def mc_goodness_fit(par, drop, niter, zlo=-1.):
# Determine goodness of fit given the parameters.
# Draw the same number of points from the model as observed, then calculate the
# loglikelihood of the drawn points. Calculate the probability that a simulated
# observation has lower loglikelihood than the observed points.
if drop == 'b':
cat1 = 'bdrops_gf_v2.cat'
cat2 = 'bdrops_udf_gf_v2.cat'
kgrid1 = mlutil.readkgrid('kernel_I.p')
kgrid2 = mlutil.readkgrid('kernel_I_udf.p')
zdf1 = 'zdgrid_bdrops.p'
zdf2 = 'zdgrid_bdrops_udf.p'
mcfile = 'M1500_to_i.txt'
mc = bl.mconvert(mcfile)
mk = mc(4.0)
chisqnulim = [0.4, 5.0]
elif drop == 'v':
cat1 = 'vdrops_gf_v2.cat'
cat2 = 'vdrops_udf_gf_v2.cat'
kgrid1 = mlutil.readkgrid('kernel_Z.p')
kgrid2 = mlutil.readkgrid('kernel_Z_udf.p')
zdf1 = 'zdgrid_vdrops.p'
zdf2 = 'zdgrid_vdrops_udf.p'
mcfile = 'M1500_to_z.txt'
mc = bl.mconvert(mcfile)
mk = mc(5.0)
chisqnulim = [0.5, 5.0]
cullflags = [0, 1, 2, 3, 4, 12, 13, 14, 18]
limits1 = array([[21.0, 26.5], [-2.0, 3.0]])
limits2 = array([[23.0, 28.5], [-2.0, 3.0]])
#limits1 = bl.limits1
#limits2 = bl.limits2
pixdx = array([0.02, 0.02])
mag1, re1, crit1 = fl.cleandata(cat1,
chisqnulim=chisqnulim[0],
magautolim=26.5,
cullflags=cullflags,
limits=limits1,
zlo=zlo,
drop=drop)
mag2, re2, crit2 = fl.cleandata(cat2,
chisqnulim=chisqnulim[1],
magautolim=28.5,
cullflags=cullflags,
limits=limits2,
zlo=zlo,
drop=drop)
data1 = array([mag1, log10(re1)])
data2 = array([mag2, log10(re2)])
N1 = len(mag1)
N2 = len(mag2)
print N1, N2, N1 + N2
model1 = bl.bivariate_lf(par, limits1, pixdx, drop, 'goods', kgrid=kgrid1, zdgridfile=zdf1,\
mcfile=mcfile, meankcorr=mk, add_interloper=True, norm=-1.)
model2 = bl.bivariate_lf(par, limits2, pixdx, drop, 'udf', kgrid=kgrid2, zdgridfile=zdf2,\
mcfile=mcfile, meankcorr=mk, add_interloper=True, norm=-1.)
sum1 = sum(model1.model.ravel()) * pixdx[0] * pixdx[1]
sum2 = sum(model2.model.ravel()) * pixdx[0] * pixdx[1]
phistar_mod = float(N1 + N2) / (sum1 + sum2)
print phistar_mod
model1.model = phistar_mod * model1.model
model2.model = phistar_mod * model2.model
logl_ref = bf.loglikelihood(data1, model1, floor=0.) + bf.loglikelihood(
data2, model2, floor=0.)
#logl_ref = bf.mlfunc(par, data1, data2, limits1, limits2, pixdx, kgrid1, kgrid2,
# 1.0, 1.0, -21.0, zdf1, zdf2, mcfile, 1, mk, 0, -1., 'phistar', drop)
print "logl_ref", logl_ref
simlogl_arr = zeros(niter) # actually -1*logL...
print "Start drawing simulated observations..."
