def fitExampleDRW(datafilename='../data/SDSSJ203817.37+003029.8.fits'):
"""
NAME:
fitExampleDRW
PURPOSE:
Fit an example of a DRW structure function GP covariance function
to an SDSS quasar in g and r
INPUT:
datafilename
OUTPUT:
HISTORY:
2010-02-21 - Written - Bovy (NYU)
"""
nu.random.seed(1)
#Get data
file = fu.table_fields(datafilename)
mjd_g = nu.array(file.mjd_g) / 365.25
g = nu.array(file.g)
err_g = nu.array(file.err_g)
mjd_r = nu.array(file.mjd_r) / 365.25
r = nu.array(file.r)
err_r = nu.array(file.err_r)
mask = (mjd_g != 0) * (g < 20.6) * (g > 19.7) #Adjust for non-g
g = g[mask]
g -= nu.mean(g)
err_g = err_g[mask]
mjd_g = mjd_g[mask]
mjd_g -= nu.amin(mjd_g)
meanErr_g = nu.mean(err_g)
r = r[mask]
r -= nu.mean(r)
err_r = err_r[mask]
mjd_r = mjd_r[mask]
mjd_r -= nu.amin(mjd_r)
meanErr_r = nu.mean(err_r)
listx = [(t, 'g') for t in mjd_g]
listx.extend([(t, 'r') for t in mjd_r])
listy = [m for m in g]
listy.extend([m for m in r])
listy = nu.array(listy)
noise = [m for m in err_g]
noise.extend([m for m in err_r])
noise = nu.array(noise)
trainSet = trainingSet(listx=listx, listy=listy, noise=noise)
SF = covarFunc(S=1., tau=.2, B=.1, C=.005, gammagr=.5, Gammagr=.01)
#Train
print "Training ..."
outcovarFunc = trainGP(trainSet, SF, useDerivs=False)
print "Best-fit:"
print outcovarFunc._dict
return None
def fitExampleDRW(datafilename="../data/SDSSJ203817.37+003029.8.fits"):
"""
NAME:
fitExampleDRW
PURPOSE:
Fit an example of a DRW structure function GP covariance function
to an SDSS quasar in g and r
INPUT:
datafilename
OUTPUT:
HISTORY:
2010-02-21 - Written - Bovy (NYU)
"""
nu.random.seed(1)
# Get data
file = fu.table_fields(datafilename)
mjd_g = nu.array(file.mjd_g) / 365.25
g = nu.array(file.g)
err_g = nu.array(file.err_g)
mjd_r = nu.array(file.mjd_r) / 365.25
r = nu.array(file.r)
err_r = nu.array(file.err_r)
mask = (mjd_g != 0) * (g < 20.6) * (g > 19.7) # Adjust for non-g
g = g[mask]
g -= nu.mean(g)
err_g = err_g[mask]
mjd_g = mjd_g[mask]
mjd_g -= nu.amin(mjd_g)
meanErr_g = nu.mean(err_g)
r = r[mask]
r -= nu.mean(r)
err_r = err_r[mask]
mjd_r = mjd_r[mask]
mjd_r -= nu.amin(mjd_r)
meanErr_r = nu.mean(err_r)
listx = [(t, "g") for t in mjd_g]
listx.extend([(t, "r") for t in mjd_r])
listy = [m for m in g]
listy.extend([m for m in r])
listy = nu.array(listy)
noise = [m for m in err_g]
noise.extend([m for m in err_r])
noise = nu.array(noise)
trainSet = trainingSet(listx=listx, listy=listy, noise=noise)
SF = covarFunc(S=1.0, tau=0.2, B=0.1, C=0.005, gammagr=0.5, Gammagr=0.01)
# Train
print "Training ..."
