def fastfit(self, **kwargs): self.fitter = LM(self, **kwargs) self.fitter.fit(**kwargs)
def slowfit(self, **kwargs): self.fitter = MCMC(self, **kwargs) self.fitter.fit(**kwargs)
class TM(Talker):
'''Transit Model object handles generation of model transit light curves.
(relies heavily on Jonathan Irwin's "eb" code, which is an updated
implementation of the classic EBOP and JKTEBOP, available at:
https://github.com/mdwarfgeek'''
def __init__(self, planet=None, star=None, instrument=None, directory=None,depthassumedforplotting=None, **kwargs):
'''Initialize the parameters of the model.'''
# setup the talker
Talker.__init__(self)
# create an empty array of parameters for eb
self.ebparams = np.zeros(eb.NPAR, dtype=np.double)
# keep track of a depth for plotting, if necessary
self.depthassumedforplotting=depthassumedforplotting
# load the model, if possible, and if a directory was given
self.directory = directory
if directory is not None and planet is None:
self.load(directory)
else:
# define the subsets of the parameters, maybe by falling to defaults
if planet is None:
planet = Planet()
if star is None:
star = Star()
if instrument is None:
instrument = Instrument()
self.planet = planet
self.star = star
self.instrument = instrument
def load(self, directory):
'''Load parameters from directory.'''
#self.directory = directory
self.speak('loading TM from {0}'.format(directory))
self.planet = Planet(directory=directory)
self.star = Star(directory=directory)
self.instrument = Instrument(directory=directory)
def save(self, directory):
'''Save parameters to directory.'''
#self.directory = directory
for x in (self.planet, self.star, self.instrument):
x.save(directory)
def linkLightCurve(self, transitlightcurve):
'''Attach a model to a light curve, defining all the TM and TLC attributes.'''
self.TLC = transitlightcurve
self.TM = self
self.TLC.TM = self
self.TLC.TLC = self.TLC
def linkRV(self, rvcurve):
'''Attach a model to a radial velocity curve, defining all the TM and RVC attributes.'''
self.RVC = rvcurve
self.TM = self
self.RVC.TM = self
self.RVC.RVC = self.RVC
#@profile
def set_ebparams(self):
'''Set up the parameters required for eb. '''
# These are the basic parameters of the model.
self.ebparams[eb.PAR_J] = self.planet.surface_brightness_ratio # J surface brightness ratio
self.ebparams[eb.PAR_RASUM] = self.planet.rsum_over_a # (R_1+R_2)/a
self.ebparams[eb.PAR_RR] = self.planet.rp_over_rs # R_2/R_1
self.ebparams[eb.PAR_COSI] = self.planet.cosi # cos i
# Mass ratio is used only for computing ellipsoidal variation and
# light travel time. Set to zero to disable ellipsoidal.
self.ebparams[eb.PAR_Q] = 0#self.planet.q
# Light travel time coefficient.
#ktot = 55.602793 # K_1+K_2 in km/s
#cltt = 1000*ktot / eb.LIGHT
# Set to zero if you don't need light travel correction (it's fairly slow
# and can often be neglected).
self.ebparams[eb.PAR_CLTT] = 0#cltt#*(self.planet.q.value == 0.0) # ktot / c
# Radiative properties of star 1.
self.ebparams[eb.PAR_LDLIN1] = self.star.u1.value # u1 star 1
self.ebparams[eb.PAR_LDNON1] = self.star.u2.value # u2 star 1
self.ebparams[eb.PAR_GD1] = self.star.gd.value # gravity darkening, std. value
self.ebparams[eb.PAR_REFL1] = self.star.albedo.value # albedo, std. value
# Spot model. Assumes spots on star 1 and not eclipsed.
self.ebparams[eb.PAR_ROT1] = 1.0# 0.636539 # rotation parameter (1 = sync.)