t1 = time.time()
for i in range(niter):
if i % 1000 == 0: print i
simdata1 = mlutil.draw_from_pdf(N1, model1, model1.limits)
simdata2 = mlutil.draw_from_pdf(N2, model2, model2.limits)
simlogl1 = bf.loglikelihood(simdata1, model1)
simlogl2 = bf.loglikelihood(simdata2, model2)
simlogl = simlogl1 + simlogl2
simlogl_arr[i] = simlogl
t2 = time.time()
dt = t2 - t1
dtmin = int(floor(dt)) / 60
dtsec = dt % 60
n_worse = sum(simlogl_arr > logl_ref)
print "%d iterations took %d min %.1f sec" % (niter, dtmin, dtsec)
print "Percentage of simulated observations with lower log-likelihood: %.2f %%" % (
100. * float(n_worse) / float(niter))
return logl_ref, simlogl_arr
def reduced_chi2(par,
drop,
mbins1,
rbins1,
mbins2,
rbins2,
chisqnulim=(0.4, 5.0),
zlo=-1):
"""
Calculate the reduced chi2 of the best-fit model
mbins, rbins should include the upper limits.
"""
if drop == 'b':
cat1 = 'bdrops_gf_v2.cat'
cat2 = 'bdrops_udf_gf_v2.cat'
kgrid1 = mlutil.readkgrid('kernel_I.p')
kgrid2 = mlutil.readkgrid('kernel_I_udf.p')
zdf1 = 'zdgrid_bdrops.p'
zdf2 = 'zdgrid_bdrops_udf.p'
mcfile = 'M1500_to_i.txt'
mc = bl.mconvert(mcfile)
mk = mc(4.0)
elif drop == 'v':
cat1 = 'vdrops_gf_v2.cat'
cat2 = 'vdrops_udf_gf_v2.cat'
kgrid1 = mlutil.readkgrid('kernel_Z.p')
kgrid2 = mlutil.readkgrid('kernel_Z_udf.p')
zdf1 = 'zdgrid_vdrops.p'
zdf2 = 'zdgrid_vdrops_udf.p'
mcfile = 'M1500_to_z.txt'
mc = bl.mconvert(mcfile)
mk = mc(5.0)
cullflags = [0, 1, 2, 3, 4, 12, 13, 14, 18, 19]
limits1 = array([[21.0, 26.5], [-2.0, 3.0]])
limits2 = array([[23.0, 28.5], [-2.0, 3.0]])
pixdx = array([0.02, 0.02])
modshape1 = (limits1[:, 1] - limits1[:, 0]) / pixdx
modshape1 = around(modshape1).astype('int')
modshape2 = (limits2[:, 1] - limits2[:, 0]) / pixdx
modshape2 = around(modshape2).astype('int')
# bin the points & tally the counts
mag1, re1, crit1 = cleandata(cat1,
chisqnulim=chisqnulim[0],
magautolim=26.5,
cullflags=cullflags,
limits=limits1,
zlo=zlo)
mag2, re2, crit2 = cleandata(cat2,
chisqnulim=chisqnulim[1],
magautolim=28.5,
cullflags=cullflags,
limits=limits2,
zlo=zlo)
bincounts1 = histogram2d(mag1, log10(re1),
bins=[mbins1, rbins1])[0].astype('float')
bincounts2 = histogram2d(mag2, log10(re2),
bins=[mbins2, rbins2])[0].astype('float')
#print bincounts1, bincounts2
# calculate the best-fit models
model1 = bl.bivariate_lf(par, limits1, pixdx, kgrid=kgrid1, zdgridfile=zdf1, \
mcfile=mcfile, drop=drop, field='goods', meankcorr=mk, add_interloper=True)
model2 = bl.bivariate_lf(par, limits2, pixdx, kgrid=kgrid2, zdgridfile=zdf2, \
mcfile=mcfile, drop=drop, field='udf', meankcorr=mk, add_interloper=True)
phistar_mod = phistar(par, drop, zlo=zlo)
model1.model = phistar_mod * model1.model
model2.model = phistar_mod * model2.model
#model1.model = ones(modshape1)/(modshape1[0]*modshape1[1]*pixdx[0]*pixdx[1])*len(mag1)
#model2.model = ones(modshape2)/(modshape2[0]*modshape2[1]*pixdx[0]*pixdx[1])*len(mag2)
print sum(model1.model.ravel()) * pixdx[0] * pixdx[1]
chi2tot = 0.
