=float(row.koi_period),
t0=float(row.koi_time0bk) % float(row.koi_period),
b=float(row.koi_impact),
e=1.123e-7,
omega=0.0,
))
system.thaw_parameter("*")
system.freeze_parameter("bodies*ln_mass")
# Load the light curves.
lcs, _ = load_light_curves_for_kic(kicid, remove_kois=not fit_kois)
# Which light curves should be fit?
fit_lcs = []
other_lcs = []
gps = []
for lc in lcs:
f = system.light_curve(lc.time, lc.texp)
if np.any(f < 1.0):
i = np.argmin(f)
inds = np.arange(len(f))
m = np.zeros(len(lc.time), dtype=bool)
m[np.sort(np.argsort(np.abs(inds - i))[:max_points])] = True
if np.any(f[~m] < system.central.flux):
m = np.ones(len(lc.time), dtype=bool)
lc.time = np.ascontiguousarray(lc.time[m])
lc.flux = np.ascontiguousarray(lc.flux[m])
lc.ferr = np.ascontiguousarray(lc.ferr[m])
fit_lcs.append(lc)
var = np.median((lc.flux - np.median(lc.flux))**2)
kernel = 1e6 * var * kernels.Matern32Kernel(2**2)
gp = george.GP(kernel,
white_noise=2 * np.log(np.mean(lc.ferr) * 1e3),
fit_white_noise=True)
gp.compute(lc.time, lc.ferr * 1e3)
gps.append(gp)
else:
other_lcs.append(lc)
model = TransitModel(args, gps, system, args["smass"], args["smass_err"],
args["srad"], args["srad_err"], fit_lcs, other_lcs)
return model
def search(kicid_and_injection=None,
lcs=None,
tau=0.6,
detrend_hw=2.0,
remove_kois=True,
grid_frac=0.25,
noise_hw=15.0,
detect_thresh=25.0,
max_fit_data=500,
max_peaks=3,
min_datapoints=10,
all_models=False,
verbose=False):
"""
:param tau:
The transit duration. (default: 0.6)
:param detrend_hw:
Half width of running window for de-trending. (default: 2.0)
:param remove_kois:
Remove data points near known KOI transits. (default: True)
:param grid_frac:
The search grid spacing as a fraction of the duration. (default: 0.25)
:param noise_hw:
Half width of running window for noise estimation. (default: 15.0)
:param detect_thresh:
Relative S/N detection threshold. (default: 25.0)
:param max_fit_data:
The maximum number of data points for fitting. (default: 500)
:param max_peaks:
The maximum number of peaks to analyze in detail. (default: 3)
:param min_datapoints:
The minimum number of in-transit data points. (default: 10)
:param verbose:
Moar printing. (default: False)
"""
if kicid_and_injection is not None:
kicid, injection = kicid_and_injection
else:
kicid, injection = None, None
inject = injection is not None
system = None
if inject:
system = transit.System(
transit.Central(q1=injection["q1"], q2=injection["q2"]))
system.add_body(
transit.Body(
radius=injection["ror"],
period=injection["period"],
b=injection["b"],
e=injection["e"],
omega=injection["omega"],
t0=injection["t0"],
))
injection["ncadences"] = 0
injection["recovered"] = False
if lcs is None and kicid is None:
raise ValueError("you must specify 'lcs' or 'kicid'")
if lcs is None:
lcs, ncad = load_light_curves_for_kic(kicid,
detrend_hw=detrend_hw,
remove_kois=remove_kois,
inject_system=system)
if inject:
injection["ncadences"] += ncad
else:
kicid = "unknown-target"
# Loop over light curves and search each one.
time = []
chunk = []
depth = []
depth_ivar = []
s2n = []
for i, lc in enumerate(lcs):
time.append(np.arange(lc.time.min(), lc.time.max(), grid_frac * tau))
d, ivar, s = cython_search(tau, time[-1], lc.time, lc.flux - 1,
1 / lc.ferr**2)
depth.append(d)
depth_ivar.append(ivar)
s2n.append(s)
chunk.append(i + np.zeros(len(time[-1]), dtype=int))
time = np.concatenate(time)
chunk = np.concatenate(chunk)
depth = np.concatenate(depth)
depth_ivar = np.concatenate(depth_ivar)
s2n = np.concatenate(s2n)
# Compute the depth S/N time series and smooth it to estimate a background
# noise level.
m = depth_ivar > 0.0
noise = np.nan + np.zeros_like(s2n)
noise[m] = running_median_trend(time[m], np.abs(s2n[m]), noise_hw)
results = SearchResults(kicid,
lcs,
tau,
detect_thresh,
time,
depth,
depth_ivar,
s2n,
noise,
injection=injection)
