def lnlike(params, bjd, mag, magerr):
a_gp, l_gp = np.exp(params[:2])
k1 = kernels.ExpSquaredKernel(l_gp)
kernel = a_gp * k1
gp = george.GP(kernel) #, solver=george.HODLRSolver)
try:
gp.compute(bjd, magerr)
except (ValueError, np.linalg.LinAlgError):
return 1e26
model = lcmodel.get_lc1(params[2:], bjd)
return gp.lnlikelihood(mag - model, quiet=True)
def get_model(self):
'''Compute the QP GP model after running the mcmc.'''
if self.GPonly:
# Compute GP model
a_gp, l_gp, G_gp, P_gp = self.hyperparams
k1 = kernels.ExpSquaredKernel(l_gp)
k2 = kernels.ExpSine2Kernel(G_gp, P_gp)
kernel = a_gp * k1 * k2
gp = george.GP(kernel)
gp.compute(self.bjdtrimbin, self.emagtrimbin)
x = np.linspace(min(self.bjd), max(self.bjd), 5e2)
mu, cov = gp.predict(self.magtrimbin, x)
std = np.sqrt(np.diag(cov))
xpred = np.linspace(max(self.bjd), max(self.bjd) + 600, 1e3)
mupred, covpred = gp.predict(self.magtrimbin, xpred)
stdpred = np.sqrt(np.diag(covpred))
else:
# Compute GP model
a_gp, l_gp = self.hyperparams
k1 = kernels.ExpSquaredKernel(l_gp)
kernel = a_gp * k1
gp = george.GP(kernel, solver=george.HODLRSolver)
gp.compute(self.bjdtrimbin, self.emagtrimbin)
x = np.linspace(min(self.bjd), max(self.bjd), 3e2)
#mu, cov = gp.predict(self.magtrimbin, x)
modelsmall = lcmodel.get_lc1(self.mcmcparams, self.bjdtrimbin)
modellarge = lcmodel.get_lc1(self.mcmcparams, x)
samples = gp.sample_conditional(self.magtrimbin - modelsmall,
x,
size=100)
mu = np.mean(samples, axis=0) + modellarge
std = np.std(samples, axis=0) + modellarge
self.modelbjd = x
self.model = mu
self.modelerr = std
self.predbjd = xpred
self.predmodel = mupred
self.predmodelerr = stdpred
def get_model(self):
'''Compute the QP GP model after running the mcmc.'''
if self.GPonly:
# Compute GP model
a_gp, l_gp, G_gp, P_gp = self.hyperparams
k1 = kernels.ExpSquaredKernel(l_gp)
k2 = kernels.ExpSine2Kernel(G_gp, P_gp)
kernel = a_gp*k1*k2
gp = george.GP(kernel)
gp.compute(self.bjdtrimbin, self.emagtrimbin)
x = np.linspace(min(self.bjd), max(self.bjd), 5e2)
mu, cov = gp.predict(self.magtrimbin, x)
std = np.sqrt(np.diag(cov))
xpred = np.linspace(max(self.bjd), max(self.bjd)+600, 1e3)
mupred, covpred = gp.predict(self.magtrimbin, xpred)
stdpred = np.sqrt(np.diag(covpred))
else:
# Compute GP model
a_gp, l_gp = self.hyperparams
k1 = kernels.ExpSquaredKernel(l_gp)
kernel = a_gp*k1
gp = george.GP(kernel, solver=george.HODLRSolver)
gp.compute(self.bjdtrimbin, self.emagtrimbin)
x = np.linspace(min(self.bjd), max(self.bjd), 3e2)
#mu, cov = gp.predict(self.magtrimbin, x)
modelsmall = lcmodel.get_lc1(self.mcmcparams, self.bjdtrimbin)
modellarge = lcmodel.get_lc1(self.mcmcparams, x)
samples = gp.sample_conditional(self.magtrimbin - modelsmall,
x, size=100)
mu = np.mean(samples, axis=0) + modellarge
std = np.std(samples, axis=0) + modellarge
self.modelbjd = x
self.model = mu
self.modelerr = std
self.predbjd = xpred
self.predmodel = mupred
self.predmodelerr = stdpred