nbins = 0
mindex1 = (mbins1 - 21.0) / 0.02
mindex1 = around(mindex1).astype('int')
rindex1 = (rbins1 - (-2.0)) / 0.02
rindex1 = around(rindex1).astype('int')
mindex2 = (mbins2 - 23.0) / 0.02
mindex2 = around(mindex2).astype('int')
rindex2 = (rbins2 - (-2.0)) / 0.02
rindex2 = around(rindex2).astype('int')
num_exp1 = [] # number of expected
num_exp2 = []
num_obs1 = bincounts1.ravel()[bincounts1.ravel() >= 5]
num_obs2 = bincounts2.ravel()[bincounts2.ravel() >= 5]
# iterate through all bins and calculate the chi2
for i in range(len(mbins1) - 1):
for j in range(len(rbins1) - 1):
if bincounts1[i, j] >= 5:
num_mod = sum(model1.model[mindex1[i]:mindex1[i + 1],
rindex1[j]:rindex1[j + 1]].ravel())
num_mod = num_mod * pixdx[0] * pixdx[1]
num_exp1 += [num_mod]
#chi2 = (bincounts1[i,j] - num_mod)**2 / num_mod
#print bincounts1[i,j], num_mod
#chi2tot += chi2
#nbins += 1
#if bincounts1[i,j] < nmin: nmin = bincounts1[i,j]
for i in range(len(mbins2) - 1):
for j in range(len(rbins2) - 1):
if bincounts2[i, j] >= 5:
num_mod = sum(model2.model[mindex2[i]:mindex2[i + 1],
rindex2[j]:rindex2[j + 1]].ravel())
num_mod = num_mod * pixdx[0] * pixdx[1]
num_exp2 += [num_mod]
#chi2 = (bincounts2[i,j] - num_mod)**2 / num_mod
#chi2tot += chi2
#nbins += 1
print "nbins", nbins
# Run chi-square test
num_exp = concatenate([num_exp1, num_exp2])
num_obs = concatenate([num_obs1, num_obs2])
ndeg = len(
num_exp) - 1 # degree of freedom = num. of contributing bins - 1
print "ndeg", ndeg
chi2, pval = stats.mstats.chisquare(num_obs, f_exp=num_exp)
print chi2, pval
#print num_exp, num_obs
#chi2nu = chi2tot / float(ndeg)
return chi2, pval, num_exp, num_obs
from numpy import *
import bivariate_lf as bl
import bivariate_fit as bf
import fit_lbg as fl
import mlutil
import zdist
import os, sys, time
from multiprocessing import Queue, Process, Pool
## initialize necessary things here
parv = array([-1.68527018, -20.50967054, 0.8757289, 0.85187255, 0.26594964])
dlimits1 = array([[24.0, 25.0], [-2.0, 3.0]])
dlimits2 = dlimits1.copy()
mlimits1 = array([[21.0, 26.5], [-2.0, 3.0]])
mlimits2 = array([[23.0, 28.5], [-2.0, 3.0]])
kgrid1 = mlutil.readkgrid('kernel_Z.p')
kgrid2 = mlutil.readkgrid('kernel_Z_udf.p')
zdgrid1 = zdist.read_zdgrid('zdgrid_vdrops.p')
zdgrid2 = zdist.read_zdgrid('zdgrid_vdrops_udf.p')
mc = bl.mconvert('M1500_to_z.txt')
logr0_arr = arange(0.7, 1.0, 0.005)
sigma_arr = arange(1.0, 1.3, 0.005)
logr0_grid, sigma_grid = meshgrid(logr0_arr, sigma_arr)
logr0_grid, sigma_grid = map(ravel, [logr0_grid, sigma_grid])
#print logr0_grid
nprocs = 3
#chunksize = float(niter) / nproc
q_logl = Queue()
q_pars = Queue()
par = parv.copy()
N = len(logr0_arr) * len(sigma_arr)
#!/usr/bin/env python
from numpy import *
from pygoods import *