# Find peaks above the fiducial threshold.
m = s2n > detect_thresh * noise
peaks = []
while np.any(m):
i = np.argmax(s2n[m])
t0 = time[m][i]
peaks.append(
dict(
kicid=kicid,
t0=t0 + 0.5 * tau,
s2n=s2n[m][i],
bkg=noise[m][i],
depth=depth[m][i],
depth_ivar=depth_ivar[m][i],
chunk=chunk[m][i],
))
m &= np.abs(time - t0) > 2 * tau
if verbose:
print("Found {0} raw peaks".format(len(peaks)))
if not len(peaks):
return results
for i, peak in enumerate(peaks):
peak["num_peaks"] = len(peaks)
peak["peak_id"] = i
if verbose and len(peaks) > max_peaks:
logging.warning("truncating peak list")
peaks = peaks[:max_peaks]
# For each peak, plot the diagnostic plots and vet.
for i, peak in enumerate(peaks):
# Vetting.
t0 = peak["t0"]
d = peak["depth"]
chunk = peak["chunk"]
lc0 = lcs[chunk]
x = lc0.raw_time
y = lc0.raw_flux
yerr = lc0.raw_ferr
ndata = np.sum(np.abs(x - t0) < 0.5 * tau)
if ndata < min_datapoints:
if verbose:
logging.warning(
"there are only {0} data points in transit".format(ndata))
continue
# Limit number of data points in chunk.
inds = np.sort(np.argsort(np.abs(t0 - x))[:max_fit_data])
x = np.ascontiguousarray(x[inds])
y = np.ascontiguousarray(y[inds])
yerr = np.ascontiguousarray(yerr[inds])
cen_x = np.ascontiguousarray(lc0.mom_cen_1[inds])
cen_y = np.ascontiguousarray(lc0.mom_cen_2[inds])
peak["data"] = (x, y, yerr, cen_x, cen_y)
if verbose:
print("{0} data points in chunk".format(len(x)))
for k in [
"channel", "skygroup", "module", "output", "quarter", "season"
]:
peak[k] = lc0.meta[k]
peak["chunk_min_time"] = x.min()
peak["chunk_max_time"] = x.max()
# Mean models:
# 1. constant
constant = np.median(y)
# 2. transit
m = np.abs(x - t0) < tau
ind = np.arange(len(x))[m][np.argmin(y[m])]
ror = np.sqrt(max(d, 1.0 - y[ind]))
system = transit.SimpleSystem(
period=3000.0,
t0=x[ind],
ror=ror,
duration=tau,
impact=0.5,
)
system.freeze_parameter("ln_period")
system.freeze_parameter("q1_param")
system.freeze_parameter("q2_param")
best = (np.inf, system.t0, 0.0, ror)
for dur in np.linspace(0.1 * tau, 2 * tau, 50):
system.duration = dur
for t0 in x[ind] + np.linspace(-0.1 * tau, 0.1 * tau, 7):
system.t0 = t0
system.ror = ror
mu = system.get_value(x)
A = np.concatenate((np.vander(x, 2).T, [mu]), axis=0).T
try:
w = np.linalg.solve(np.dot(A.T, A), np.dot(A.T, y))
except np.linalg.LinAlgError:
continue
d = np.sum((y - np.dot(A, w))**2)
if d < best[0]:
best = (d, t0, dur, np.sqrt(w[-1]) * ror)
system.t0 = best[1]
system.duration = best[2]
system.ror = best[3]
# 3. step
best = (np.inf, 0, 0.0, 0.0)
n = 2
for ind in range(1, len(y) - 2 * n):
a = slice(ind, ind + n)
b = slice(ind + n, ind + 2 * n)
step = StepModel(
value1=np.mean(y[:ind]),
value2=np.mean(y[ind + 2 * n:]),
height1=np.mean(y[a]) - np.mean(y[:ind]),
height2=np.mean(y[ind + 2 * n:]) - np.mean(y[b]),
log_width_plus=0.0,
log_width_minus=0.0,
t0=0.5 * (x[ind + n - 1] + x[ind + n]),
)
best_minus = (np.inf, 0.0)
m = x < step.t0
for w in np.linspace(-4, 2, 20):
step.log_width_minus = w
d = np.sum((y[m] - step.get_value(x[m]))**2)
if d < best_minus[0]:
best_minus = (d, w)