outcovarFunc = trainGP(trainSet, SF, useDerivs=False)
print "Best-fit:"
print outcovarFunc._dict
return None
def singleBandSampleExample(
datafilename='../data/SDSSJ203817.37+003029.8.fits',
band='g',
nsamples=1000,
basefilename='SDSSJ203817.37+003029.8'):
"""
NAME:
singleBandSampleExample
PURPOSE:
Sample an example of a power-law structure function GP covariance
function to an SDSS quasar
INPUT:
datafilename
band - band to fit
nsamples - number of samples to use
basefilename
OUTPUT:
HISTORY:
2010-08-08 - Written - Bovy (NYU)
"""
nu.random.seed(1)
#Get data
file = fu.table_fields(datafilename)
if band == 'u':
mjd_g = nu.array(file.mjd_u) / 365.25
g = nu.array(file.u)
err_g = nu.array(file.err_u)
elif band == 'g':
mjd_g = nu.array(file.mjd_g) / 365.25
g = nu.array(file.g)
err_g = nu.array(file.err_g)
elif band == 'r':
mjd_g = nu.array(file.mjd_r) / 365.25
g = nu.array(file.r)
err_g = nu.array(file.err_r)
elif band == 'i':
mjd_g = nu.array(file.mjd_i) / 365.25
g = nu.array(file.i)
err_g = nu.array(file.err_i)
elif band == 'z':
mjd_g = nu.array(file.mjd_z) / 365.25
g = nu.array(file.z)
err_g = nu.array(file.err_z)
mjd_r = nu.array(file.mjd_r) / 365.25
r = nu.array(file.r)
err_r = nu.array(file.err_r)
mjd_i = nu.array(file.mjd_i) / 365.25
i = nu.array(file.i)
err_i = nu.array(file.err_i)
mask = (mjd_g != 0) * (g < 20.6) * (g > 19.7)
g = g[mask]
g -= nu.mean(g)
err_g = err_g[mask]
mjd_g = mjd_g[mask]
mjd_g -= nu.amin(mjd_g)
meanErr_g = nu.mean(err_g)
r = r[mask]
r -= nu.mean(r)
err_r = err_r[mask]
mjd_r = mjd_r[mask]
mjd_r -= nu.amin(mjd_r)
meanErr_r = nu.mean(err_r)
i = i[mask]
i -= nu.mean(i)
err_i = err_i[mask]
mjd_i = mjd_i[mask]
mjd_i -= nu.amin(mjd_i)
meanErr_i = nu.mean(err_i)
savefilename = basefilename + '.sav'
if os.path.exists(savefilename):
savefile = open(savefilename, 'rb')
gammas = pickle.load(savefile)
if 'gr' in basefilename:
logGammas = pickle.load(savefile)
gammagrs = pickle.load(savefile)
logGammagrs = pickle.load(savefile)
else:
logAs = pickle.load(savefile)
out = pickle.load(savefile)
savefile.close()
else:
if 'gri' in basefilename:
raise NotImplementedError
from powerlawSFgr import covarFunc
params = {
'logGamma': array([-7.79009776]),
'logGammagr': array([-28.0487848]),
'gamma': array([0.45918053]),
'gammagr': array([0.21333858])
}
SF = covarFunc(**params)
listx = [(t, 'g') for t in mjd_g]
listx.extend([(t, 'r') for t in mjd_r])
listx.extend([(t, 'i') for t in mjd_i])
listy = [m for m in g]
listy.extend([m for m in r])
listy.extend([m for m in i])
listy = nu.array(listy)
noise = [m for m in err_g]
noise.extend([m for m in err_r])
noise.extend([m for m in err_i])
noise = nu.array(noise)
trainSet = trainingSet(listx=listx, listy=listy, noise=noise)
elif 'gr' in basefilename:
raise NotImplementedError
from powerlawSFgr import covarFunc
params = {
'logGamma': array([-7.79009776]),
'logGammagr': array([-28.0487848]),
'gamma': array([0.45918053]),
'gammagr': array([0.21333858])
}
SF = covarFunc(**params)
listx = [(t, 'g') for t in mjd_g]
listx.extend([(t, 'r') for t in mjd_r])
listy = [m for m in g]
listy.extend([m for m in r])
listy = nu.array(listy)
noise = [m for m in err_g]
noise.extend([m for m in err_r])
noise = nu.array(noise)
trainSet = trainingSet(listx=listx, listy=listy, noise=noise)
else:
from gp.powerlawSF import covarFunc
trainSet = trainingSet(listx=mjd_g, listy=g, noise=err_g)
params = [0., 0.]
