def fit_spectrophotometry(igal,
sim='lgal',
noise='bgs0_legacy',
method='ifsps',
model='emulator',
nwalkers=100,
burnin=100,
niter='adaptive',
maxiter=200000,
opt_maxiter=100,
overwrite=False,
postprocess=False,
justplot=False):
''' Fit Lgal spectra. `noise` specifies whether to fit spectra without noise or
with BGS-like noise. Produces an MCMC chain and, if not on nersc, a corner plot of the posterior.
:param igal:
index of Lgal galaxy within the spectral_challenge
:param noise:
If 'bgs1'...'bgs8', fit BGS-like spectra. (default: 'none')
:param justplot:
If True, skip the fitting and plot the best-fit. This is mainly implemented
because I'm having issues plotting in NERSC. (default: False)
'''
noise_spec = noise.split('_')[0]
noise_photo = noise.split('_')[1]
# read noiseless Lgal spectra of the spectral_challenge mocks
specs, meta = Data.Spectra(sim=sim,
noise=noise_spec,
lib='bc03',
sample='mini_mocha')
# read Lgal photometry of the mini_mocha mocks
photo, _ = Data.Photometry(sim=sim,
noise=noise_photo,
lib='bc03',
sample='mini_mocha')
if meta['redshift'][igal] < 0.101:
# current Speculator wavelength doesn't extend far enough
# current Speculator wavelength doesn't extend far enough
# current Speculator wavelength doesn't extend far enough
# current Speculator wavelength doesn't extend far enough
# current Speculator wavelength doesn't extend far enough
return None
w_obs = specs['wave']
flux_obs = specs['flux'][igal]
ivar_obs = specs['ivar'][igal]
if method == 'ispeculator':
photo_obs = photo['flux'][igal, :3]
photo_ivar_obs = photo['ivar'][igal, :3]
else:
photo_obs = photo['flux'][igal, :5]
photo_ivar_obs = photo['ivar'][igal, :5]
# get fiber flux factor prior range based on measured fiber flux
f_fiber_true = (photo['fiberflux_r_true'][igal] /
photo['flux_r_true'][igal])
prior_width = np.max([
0.05,
5. * photo['fiberflux_r_ivar'][igal]**-0.5 / photo['flux'][igal, 1]
])
f_fiber_min = (
photo['fiberflux_r_meas'][igal]) / photo['flux'][igal, 1] - prior_width
f_fiber_max = (
photo['fiberflux_r_meas'][igal]) / photo['flux'][igal, 1] + prior_width
f_fiber_prior = [f_fiber_min, f_fiber_max]
print('--- input ---')
print('z = %f' % meta['redshift'][igal])
print('log M* total = %f' % meta['logM_total'][igal])
print('log M* fiber = %f' % meta['logM_fiber'][igal])
print('f_fiber = %f' % f_fiber_true)
print('log SFR 100myr = %f' % np.log10(meta['sfr_100myr'][igal]))
print('log Z_MW = %f' % np.log10(meta['Z_MW'][igal]))
f_mcmc = os.path.join(
UT.dat_dir(), 'mini_mocha', method,
'%s.specphoto.noise_%s.%s.%i.mcmc.hdf5' % (sim, noise, model, igal))
if method == 'ifsps':
ifitter = Fitters.iFSPS(model_name=model)
elif method == 'ispeculator':
ifitter = Fitters.iSpeculator(model_name=model)
print('--- mcmc ---')
if (justplot or not overwrite) and os.path.isfile(f_mcmc):
print(' reading... %s' % os.path.basename(f_mcmc))
# read in mcmc chain
fmcmc = h5py.File(f_mcmc, 'r')
mcmc = {}
for k in fmcmc.keys():
mcmc[k] = fmcmc[k][...]
fmcmc.close()
else:
if os.path.isfile(f_mcmc) and not overwrite:
print("** CAUTION: %s already exists **" %
os.path.basename(f_mcmc))
print(' writing... %s' % os.path.basename(f_mcmc))
# initiating fit
prior = ifitter._default_prior(f_fiber_prior=f_fiber_prior)
mcmc = ifitter.MCMC_spectrophoto(w_obs,
flux_obs,
ivar_obs,
photo_obs,
photo_ivar_obs,
meta['redshift'][igal],
mask='emline',
prior=prior,
nwalkers=nwalkers,
burnin=burnin,
niter=niter,
maxiter=maxiter,
opt_maxiter=opt_maxiter,
writeout=f_mcmc,
silent=False)
if postprocess:
print('--- postprocessing ---')
f_post = f_mcmc.replace('.mcmc.hdf5', '.postproc.hdf5')
mcmc = ifitter.postprocess(mcmc_output=mcmc, writeout=f_post)
i_fib = list(mcmc['theta_names'].astype(str)).index('f_fiber')
print('log M* total = %f' % mcmc['theta_med'][0])
print('log M* fiber = %f' %
(mcmc['theta_med'][0] + np.log10(mcmc['theta_med'][i_fib])))
print('---------------')
try:
# plotting on nersc never works.
if os.environ['NERSC_HOST'] == 'cori': return None
except KeyError:
labels = [lbl_dict[_t] for _t in mcmc['theta_names'].astype(str)]
truths = [None for _ in labels]
truths[0] = meta['logM_total'][igal]
truths[i_fib] = f_fiber_true
if postprocess:
i_sfr = list(
mcmc['theta_names'].astype(str)).index('logsfr.100myr')
truths[i_sfr] = np.log10(meta['sfr_100myr'][igal])
i_zw = list(mcmc['theta_names'].astype(str)).index('logz.mw')
truths[i_zw] = np.log10(meta['Z_MW'][igal])
print('log SFR 100myr = %f' %
np.median(mcmc['mcmc_chain'][:, i_sfr]))
print('log Z_MW = %f' % np.median(mcmc['mcmc_chain'][:, i_zw]))
# corner plot of the posteriors
fig = DFM.corner(mcmc['mcmc_chain'],
range=mcmc['prior_range'],
quantiles=[0.16, 0.5, 0.84],
levels=[0.68, 0.95],
nbin=40,
smooth=True,
truths=truths,
labels=labels,
label_kwargs={'fontsize': 20})
if postprocess:
fig.savefig(f_post.replace('.hdf5', '.png'), bbox_inches='tight')
else:
fig.savefig(f_mcmc.replace('.hdf5', '.png'), bbox_inches='tight')
plt.close()
fig = plt.figure(figsize=(18, 5))
gs = mpl.gridspec.GridSpec(1,
2,
figure=fig,
width_ratios=[1, 3],
wspace=0.2)
sub = plt.subplot(gs[0])
sub.errorbar(np.arange(len(photo_obs)),
photo_obs,
yerr=photo_ivar_obs**-0.5,
fmt='.k',
label='data')
sub.scatter(np.arange(len(photo_obs)),
mcmc['flux_photo_model'],
c='C1',
label='model')
sub.legend(loc='upper left',
markerscale=3,
handletextpad=0.2,
fontsize=15)
sub.set_xticks([0, 1, 2, 3, 4])
sub.set_xticklabels(['$g$', '$r$', '$z$', 'W1', 'W2'])
sub.set_xlim(-0.5, len(photo_obs) - 0.5)
sub = plt.subplot(gs[1])
sub.plot(w_obs, flux_obs, c='k', lw=1, label='data')
sub.plot(mcmc['wavelength_model'],
mcmc['flux_spec_model'],
c='C1',
ls='--',
lw=1,
label=method)
sub.legend(loc='upper right', fontsize=15)
sub.set_xlabel('wavelength [$A$]', fontsize=20)
sub.set_xlim(3600., 9800.)
sub.set_ylim(-1., 5.)
if postprocess:
fig.savefig(f_post.replace('.hdf5', '.bestfit.png'),
bbox_inches='tight')
else:
fig.savefig(f_mcmc.replace('.hdf5', '.bestfit.png'),
bbox_inches='tight')
plt.close()
return None
def fit_photometry(igal,
sim='lgal',
noise='legacy',
method='ifsps',
model='emulator',
nwalkers=100,
burnin=100,
niter='adaptive',
maxiter=200000,
opt_maxiter=100,
overwrite=False,
postprocess=False,
justplot=False):
''' Fit simulated photometry. `noise` specifies whether to fit spectra without noise or
with legacy-like noise. `dust` specifies whether to if spectra w/ dust or not.
Produces an MCMC chain and, if not on nersc, a corner plot of the posterior.
:param igal:
index of Lgal galaxy within the spectral_challenge
:param noise:
If 'none', fit noiseless photometry.