best_plus = (np.inf, 0.0)
m = x >= step.t0
for w in np.linspace(-4, 2, 20):
step.log_width_plus = w
d = np.sum((y[m] - step.get_value(x[m]))**2)
if d < best_plus[0]:
best_plus = (d, w)
d = best_minus[0] + best_plus[0]
if d < best[0]:
best = (d, ind, best_minus[1], best_plus[1])
_, ind, wm, wp = best
a = slice(ind, ind + n)
b = slice(ind + n, ind + 2 * n)
step = StepModel(
value1=np.mean(y[:ind]),
value2=np.mean(y[ind + 2 * n:]),
height1=np.mean(y[a]) - np.mean(y[:ind]),
height2=np.mean(y[ind + 2 * n:]) - np.mean(y[b]),
log_width_plus=wp,
log_width_minus=wm,
t0=0.5 * (x[ind + n - 1] + x[ind + n]),
)
check_gradient(step, x)
# 4. box:
inds = np.argsort(np.diff(y))
inds = np.sort([inds[0], inds[-1]])
boxes = []
for tmn, tmx in (0.5 * (x[inds] + x[inds + 1]), (t0 - 0.5 * tau,
t0 + 0.5 * tau)):
boxes.append(BoxModel(tmn, tmx, data=(x, y)))
check_gradient(boxes[-1], x)
# # 5. vee
# vee = VeeModel(depth=1., t0=system.t0, log_b=-0.5,
# log_a=np.log(0.1/24.))
# check_gradient(vee, x)
# Loop over models and compare them.
models = OrderedDict([
("transit", system),
# ("vee", vee),
("box1", boxes[1]),
("step", step),
("gp", constant),
("box2", boxes[0]),
])
peak["gps"] = OrderedDict()
peak["pred_cens"] = []
for name, mean_model in models.items():
kernel = np.var(y) * kernels.Matern32Kernel(2**2)
gp = george.GP(kernel,
mean=mean_model,
fit_mean=True,
white_noise=2 * np.log(np.mean(yerr)),
fit_white_noise=True)
gp.compute(x, yerr)
# Set up some bounds.
bounds = gp.get_bounds()
n = gp.get_parameter_names()
if name == "transit":
bounds[n.index("mean:ln_duration")] = np.log(
(system.duration / 2.0, system.duration * 2.0))
bounds[n.index("mean:t0")] = (t0 - 0.5 * tau, t0 + 0.5 * tau)
if "mean:impact" in n:
bounds[n.index("mean:impact")] = (0, 1.0)
if "mean:q1_param" in n:
bounds[n.index("mean:q1_param")] = (-10, 10)
bounds[n.index("mean:q2_param")] = (-10, 10)
bounds[n.index("kernel:k2:ln_M_0_0")] = (np.log(0.1), None)
bounds[n.index("white:value")] = (2 *
np.log(0.5 * np.median(yerr)),
None)
bounds[n.index("kernel:k1:ln_constant")] = \
(2*np.log(np.median(yerr)), None)
# Optimize.
initial_vector = np.array(gp.get_vector())
r = minimize(gp.nll,
gp.get_vector(),
jac=gp.grad_nll,
args=(y, ),
method="L-BFGS-B",
bounds=bounds)
gp.set_vector(r.x)
peak["gps"][name] = (gp, y)
# Compute the -0.5*BIC.
peak["lnlike_{0}".format(name)] = -r.fun
peak["bic_{0}".format(
name)] = -r.fun - 0.5 * len(r.x) * np.log(len(x))
if verbose:
print("Peak {0}:".format(i))
print("Model: '{0}'".format(name))
print("Converged: {0}".format(r.success))
if not r.success:
print("Message: {0}".format(r.message))
print("Log-likelihood: {0}".format(
peak["lnlike_{0}".format(name)]))
print("-0.5 * BIC: {0}".format(peak["bic_{0}".format(name)]))
print("Parameters:")
for k, v0, v in zip(gp.get_parameter_names(), initial_vector,
gp.get_vector()):
print(" {0}: {1:.4f} -> {2:.4f}".format(k, v0, v))
print()
# Initialize one of the boxes using the transit shape.
if name == "transit" and np.any(system.get_value(x) < 1.0):
depth = 1.0 - float(system.get_value(system.t0))