SF = covarFunc(gamma=.2, A=0.000524) #Power
#SF= covarFunc(a=.0001,l=.001) #OU
#Train
print "Training ..."
outcovarFunc = trainGP(trainSet, SF, useDerivs=False)
print "Best-fit:"
print outcovarFunc._dict
print "Sampling ..."
out = sampleGP(trainSet,
outcovarFunc,
nsamples=nsamples,
step=[0.2 for ii in range(len(params))])
gammas = nu.array([o._dict['gamma'] for o in out]).reshape(len(out))
if 'gr' in basefilename:
gammagrs = nu.array([o._dict['gammagr']
for o in out]).reshape(len(out))
logGammas = nu.array([o._dict['logGamma']
for o in out]).reshape(len(out))
logGammagrs = nu.array([o._dict['logGammagr']
for o in out]).reshape(len(out))
else:
logAs = nu.array([o._dict['logA'] for o in out]).reshape(len(out))
print nu.mean(logAs), nu.sqrt(nu.var(logAs))
print nu.mean(gammas), nu.sqrt(nu.var(gammas))
#Save
savefile = open(savefilename, 'wb')
pickle.dump(gammas, savefile)
if 'gr' in basefilename:
pickle.dump(logGammas, savefile)
pickle.dump(gammagrs, savefile)
pickle.dump(logGammagrs, savefile)
else:
pickle.dump(logAs, savefile)
pickle.dump(out, savefile)
savefile.close()
#if not 'gr' in basefilename:
# logGammas= logAs/gammas
#Plot
bovy_plot.bovy_print()
bovy_plot.scatterplot(logAs / 2.,
gammas,
'k,',
ylabel=r'$\gamma$',
xlabel=r'\log A',
xrange=[-9.21 / 2., 0.],
yrange=[0., 1.25],
bins=50,
onedhists=True)
bovy_plot.bovy_end_print(basefilename + '_sample_2d.png')
def fitExamplegr(datafilename='../data/SDSSJ203817.37+003029.8.fits'):
"""
NAME:
fitExamplegr
PURPOSE:
Fit an example of a power-law structure function GP covariance function
to an SDSS quasar in g and r
INPUT:
datafilename
OUTPUT:
HISTORY:
2010-02-21 - Written - Bovy (NYU)
"""
nu.random.seed(1)
#Get data
file= fu.table_fields(datafilename)
mjd_g= nu.array(file.mjd_g)/365.25
g= nu.array(file.g)
err_g= nu.array(file.err_g)
mjd_r= nu.array(file.mjd_r)/365.25
r= nu.array(file.r)
err_r= nu.array(file.err_r)
mask= (mjd_g != 0)*(g < 20.6)*(g > 19.7)#Adjust for non-g
g= g[mask]
g-= nu.mean(g)
err_g= err_g[mask]
mjd_g= mjd_g[mask]
mjd_g-= nu.amin(mjd_g)
meanErr_g= nu.mean(err_g)
r= r[mask]
r-= nu.mean(r)
err_r= err_r[mask]
mjd_r= mjd_r[mask]
mjd_r-= nu.amin(mjd_r)
meanErr_r= nu.mean(err_r)
listx= [(t,'g') for t in mjd_g]
listx.extend([(t,'r') for t in mjd_r])
listy= [m for m in g]
listy.extend([m for m in r])
listy= nu.array(listy)
noise= [m for m in err_g]
noise.extend([m for m in err_r])
noise= nu.array(noise)
trainSet= trainingSet(listx=listx,listy=listy,noise=noise)
#SF= covarFunc(Gamma=0.0058,gamma=0.5,Gammagr=0.0058,gammagr=0.5)
SF= covarFunc(logl=-1.37742591,loga=-3.47341754,
loglgr=-2.3777,logagr= -1.)