self.ebparams[eb.PAR_FSPOT1] = 0.0 # fraction of spots eclipsed
self.ebparams[eb.PAR_OOE1O] = 0.0 # base spottedness out of eclipse
self.ebparams[eb.PAR_OOE11A] = 0.0#0.006928 # *sin
self.ebparams[eb.PAR_OOE11B] = 0.0 # *cos
# PAR_OOE12* are sin(2*rot*omega) on star 1,
# PAR_OOE2* are for spots on star 2.
# Assume star 2 is the same as star 1 but without spots.
self.ebparams[eb.PAR_LDLIN2] = self.ebparams[eb.PAR_LDLIN1]
self.ebparams[eb.PAR_LDNON2] = self.ebparams[eb.PAR_LDNON1]
self.ebparams[eb.PAR_GD2] = self.ebparams[eb.PAR_GD1]
self.ebparams[eb.PAR_REFL2] = self.ebparams[eb.PAR_REFL1]
# Orbital parameters.
self.ebparams[eb.PAR_ECOSW] = self.planet.ecosw.value # ecosw
self.ebparams[eb.PAR_ESINW] = self.planet.esinw.value # esinw
self.ebparams[eb.PAR_P] = self.planet.period.value # period
self.ebparams[eb.PAR_T0] = self.planet.t0.value + self.planet.dt.value # T0 (epoch of primary eclipse), with an offset of dt applied
# OTHER NOTES:
#
# To do standard transit models (a'la Mandel & Agol),
# set J=0, q=0, cltt=0, albedo=0.
# This makes the secondary dark, and disables ellipsoidal and reflection.
#
# The strange parameterization of radial velocity is to retain the
# flexibility to be able to model just light curves, SB1s, or SB2s.
#
# For improved precision, it's best to subtract most of the "DC offset"
# from T0 and the time array (e.g. take off the nominal value of T0 or
# the midtime of the data array) and add it back on at the end when
# printing self.ebparams[eb.PAR_T0] and vder[eb.PAR_TSEC]. Likewise the period
# can cause scaling problems in minimization routines (because it has
# to be so much more precise than the other parameters), and may need
# similar treatment.
def stellar_rv(self, rvc=None, t=None):
self.set_ebparams()
# by default, will use the linked RVC, but could use a custom one (e.g. high-resolution for plotting), or just times
if rvc is None:
rvc = self.RVC
if t is None:
t = rvc.bjd
# make sure the types are okay for Jonathan's inputs
typ = np.empty_like(t, dtype=np.uint8)
typ.fill(eb.OBS_VRAD1)
rv = self.planet.semiamplitude.value*eb.model(self.ebparams, t, typ) + self.star.gamma.value
assert(np.isfinite(rv).all())
return rv
#@profile
def planet_model(self, tlc=None, t=None):
'''Model of the planetary transit.'''
self.set_ebparams()
# by default, will use the linked TLC, but could use a custom TLC
# (e.g. a high-resolution one, for plotting)
if tlc is None:
tlc = self.TLC
# if called with a "t=" keyword set, then will use those custom times
if t is None:
t = tlc.bjd
# make sure the types are okay for Jonathan's inputs
typ = np.empty_like(t, dtype=np.uint8)
typ.fill(eb.OBS_MAG)
# work in relative flux (rather than magnitudes) -- why did I do this?
return 10**(-0.4*eb.model(self.ebparams, t, typ))
##@profile
def instrument_model(self, tlc=None):
'''Model of the instrument.'''
# by default, use the real TLC
if tlc is None:
tlc = self.TLC
# use the instrument to spit out a corrective model
return self.instrument.model(tlc)
##@profile
def model(self, tlc=None):
'''Model including both instrument and planetary transit.'''
# create the complete model, for either real or fake light curve
return self.planet_model(tlc=tlc)*self.instrument_model(tlc=tlc)
@property
def parameters(self):
'''Return a list of the parameters that are variable.'''
try:
assert(len(self._parameters) == len(self.floating))
return self._parameters
except:
self.defineParameterList()
return self._parameters
@parameters.setter
def parameters(self, **kwargs):
pass
def defineParameterList(self):
'''Set up the parameter list, by pulling the variable parameters out of the subsets.'''