models["box1"].mn = system.t0 - 0.5 * system.duration
models["box1"].mx = system.t0 + 0.5 * system.duration
# models["vee"].t0 = system.t0
# models["vee"].log_b = np.log(0.5*system.duration)
# models["vee"].depth = depth
# Fit the centroids.
tm = (1.0 - system.get_value(x)) / depth
A = np.vander(tm, 2)
AT = A.T
offset = 0.0
offset_err = 0.0
for ind, c in enumerate((cen_x, cen_y)):
err = np.median(np.abs(np.diff(c)))
kernel = np.var(c) * kernels.Matern32Kernel(2**2)
gp = george.GP(kernel,
white_noise=2 * np.log(np.mean(err)),
fit_white_noise=True,
mean=CentroidModel(tm, a=0.0, b=0.0),
fit_mean=True)
gp.compute(x, err)
r = minimize(gp.nll,
gp.get_vector(),
jac=gp.grad_nll,
args=(c - np.mean(c), ),
method="L-BFGS-B")
gp.set_vector(r.x)
C = gp.get_matrix(x)
C[np.diag_indices_from(C)] += err**2
alpha = np.linalg.solve(C, A)
ATA = np.dot(AT, alpha)
mu = np.mean(c)
a = np.linalg.solve(C, c - mu)
w = None
for _ in range(10):
try:
w = np.linalg.solve(ATA, np.dot(AT, a))
except np.linalg.LinAlgError:
ATA[np.diag_indices_from(ATA)] *= 1 + 1e-5
w = None
else:
break
if w is None:
raise np.linalg.LinAlgError("Couldn't fix ATA")
offset += w[0]**2
offset_err = np.linalg.inv(ATA)[0, 0] * w[0]**2
peak["pred_cens"].append(np.dot(A, w) + mu)
offset_err = np.sqrt(offset_err / offset)
offset = np.sqrt(offset)
peak["centroid_offset"] = offset
peak["centroid_offset_err"] = offset_err
elif name == "transit":
peak["centroid_offset"] = 0.0
peak["centroid_offset_err"] = np.inf
if (peak["bic_{0}".format(name)] > peak["bic_transit"]
and not all_models):
break
# Deal with outliers.
if name != "gp":
continue
N = len(r.x) + 1
peak["lnlike_outlier"] = -r.fun
peak["bic_outlier"] = -r.fun - 0.5 * N * np.log(len(x))
best = (-r.fun, 0)
for j in np.arange(len(x))[np.abs(x - t0) < 0.5 * tau]:
y0 = np.array(y)
y0[j] = np.median(y[np.arange(len(y)) != j])
ll = gp.lnlikelihood(y0)
if ll > best[0]:
best = (ll, j)
# Optimize the outlier model:
m = np.arange(len(y)) != best[1]
y0 = np.array(y)
y0[~m] = np.median(y0[m])
kernel = np.var(y0) * kernels.Matern32Kernel(2**2)
gp = george.GP(kernel,
mean=np.median(y0),
fit_mean=True,
white_noise=2 * np.log(np.mean(yerr)),
fit_white_noise=True)
gp.compute(x, yerr)
r = minimize(gp.nll,
gp.get_vector(),
jac=gp.grad_nll,
args=(y0, ),
method="L-BFGS-B",
bounds=bounds)
gp.set_vector(r.x)
peak["lnlike_outlier"] = -r.fun
peak["bic_outlier"] = -r.fun - 0.5 * N * np.log(len(x))
peak["gps"]["outlier"] = (gp, y0)
if verbose:
print("Peak {0}:".format(i))
print("Model: 'outlier'")
print("Converged: {0}".format(r.success))
print("Log-likelihood: {0}".format(peak["lnlike_outlier"]))
print("-0.5 * BIC: {0}".format(peak["bic_outlier"]))
print("Parameters:")
for k, v in zip(gp.get_parameter_names(), gp.get_vector()):
print(" {0}: {1:.4f}".format(k, v))
print()
# Save the transit parameters.
peak["transit_impact"] = system.impact
peak["transit_duration"] = system.duration
peak["transit_ror"] = system.ror
peak["transit_time"] = system.t0
peak["transit_depth"] = 1.0 - float(system.get_value(system.t0))
# Accept the peak?
accept_bic = all(
peak["bic_transit"] >= peak.get("bic_{0}".format(k), -np.inf)
for k in models) and (peak["bic_transit"] > peak["bic_outlier"])
accept_time = ((peak["transit_time"] - 1.0 * peak["transit_duration"] >
peak["chunk_min_time"]) and
(peak["transit_time"] + 1.0 * peak["transit_duration"] <
peak["chunk_max_time"]))
accept = accept_bic and accept_time
peak["accept_bic"] = accept_bic
peak["accept_time"] = accept_time
# Save the injected parameters.
if inject:
for k in ["t0", "period", "ror", "b", "e", "omega"]:
peak["injected_{0}".format(k)] = injection[k]