import bivariate_lf as bl
import bivariate_fit as bf
import fit_lbg as fl
import mlutil
parb = array([-1.61711035, -20.53430348, 0.81505052, 0.76553555, 0.20435599])
parv = array([-1.68527018, -20.50967054, 0.8757289, 0.85187255, 0.26594964])
kgrid1 = mlutil.readkgrid('kernel_I.p')
kgrid2 = mlutil.readkgrid('kernel_I_udf.p')
kgrid3 = mlutil.readkgrid('kernel_Z.p')
kgrid4 = mlutil.readkgrid('kernel_Z_udf.p')
mci = bl.mconvert('M1500_to_i.txt')
mcz = bl.mconvert('M1500_to_z.txt')
def logl_sigma_bdrops(sarr=arange(0.3, 1.0, 0.05)):
mag1, re1, crit1 = fl.cleandata('bdrops_gf_v2.cat',
chisqnulim=0.4,
magautolim=26.5,
limits=bl.limits1,
drop='b')
mag2, re2, crit2 = fl.cleandata('bdrops_udf_gf_v2.cat',
chisqnulim=5.0,
magautolim=28.5,
limits=bl.limits2,
drop='b')
data1 = array([mag1, log10(re1)])
def plot_sizedist(parb, parv):
mod_lograrr = arange(bl.limits1[1][0], bl.limits1[1][1], 0.02)
magb1, reb1, critb1 = fl.cleandata('bdrops_gf_v2.cat',
drop='b',
limits=bl.limits1,
zlo=3.0)
magb2, reb2, critb2 = fl.cleandata('bdrops_udf_gf_v2.cat', chisqnulim=5.0, magautolim=28.5,\
limits=bl.limits2, drop='b', zlo=3.0)
kgridb1 = mlutil.readkgrid('kernel_I.p')
kgridb2 = mlutil.readkgrid('kernel_I_udf.p')
mci = bl.mconvert('M1500_to_i.txt')
reb = concatenate((reb1, reb2))
modelb1 = bl.bivariate_lf(parb,
bl.limits1,
bl.pixdx,
'b',
'goods',
kgrid=kgridb1,
zdgridfile='zdgrid_bdrops.p',
mcfile='M1500_to_i.txt',
meankcorr=mci(4.0))
modelb2 = bl.bivariate_lf(parb,
bl.limits2,
bl.pixdx,
'b',
'udf',
kgrid=kgridb2,
zdgridfile='zdgrid_bdrops_udf.p',
mcfile='M1500_to_i.txt',
meankcorr=mci(4.0))
sizedist_b = (modelb1.model.sum(axis=0) +
modelb2.model.sum(axis=0)) / 10.**mod_lograrr
magv1, rev1, critv1 = fl.cleandata('vdrops_gf_v2.cat',
chisqnulim=0.5,
drop='v',
limits=bl.limits1,
zlo=4.0)
magv2, rev2, critv2 = fl.cleandata('vdrops_udf_gf_v2.cat', chisqnulim=5.0, magautolim=28.5,\
limits=bl.limits2, drop='v', zlo=4.0)
rev = concatenate((rev1, rev2))
kgridv1 = mlutil.readkgrid('kernel_Z.p')
kgridv2 = mlutil.readkgrid('kernel_Z_udf.p')
mcz = bl.mconvert('M1500_to_z.txt')
modelv1 = bl.bivariate_lf(parv,
bl.limits1,
bl.pixdx,
'v',
'goods',
kgrid=kgridv1,
zdgridfile='zdgrid_vdrops.p',
mcfile='M1500_to_z.txt',
meankcorr=mcz(5.0))
modelv2 = bl.bivariate_lf(parv,
bl.limits2,
bl.pixdx,
'v',
'udf',
kgrid=kgridv2,
zdgridfile='zdgrid_vdrops_udf.p',
mcfile='M1500_to_z.txt',
meankcorr=mcz(5.0))
sizedist_v = (modelv1.model.sum(axis=0) +
modelb2.model.sum(axis=0)) / 10.**mod_lograrr
# fit the uncorrected size distribution with lognormal function
xout_b = fln.fit_lognormal(drop='b')
print xout_b
xout_v = fln.fit_lognormal(drop='v')
print xout_v
# plot
fig = plt.figure(figsize=(8, 10))
ax1 = fig.add_subplot(211)
rarr = arange(0.001, 41., 1.)