If 'legacy', fit Legacy-like photometry. (default: 'none')
:param justplot:
If True, skip the fitting and plot the best-fit. This is mainly implemented
because I'm having issues plotting in NERSC. (default: False)
'''
# read Lgal photometry of the mini_mocha mocks
photo, meta = Data.Photometry(sim=sim,
noise=noise,
lib='bc03',
sample='mini_mocha')
if meta['redshift'][igal] < 0.101:
# current Speculator wavelength doesn't extend far enough
# current Speculator wavelength doesn't extend far enough
# current Speculator wavelength doesn't extend far enough
# current Speculator wavelength doesn't extend far enough
# current Speculator wavelength doesn't extend far enough
return None
print('--- input ---')
print('z = %f' % meta['redshift'][igal])
print('log M* total = %f' % meta['logM_total'][igal])
print('log SFR 100myr = %f' % np.log10(meta['sfr_100myr'][igal]))
print('log Z MW = %f' % np.log10(meta['Z_MW'][igal]))
if method == 'ispeculator':
photo_obs = photo['flux'][igal, :3]
ivar_obs = photo['ivar'][igal, :3]
else:
photo_obs = photo['flux'][igal, :5]
ivar_obs = photo['ivar'][igal, :5]
f_mcmc = os.path.join(
UT.dat_dir(), 'mini_mocha', method,
'%s.photo.noise_%s.%s.%i.mcmc.hdf5' % (sim, noise, model, igal))
# initiating fit
if method == 'ifsps':
ifitter = Fitters.iFSPS(model_name=model)
elif method == 'ispeculator':
ifitter = Fitters.iSpeculator(model_name=model)
print('--- bestfit ---')
if (justplot or not overwrite) and os.path.isfile(f_mcmc):
# read in best-fit file with mcmc chain
print(' reading in %s' % f_mcmc)
fmcmc = h5py.File(f_mcmc, 'r')
mcmc = {}
for k in fmcmc.keys():
mcmc[k] = fmcmc[k][...]
fmcmc.close()
else:
if os.path.isfile(f_mcmc) and not overwrite:
print("** CAUTION: %s already exists **" % os.path.basename(f_bf))
print(' writing %s' % f_mcmc)
prior = ifitter._default_prior(f_fiber_prior=None)
mcmc = ifitter.MCMC_photo(photo_obs,
ivar_obs,
meta['redshift'][igal],
bands='desi',
prior=prior,
nwalkers=nwalkers,
burnin=burnin,
niter=niter,
maxiter=maxiter,
opt_maxiter=opt_maxiter,
writeout=f_mcmc,
silent=False)
if postprocess:
print('--- postprocessing ---')
f_post = f_mcmc.replace('.mcmc.hdf5', '.postproc.hdf5')
mcmc = ifitter.postprocess(mcmc_output=mcmc, writeout=f_post)
print('log M* total = %f' % mcmc['theta_med'][0])
print('---------------')
try:
# plotting on nersc never works.
if os.environ['NERSC_HOST'] == 'cori': return None
except KeyError:
labels = [lbl_dict[_t] for _t in mcmc['theta_names'].astype(str)]
truths = [None for _ in labels]
truths[0] = meta['logM_total'][igal]
if postprocess:
i_sfr = list(
mcmc['theta_names'].astype(str)).index('logsfr.100myr')
truths[i_sfr] = np.log10(meta['sfr_100myr'][igal])
i_zmw = list(mcmc['theta_names'].astype(str)).index('logz.mw')
truths[i_zmw] = np.log10(meta['Z_MW'][igal])
print('log SFR 100myr = %f' %
np.median(mcmc['mcmc_chain'][:, i_sfr]))
print('log Z MW = %f' % np.median(mcmc['mcmc_chain'][:, i_zmw]))
fig = DFM.corner(mcmc['mcmc_chain'],
range=mcmc['prior_range'],
quantiles=[0.16, 0.5, 0.84],
levels=[0.68, 0.95],
nbin=40,
smooth=True,
truths=truths,
labels=labels,
label_kwargs={'fontsize': 20})
if postprocess:
fig.savefig(f_post.replace('.hdf5', '.png'), bbox_inches='tight')
else:
fig.savefig(f_mcmc.replace('.hdf5', '.png'), bbox_inches='tight')
plt.close()
fig = plt.figure(figsize=(5, 3))
sub = fig.add_subplot(111)
sub.errorbar(np.arange(len(photo_obs)),
photo_obs,
yerr=ivar_obs**-0.5,
fmt='.k',
label='data')
sub.scatter(np.arange(len(photo_obs)),
mcmc['flux_photo_model'],
c='C1',
label='model')
sub.legend(loc='upper left',
markerscale=3,
handletextpad=0.2,
fontsize=15)
sub.set_xticks([0, 1, 2])
sub.set_xticklabels(['$g$', '$r$', '$z$'])
sub.set_xlim(-0.5, len(photo_obs) - 0.5)
if postprocess:
fig.savefig(f_post.replace('.hdf5', '.bestfit.png'),
bbox_inches='tight')
else:
fig.savefig(f_mcmc.replace('.hdf5', '.bestfit.png'),
bbox_inches='tight')
plt.close()
return None
def fit_spectrophotometry(igal, noise='none', nwalkers=100, burnin=100, niter=1000, overwrite=False, justplot=False):
''' Fit Lgal spectra. `noise` specifies whether to fit spectra without noise or
with BGS-like noise. Produces an MCMC chain and, if not on nersc, a corner plot of the posterior.