#Train
print "Training ..."
outcovarFunc= trainGP(trainSet,SF,useDerivs=False)
print "Best-fit:"
print outcovarFunc._dict
return None
def singleBandSampleExample(
datafilename="SDSSJ000051.56+001202.5.fit", band="g", nsamples=1000, basefilename="SDSSJ203817.37+003029.8"
):
"""
NAME:
singleBandSampleExample
PURPOSE:
Sample an example of a power-law structure function GP covariance
function to an SDSS quasar
INPUT:
datafilename
band - band to fit
nsamples - number of samples to use
basefilename
OUTPUT:
HISTORY:
2010-08-08 - Written - Bovy (NYU)
"""
nu.random.seed(1)
# Get data
file = fu.table_fields(datafilename)
if band == "u":
mjd_g = nu.array(file.mjd_u) / 365.25
g = nu.array(file.u)
err_g = nu.array(file.err_u)
elif band == "g":
mjd_g = nu.array(file.mjd_g) / 365.25
g = nu.array(file.g)
err_g = nu.array(file.err_g)
elif band == "r":
mjd_g = nu.array(file.mjd_r) / 365.25
g = nu.array(file.r)
err_g = nu.array(file.err_r)
elif band == "i":
mjd_g = nu.array(file.mjd_i) / 365.25
g = nu.array(file.i)
err_g = nu.array(file.err_i)
elif band == "z":
mjd_g = nu.array(file.mjd_z) / 365.25
g = nu.array(file.z)
err_g = nu.array(file.err_z)
mjd_r = nu.array(file.mjd_r) / 365.25
r = nu.array(file.r)
err_r = nu.array(file.err_r)
mjd_i = nu.array(file.mjd_i) / 365.25
i = nu.array(file.i)
err_i = nu.array(file.err_i)
mask = (mjd_g != 0) * (g < 20.6) * (g > 19.7)
g = g[mask]
g -= nu.mean(g)
err_g = err_g[mask]
mjd_g = mjd_g[mask]
mjd_g -= nu.amin(mjd_g)
meanErr_g = nu.mean(err_g)
r = r[mask]
r -= nu.mean(r)
err_r = err_r[mask]
mjd_r = mjd_r[mask]
mjd_r -= nu.amin(mjd_r)
meanErr_r = nu.mean(err_r)
i = i[mask]
i -= nu.mean(i)
err_i = err_i[mask]
mjd_i = mjd_i[mask]
mjd_i -= nu.amin(mjd_i)
meanErr_i = nu.mean(err_i)
savefilename = basefilename + ".sav"
if os.path.exists(savefilename):
savefile = open(savefilename, "rb")
gammas = pickle.load(savefile)
if "gr" in basefilename:
logGammas = pickle.load(savefile)
gammagrs = pickle.load(savefile)
logGammagrs = pickle.load(savefile)
else:
logAs = pickle.load(savefile)
out = pickle.load(savefile)
savefile.close()
else:
if "gri" in basefilename:
raise NotImplementedError
from powerlawSFgr import covarFunc
params = {
"logGamma": array([-7.79009776]),
"logGammagr": array([-28.0487848]),
"gamma": array([0.45918053]),
"gammagr": array([0.21333858]),
}
SF = covarFunc(**params)
listx = [(t, "g") for t in mjd_g]
listx.extend([(t, "r") for t in mjd_r])
listx.extend([(t, "i") for t in mjd_i])
listy = [m for m in g]
listy.extend([m for m in r])
listy.extend([m for m in i])
listy = nu.array(listy)
noise = [m for m in err_g]
noise.extend([m for m in err_r])
noise.extend([m for m in err_i])
noise = nu.array(noise)
trainSet = trainingSet(listx=listx, listy=listy, noise=noise)
elif "gr" in basefilename:
raise NotImplementedError
from powerlawSFgr import covarFunc
params = {
"logGamma": array([-7.79009776]),
"logGammagr": array([-28.0487848]),
"gamma": array([0.45918053]),
"gammagr": array([0.21333858]),
}
SF = covarFunc(**params)
listx = [(t, "g") for t in mjd_g]
listx.extend([(t, "r") for t in mjd_r])
listy = [m for m in g]
listy.extend([m for m in r])
listy = nu.array(listy)
noise = [m for m in err_g]
noise.extend([m for m in err_r])
noise = nu.array(noise)
trainSet = trainingSet(listx=listx, listy=listy, noise=noise)
else:
from gp.powerlawSF import covarFunc
trainSet = trainingSet(listx=mjd_g, listy=g, noise=err_g)
params = [0.0, 0.0]
SF = covarFunc(gamma=0.2, A=0.000524) # Power
# SF= covarFunc(a=.0001,l=.001) #OU
# Train
print "Training ..."