pl_rarr = arange(0.001, 41., 0.01)
h1 = ax1.hist(reb, rarr, color='gray', ec='none')
fb = fln.lognormal(xout_b[0], xout_b[1], pl_rarr)
ax1.plot(pl_rarr,
fb * max(h1[0]) / max(fb),
color='black',
label=r'$\sigma=%.2f$' % xout_b[1])
ax1.plot(10.**mod_lograrr,
sizedist_b * max(h1[0]) / max(sizedist_b),
color='red',
label=r'$\sigma=%.2f$; corrected' % parb[3])
ax2 = fig.add_subplot(212)
h2 = ax2.hist(rev, rarr, color='gray', ec='none')
fv = fln.lognormal(xout_v[0], xout_v[1], pl_rarr)
ax2.plot(pl_rarr,
fv * max(h2[0]) / max(fv),
color='black',
label=r'$\sigma=%.2f$' % xout_v[1])
ax2.plot(10.**mod_lograrr,
sizedist_v * max(h2[0]) / max(sizedist_v),
color='red',
label=r'$\sigma=%.2f$; corrected' % parv[3])
ax1.set_xlim(0, 40)
ax2.set_xlim(0, 40)
ax1.set_xlabel('Re [0.03" / pixel]')
ax2.set_xlabel('Re [0.03" / pixel]')
ax1.legend(loc=1)
ax2.legend(loc=1)
ax1.set_title('B-dropouts (z~4)')
ax2.set_title('V-dropouts (z~5)')
return fig
def show_model(par,
drop,
field,
newfig=True,
axCent=None,
fig1=None,
lfbw=0.2,
sdbw=0.2,
colors=['blue', 'green', 'black', 'red']):
if drop == 'b':
zmean = 4.0
mc = bl.mconvert('M1500_to_i.txt')
if field == 'goods':
cat = 'bdrops_gf_v3.cat'
zdgrid = zdist.read_zdgrid('zdgrid/zdgrid_bdrops_nolya.p')
limits = limitsb1
magautolim = 26.5
chisqnulim = 0.4
reerrlim = 0.6
kgridfile = 'tfkernel/kernel_I.p'
dataset = 'GOODS'
elif field == 'udf':
cat = 'bdrops_udf_gf_v3.cat'
zdgrid = zdist.read_zdgrid('zdgrid/zdgrid_bdrops_udf_nolya.p')
limits = limitsb2
magautolim = 28.5
chisqnulim = 5.0
reerrlim = 0.6
kgridfile = 'tfkernel/kernel_I_udf.p'
dataset = 'HUDF'
elif drop == 'v':
zmean = 5.0
mc = bl.mconvert('M1500_to_z.txt')
if field == 'goods':
cat = 'vdrops_gf_v2.cat'
zdgrid = zdist.read_zdgrid('zdgrid/zdgrid_vdrops_nolya_bston5.p')
limits = limitsv1
magautolim = 26.5
chisqnulim = 0.5
reerrlim = 0.6
kgridfile = 'tfkernel/kernel_Z.p'
dataset = 'GOODS'
elif field == 'udf':
cat = 'vdrops_udf_gf_v2.cat'
zdgrid = zdist.read_zdgrid(
'zdgrid/zdgrid_vdrops_udf_nolya_bston5.p')
limits = limitsv2
magautolim = 28.5
chisqnulim = 5.0
reerrlim = 0.6
kgridfile = 'tfkernel/kernel_Z_udf.p'
dataset = 'HUDF'
kgrid = mlutil.readkgrid(kgridfile)
meankcorr = mc(zmean)
fp = matplotlib.font_manager.FontProperties(size=9)
if newfig:
fig1 = plt.figure(figsize=(10, 15))
axCent = plt.subplot(111)
divider = make_axes_locatable(axCent)
restlim = array([[-26.5, -15.5], [-0.5, 2.0]])
pixdx = array([0.02, 0.02])
if drop == 'b': z0 = 3.0