# define a list containing the keys all the parameters that float
self.floating = []
list = []
for x in (self.planet, self.star, self.instrument):
d = x.__dict__
for key in d.keys():
try:
if d[key].fixed == False:
self.floating.append(key)
list.append(d[key])
except:
pass
self._parameters = np.array(list)
def fromArray(self, array):
'''Use an input array to assign the internal parameter attributes.'''
count = 0
for parameter in self.parameters:
parameter.value = array[count]
count += 1
#print count
#print array
assert(count > 0)
def toArray(self):
'''Define an parameter array, by pulling them out of the internal parameter attributes.'''
list = []
parinfolist = []
for parameter in self.parameters:
list.append(parameter.value)
parinfolist.append(parameter.parinfo)
return list, parinfolist
def plotPhased(self, smooth=None, **kw):
'''Plot the light curve model, phased with the planetary period.'''
t_phased = self.planet.timefrommidtransit(self.smooth_phased_tlc.bjd)
assert(len(t_phased) == len(self.model(self.smooth_phased_tlc)))
toplot = self.planet_model(self.smooth_phased_tlc)
if smooth is not None:
cadence = np.mean(self.smooth_phased_tlc.bjd[1:] - self.smooth_phased_tlc.bjd[:-1])
n = np.round(smooth/cadence).astype(np.int)
kernel = np.ones(n)/n
toplot = np.convolve(toplot, kernel, 'same')
else:
toplot = self.planet_model(self.smooth_phased_tlc)
plt.plot(t_phased, toplot, **kw)
'''KLUDGED COMMENTED OUT!'''
#try:
# for phased in [self.line_phased[0], self.line_phased_zoom[0]]:
# phased.set_data(t_phased, self.model(self.smooth_phased_tlc))
#except:
# self.line_phased = self.TLC.ax_phased.plot(t_phased,self.model(self.smooth_phased_tlc), **self.kw)
# self.line_phased_zoom = self.TLC.ax_phased_zoom.plot(t_phased, self.model(self.smooth_phased_tlc), **self.kw)'''
def plotUnphased(self):
'''Plot the light curve model, linear in time.'''
t_unphased = self.smooth_unphased_tlc.bjd - self.planet.t0.value
assert(len(t_unphased) == len(self.model(self.smooth_unphased_tlc)))
try:
for unphased in [self.line_unphased[0], self.line_unphased_zoom[0]]:
unphased.set_data(t_unphased, self.model(self.smooth_unphased_tlc))
except:
self.line_unphased = self.TLC.ax_unphased.plot(t_unphased, self.model(self.smooth_unphased_tlc), **self.kw)
self.line_unphased_zoom = self.TLC.ax_unphased_zoom.plot(t_unphased, self.model(self.smooth_unphased_tlc), **self.kw)
def plotDiagnostics(self):
'''Plot the light curve model, linear in time.'''
t_unphased = self.smooth_unphased_tlc.bjd - self.planet.t0.value
assert(len(t_unphased) == len(self.model(self.smooth_unphased_tlc)))
#try:
# self.line_raw.set_data(t_unphased, self.model(self.smooth_unphased_tlc))
# self.line_corrected.set_data(t_unphased, self.planet_model(self.smooth_unphased_tlc))
# self.line_residuals.set_data(t_unphased, ppm*np.zeros_like(t_unphased))
# self.line_instrument.set_data(t_unphased, ppm*self.instrument_model(self.smooth_unphased_tlc))
#except:
# NOT USING?!?
self.line_raw = self.TLC.ax_raw.plot(t_unphased, self.model(self.smooth_unphased_tlc), **kw)[0]
self.line_corrected = self.TLC.ax_corrected.plot(t_unphased, self.planet_model(self.smooth_unphased_tlc), **kw)[0]
self.line_residuals = self.TLC.ax_residuals.plot(t_unphased, ppm*np.zeros_like(t_unphased), **kw)[0]
self.line_instrument = self.TLC.ax_instrument.plot(t_unphased, ppm*self.instrument_model(self.smooth_unphased_tlc), **kw)[0]
def plot(self):
'''Plot the model lines over the existing light curve structures.'''