outcovarFunc = trainGP(trainSet, SF, useDerivs=False)
print "Best-fit:"
print outcovarFunc._dict
print "Sampling ..."
out = sampleGP(trainSet, outcovarFunc, nsamples=nsamples, step=[0.2 for ii in range(len(params))])
gammas = nu.array([o._dict["gamma"] for o in out]).reshape(len(out))
if "gr" in basefilename:
gammagrs = nu.array([o._dict["gammagr"] for o in out]).reshape(len(out))
logGammas = nu.array([o._dict["logGamma"] for o in out]).reshape(len(out))
logGammagrs = nu.array([o._dict["logGammagr"] for o in out]).reshape(len(out))
else:
logAs = nu.array([o._dict["logA"] for o in out]).reshape(len(out))
print nu.mean(logAs), nu.sqrt(nu.var(logAs))
print nu.mean(gammas), nu.sqrt(nu.var(gammas))
# Save
savefile = open(savefilename, "wb")
pickle.dump(gammas, savefile)
if "gr" in basefilename:
pickle.dump(logGammas, savefile)
pickle.dump(gammagrs, savefile)
pickle.dump(logGammagrs, savefile)
else:
pickle.dump(logAs, savefile)
pickle.dump(out, savefile)
savefile.close()
# if not 'gr' in basefilename:
# logGammas= logAs/gammas
# Plot
bovy_plot.bovy_print()
bovy_plot.scatterplot(
logAs / 2.0,
gammas,
"k,",
ylabel=r"$\gamma$",
xlabel=r"\log A",
xrange=[-9.21 / 2.0, 0.0],
yrange=[0.0, 1.25],
bins=50,
onedhists=True,
)
bovy_plot.bovy_end_print(basefilename + "_sample_2d.png")
def fitExample(N=None,datafilename='../data/SDSSJ203817.37+003029.8.fits',
band='g'):
"""
NAME:
fitExample
PURPOSE:
Fit an example of a power-law structure function GP covariance function
to an SDSS quasar
INPUT:
N - use a random subsample of N points
datafilename
band - band to fit
OUTPUT:
HISTORY:
2010-02-21 - Written - Bovy (NYU)
"""
nu.random.seed(1)
#Get data
file= fu.table_fields(datafilename)
if band == 'u':
mjd_g= nu.array(file.mjd_u)/365.25
g= nu.array(file.u)
err_g= nu.array(file.err_u)
elif band == 'g':
mjd_g= nu.array(file.mjd_g)/365.25
g= nu.array(file.g)
err_g= nu.array(file.err_g)
elif band == 'r':
mjd_g= nu.array(file.mjd_r)/365.25
g= nu.array(file.r)
err_g= nu.array(file.err_r)
elif band == 'i':
mjd_g= nu.array(file.mjd_i)/365.25
g= nu.array(file.i)
err_g= nu.array(file.err_i)
elif band == 'z':
mjd_g= nu.array(file.mjd_z)/365.25
g= nu.array(file.z)
err_g= nu.array(file.err_z)
mask= (mjd_g != 0)*(g < 20.6)*(g > 19.7)
g= g[mask]
g-= nu.mean(g)
err_g= err_g[mask]
mjd_g= mjd_g[mask]
mjd_g-= nu.amin(mjd_g)
meanErr_g= nu.mean(err_g)
trainSet= trainingSet(listx=mjd_g,listy=g,noise=err_g)
SF= covarFunc(a=.0001,l=.001)
#SF= covarFunc(gamma=.2,A=0.000524)
#SF= covarFunc(a=0.18,l=0.005)
#Train
print "Training ..."
outcovarFunc= trainGP(trainSet,SF,useDerivs=False)
print "Best-fit:"
print outcovarFunc._dict
return None