else: z0 = 4.0
zd_flat = zdist.zdgrid(-25.0, -15.0, 0.5, -0.6, 1.8, 0.2, z0, 6.0, 0.1,
drop, zdgrid.area)
zd_flat.flat_zdgrid(zlo=zmean - 0.5, zhi=zmean + 0.5)
f = open('zdgrid_flat.p', 'w')
cPickle.dump(zd_flat, f, 2)
f.close()
model0 = bl.bivariate_lf(par, limits, pixdx, drop, field, mc=mc, meankcorr=meankcorr,\
zdgrid=None)
V0 = zdgrid.dVdz[(zdgrid.zarr >= (zmean - 0.5))
& (zdgrid.zarr <= (zmean + 0.5))]
model0.model = model0.model * sum(V0)
model_ds = bl.bivariate_lf(par, limits, pixdx, drop, field, kgrid=None, meankcorr=meankcorr,\
zdgrid=zdgrid, mc=mc, norm=-1)
model = bl.bivariate_lf(par,
limits,
pixdx,
drop,
field,
kgrid=kgrid,
meankcorr=meankcorr,
M0=-21.0,
zdgrid=zdgrid,
mc=mc,
norm=-1)
# Show model + data points
#axCent.imshow(model.model.swapaxes(0,1), origin = 'lower', vmin = vmin, vmax = vmax,
# aspect = 'auto')
mag, re, crit = fl.cleandata(cat,
chisqnulim=chisqnulim,
reerr_ratiolim=reerrlim,
limits=limits,
drop=drop)
npts = len(mag)
print npts
axCent.scatter(mag, log10(re), s=4, color='black')
axCent.contour(arange(limits[0][0],limits[0][1],pixdx[0]),\
arange(limits[1][0],limits[1][1],pixdx[1]),model.model.swapaxes(0,1),6,colors=colors[0])
yf = 0.5
axCent.set_yticks(arange(limits[1][0], limits[1][1], 0.5))
axCent.set_yticklabels(arange(limits[1][0], limits[1][1], 0.5))
if drop == 'b':
axCent.set_xlabel(r'GALFIT MAG in $i_{775}$')
axCent.set_ylabel(r'GALFIT $\log_{10}(R_e)$ in $i_{775}$ [pixel]')
elif drop == 'v':
axCent.set_xlabel(r'GALFIT MAG in $z_{850}$')
axCent.set_ylabel(r'GALFIT $\log_{10}(R_e)$ in $z_{850}$ [pixel]')
#plt.suptitle(r'$\vec{\mathbf{P}}=[%.2f, %.2f, %.2f$",$ %.2f, %.2f]$' % (par[0],par[1],
# 10.**par[2]*0.03,par[3],par[4]),size=20)
#axCent.text(0.1,0.1,"par = [%.2f,%.2f,%.2f,%.2f,%.2f]"%tuple(par),transform=axCent.transAxes,
# color='black')
# plot a straight line through the observed points
ydata = findline(mag, log10(re))
xr = arange(limits[0][0], limits[0][1], 0.02)
axCent.plot(xr, ydata[0] * xr + ydata[1], ':', c='black')
# plot the straight line corresponding to the power-law relation with logR0 and beta
m0 = -21.0 + mc(zmean)
b = par[2] + 0.4 * par[4] * m0
axCent.plot(xr, -0.4 * par[4] * xr + b, '--', lw=2.5, c=colors[3])
axCent.text(0.1,
0.9,
dataset,
transform=axCent.transAxes,
color='black',
size=14)
# Show LF
axLF = divider.append_axes("top", size=1.2, pad=0.0, sharex=axCent)
n, bins = histogram(mag, arange(limits[0, 0], limits[0, 1] + lfbw, lfbw))
nerr = [sqrt(n), sqrt(n)]