self.kw = {'color':self.TLC.colors['lines'], 'linewidth':3, 'alpha':1.0}
self.plotPhased()
self.plotUnphased()
if self.depthassumedforplotting is None:
self.depthassumedforplotting = self.planet.rp_over_rs.value**2
# need to work on displaying the parameters...
try:
xtext = self.planet.duration/2.0*1.1
posttransit = self.planet.timefrommidtransit(bjd) > self.planet.duration/2.0
ytext = np.mean(self.model()[posttransit]) - self.planet.rp_over_rs**2/2.0
instrument_string = "Instrumental Parameters"
self.TLC.ax_raw.text(xtext, ytext, instrument_string)
except:
pass
##@profile
def deviates(self, p, fjac=None, plotting=False):
'''Return the normalized deviates (an input for mpfit).'''
# populate the parameter attributes, using the input array
self.fromArray(p)
# if necessary, plot the light curve along with this step of the deviates calculation
status =0
if plotting:
self.TLC.LightcurvePlots()
ok = (self.TLC.bad == 0).nonzero()
devs = (self.TLC.flux[ok] - self.TM.model()[ok])/self.TLC.uncertainty[ok]
# add limb darkening priors
try:
prioru1 = (self.star.u1.value - self.u1prior_value)/self.u1prior_uncertainty
prioru2 = (self.star.u2.value - self.u2prior_value)/self.u2prior_uncertainty
devs = np.append(devs, prioru1)
devs = np.append(devs, prioru2)
#print '==============================='
#print 'u1: ({value} - {center})/{uncertainty}'.format(value = self.star.u1.value, center=self.u1prior_value, uncertainty =self.u1prior_uncertainty)
#print 'u2: ({value} - {center})/{uncertainty}'.format(value = self.star.u2.value, center=self.u2prior_value, uncertainty =self.u2prior_uncertainty)
except:
pass
# mpfit wants a list with the first element containing a status code
return [status,devs]
def applyLDpriors(self):
self.speak('using atmosphere model as prior on LD coefficients')
self.u1prior_value = self.star.u1.value + 0.0
self.u2prior_value = self.star.u2.value + 0.0
self.u1prior_uncertainty = self.star.u1.uncertainty + 0.0
self.u2prior_uncertainty = self.star.u2.uncertainty + 0.0
##@profile
def lnprob(self, p):
"""Return the log posterior, calculated from the TM.deviates function (which may have included some conjugate Gaussian priors.)"""
chisq = np.sum(self.deviates(p)[-1]**2)/2.0
N = np.sum(self.TLC.bad == 0)
# sum the deviates into a chisq-like thing
lnlikelihood = -N * np.log(self.instrument.rescaling.value) - chisq/self.instrument.rescaling.value**2
if np.isfinite(lnlikelihood) == False:
lnlikelihood = -1e9
# initialize an empty constraint, which could freak out if there's something bad about this fit
constraints = 0.0
# loop over the parameters
for parameter in self.parameters:
# if a parameter is outside its allowed range, then make the constraint very strong!
inside = (parameter.value < parameter.limits[1]) & (parameter.value > parameter.limits[0])
try:
assert(inside)
except AssertionError:
constraints -= 1e6
# return the constrained likelihood
return lnlikelihood + constraints
def fastfit(self, **kwargs):
self.fitter = LM(self, **kwargs)
self.fitter.fit(**kwargs)
def slowfit(self, **kwargs):
self.fitter = MCMC(self, **kwargs)
self.fitter.fit(**kwargs)
def __repr__(self):
return self.TLC.__repr__().replace('TLC', 'TM')