# Check for recovery.
p = injection["period"]
d = (peak["transit_time"] - injection["t0"] +
0.5 * p) % p - 0.5 * p
peak["is_injection"] = np.abs(d) < peak["transit_duration"]
results.injection["recovered"] |= accept
# Save the peak.
results.peaks.append(peak)
return results
from __future__ import division, print_function
import time
import numpy as np
import matplotlib.pyplot as pl
import emcee
from kplr import EXPOSURE_TIMES
import transit
texp = EXPOSURE_TIMES[1] / 86400.0
s = transit.System(transit.Central(dilution=0.05))
body = transit.Body(radius=0.2, mass=0.0, period=4.0, t0=2, b=0.0, e=0.4,
omega=0.5*np.pi + 0.01)
s.add_body(body)
s.thaw_parameter("*")
print(s.get_parameter_names())
x = np.linspace(0, 10.0, 1000)
yerr = 5e-4 * np.ones_like(x)
y = s.light_curve(x) + yerr * np.random.randn(len(x))
pl.plot(x, y, ".k")
pl.savefig("data.png")
p0 = s.get_vector()
def inject(kicid, rng=6):
# Download the data.
client = kplr.API()
kic = client.star(kicid)
lcs = kic.get_light_curves(short_cadence=False)
lc = lcs[np.random.randint(len(lcs))]
# Read the data.
data = lc.read()
t = data["TIME"]
f = data["SAP_FLUX"]
fe = data["SAP_FLUX_ERR"]
q = data["SAP_QUALITY"]
# Remove missing points.
m = np.isfinite(t) * np.isfinite(f) * np.isfinite(fe) * (q == 0)
t, f, fe = t[m], f[m], fe[m]
t -= t.min()
# Build the transit system.
s = transit.System(
transit.Central(q1=np.random.rand(), q2=np.random.rand()))
body = transit.Body(period=365.25,
b=np.random.rand(),
r=0.04,
t0=np.random.uniform(t.max()))
s.add_body(body)
# Compute the transit model.
texp = kplr.EXPOSURE_TIMES[1] / 86400.0 # Long cadence exposure time.
model = s.light_curve(t, texp=texp)
f *= model
# Trim the dataset to include data only near the transit.
m = np.abs(t - body.t0) < rng
t, f, fe = t[m], f[m], fe[m]
t -= body.t0
# Save the injection as a FITS light curve.
dt = [("TIME", float), ("SAP_FLUX", float), ("SAP_FLUX_ERR", float)]
data = np.array(zip(t, f, fe), dtype=dt)
hdr = dict(b=body.b,
period=body.period,
r=body.r,
t0=0.0,
q1=s.central.q1,
q2=s.central.q2)
fitsio.write("{0}-injection.fits".format(kicid),
data,
header=hdr,
clobber=True)
# Plot the light curve.
ppm = (f / np.median(f) - 1) * 1e6
fig = pl.figure(figsize=(6, 6))
ax = fig.add_subplot(111)
ax.plot(t, ppm, ".k")
ax.set_xlim(-rng, rng)
ax.set_xlabel("time since transit [days]")
ax.set_ylabel("relative flux [ppm]")
ax.set_title("raw light curve")
fig.subplots_adjust(left=0.2, bottom=0.2, top=0.9, right=0.9)
fig.savefig("{0}-raw.pdf".format(kicid))
# Throw away the datasets without transits.
period, t0 = 295.963, 138.91
hp = 0.5 * period
selection = lambda lc: np.any(
np.abs((lc[0] - t0 + hp) % period - hp) < 1.0)
light_curves = filter(selection, light_curves)
print(len(light_curves))
# Plot the raw data.
for lc in light_curves:
pl.plot(lc[0], lc[1], ".", ms=3)
pl.savefig("raw_data.png")
# Set up the initial system.
system = transit.System(transit.Central(radius=0.95))
planet = transit.Body(r=2.03 * 0.01, period=period, t0=t0, b=0.9)
system.add_body(planet)
texp = kplr.EXPOSURE_TIMES[1] / 60. / 60. / 24.
mean_function = partial(system.light_curve, texp=texp)
# Set up the Gaussian processes.
pl.clf()
offset = 0.001
models = []
for i, lc in enumerate(light_curves):
dt = np.median(np.diff(lc[0])) * integrated_time(lc[1])
kernel = np.var(lc[1]) * kernels.Matern32Kernel(dt**2)
gp = george.GP(kernel, mean=mean_function, solver=george.HODLRSolver)
gp.compute(lc[0], lc[2])
models.append((gp, lc[1]))