for i in range(len(n)):
if n[i] == 1: nerr[0][i] = 1. - 1.e-3
axLF.errorbar(bins[:-1]+lfbw/2., n, yerr=nerr, fmt='.', ms=14.,\
mfc='black', ls='None', mec='black', ecolor='black', capsize=6)
LF = model.model.sum(axis=1) # LF here contains the volume already
LFtot = sum(LF) * pixdx[0] / lfbw
normfactor = npts / LFtot # normalize the LF to predict the total number of points
LF = LF * normfactor
LF0 = model0.model.sum(axis=1)
LF0 = LF0 * normfactor
LF_ds = model_ds.model.sum(axis=1)
LF_ds = LF_ds * normfactor
axLF.semilogy(arange(limits[0,0],limits[0,1],pixdx[0]), LF, color = colors[2],\
nonposy='mask', label='GALFIT TF')
axLF.semilogy(arange(limits[0,0],limits[0,1],pixdx[0]), LF_ds, color = colors[1],\
ls=':',lw=2,nonposy='mask',label='w/ dropout sel. kernel')
axLF.semilogy(arange(limits[0,0],limits[0,1],pixdx[0]), LF0, color=colors[0],\
ls='--',lw=2,nonposy='mask',label='Schechter')
axLF.set_yticks([1.e-2, 1., 1.e2])
axLF.set_ylim(1.e-2, max(n) * 50.)
xf = 1.0
axCent.set_xticks(arange(limits[0][0], limits[0][1] + 1., 1.))
axCent.set_xticklabels(arange(limits[0][0], limits[0][1] + 1., 1.))
axCent.set_xlim(limits[0][0], limits[0][1])
#axLF.legend(loc='upper left', prop=fp)
# Show size distribution
axSD = divider.append_axes("right", size="35%", pad=0.0, sharey=axCent)
n, bins = histogram(log10(re),
arange(limits[1, 0], limits[1, 1] + sdbw, sdbw))
nerr = [sqrt(n), sqrt(n)]
for i in range(len(n)):
if n[i] == 1: nerr[0][i] = 1. - 1.e-3
axSD.errorbar(n, bins[:-1]+sdbw/2., xerr=nerr, fmt='.', ms=14,\
mfc='black', ls='None', mec='black', ecolor='black', capsize=6)
SD = model.model.sum(axis=0)
SDtot = sum(SD) * pixdx[1] / sdbw
normfactor = npts / SDtot
SD = SD * normfactor
sizer = arange(limits[1, 0], limits[1, 1], pixdx[1])
axSD.semilogx(SD, sizer, color=colors[2], label='GALFIT TF')
SD0 = model0.model.sum(axis=0)
SD0 = SD0 * normfactor
SD_ds = model_ds.model.sum(axis=0)
SD_ds = SD_ds * normfactor
axSD.semilogx(SD_ds,
sizer,
color=colors[1],
ls=':',
lw=2,
label='w/ dropout sel. kernel')
axSD.semilogx(SD0,
sizer,
color=colors[0],
ls='--',
lw=2,
label='lognormal')
axSD.set_xticks([1., 10., 1.e2])
axSD.set_xlim(1.e-2, max(SD) * 5)
#axSD.legend(loc='lower right', prop=fp)
#axCent.set_ylim(limits[1][0], limits[1][1])
axCent.set_ylim(-0.6, 1.8)
plt.draw()
fig1.show()
for tl in axLF.get_xticklabels():
tl.set_visible(False)
for tl in axSD.get_yticklabels():
tl.set_visible(False)
return model, axCent, axSD, axLF, fig1