:param igal:
index of Lgal galaxy within the spectral_challenge
:param noise:
If 'none', fit noiseless spectra.
If 'bgs1'...'bgs8', fit BGS-like spectra. (default: 'none')
:param justplot:
If True, skip the fitting and plot the best-fit. This is mainly implemented
because I'm having issues plotting in NERSC. (default: False)
'''
if noise != 'none':
noise_spec = noise.split('_')[0]
noise_photo = noise.split('_')[1]
else:
noise_spec = 'none'
noise_photo = 'none'
# read noiseless Lgal spectra of the spectral_challenge mocks
specs, meta = Data.Spectra(sim='lgal', noise=noise_spec, lib='bc03', sample='mini_mocha')
# read Lgal photometry of the mini_mocha mocks
photo, _ = Data.Photometry(sim='lgal', noise=noise_photo, lib='bc03', sample='mini_mocha')
model = 'vanilla'
w_obs = specs['wave']
flux_obs = specs['flux'][igal]
if noise_spec != 'none':
ivar_obs = specs['ivar'][igal]
else:
ivar_obs = np.ones(len(w_obs))
photo_obs = photo['flux'][igal,:5]
if noise_photo != 'none':
photo_ivar_obs = photo['ivar'][igal,:5]
else:
photo_ivar_obs = np.ones(photo_obs.shape[0])
# get fiber flux factor prior range based on measured fiber flux
f_fiber_true = (photo['fiberflux_r_meas'][igal]/photo['flux_r_true'][igal])
f_fiber_min = (photo['fiberflux_r_meas'][igal] - 3.*photo['fiberflux_r_ivar'][igal]**-0.5)/photo['flux'][igal,1]
f_fiber_max = (photo['fiberflux_r_meas'][igal] + 3.*photo['fiberflux_r_ivar'][igal]**-0.5)/photo['flux'][igal,1]
f_fiber_prior = [f_fiber_min, f_fiber_max]
print(f_fiber_prior)
print(f_fiber_true)
print((photo['fiberflux_r_true'][igal]/photo['flux_r_true'][igal]))
truths = [meta['logM_total'][igal], np.log10(meta['Z_MW'][igal]), meta['t_age_MW'][igal], None, None, f_fiber_true]
labels = ['$\log M_*$', '$\log Z$', r'$t_{\rm age}$', 'dust2', r'$\tau$', r'$f_{\rm fiber}$']
print('--- input ---')
print('z = %f' % meta['redshift'][igal])
print('log M* total = %f' % meta['logM_total'][igal])
print('log M* fiber = %f' % meta['logM_fiber'][igal])
print('MW Z = %f' % meta['Z_MW'][igal])
print('MW tage = %f' % meta['t_age_MW'][igal])
f_bf = os.path.join(UT.dat_dir(), 'mini_mocha', 'ifsps', 'lgal.specphoto.noise_%s.%s.%i.hdf5' % (noise, model, igal))
if not justplot:
if os.path.isfile(f_bf):
if not overwrite:
print("** CAUTION: %s already exists **" % os.path.basename(f_bf))
# initiating fit
ifsps = Fitters.iFSPS(model_name=model, prior=None)
bestfit = ifsps.MCMC_spectrophoto(
w_obs,
flux_obs,
ivar_obs,
photo_obs,
photo_ivar_obs,
meta['redshift'][igal],
f_fiber_prior=f_fiber_prior,
mask='emline',
nwalkers=nwalkers,
burnin=burnin,
niter=niter,
writeout=f_bf,
silent=False)
else:
# read in best-fit file with mcmc chain
fbestfit = h5py.File(f_bf, 'r')
bestfit = {}
for k in fbestfit.keys():
bestfit[k] = fbestfit[k][...]
print('--- bestfit ---')
print('written to %s' % f_bf)
print('log M* total = %f' % bestfit['theta_med'][0])
print('log M* fiber = %f' % (bestfit['theta_med'][0] + np.log10(bestfit['theta_med'][-1])))
print('log Z = %f' % bestfit['theta_med'][1])
print('---------------')
try:
# plotting on nersc never works.
if os.environ['NERSC_HOST'] == 'cori': return None
except KeyError:
# corner plot of the posteriors
fig = DFM.corner(bestfit['mcmc_chain'], range=bestfit['priors'], quantiles=[0.16, 0.5, 0.84],
levels=[0.68, 0.95], nbin=40, smooth=True,
truths=truths, labels=labels, label_kwargs={'fontsize': 20})
fig.savefig(f_bf.replace('.hdf5', '.png'), bbox_inches='tight')
return None
spec_obs_non = specs_non['flux'][0]
spec_ivar_obs_non = np.ones(w_obs_non.shape[0])
t0 = time.time()
# read Lgal photometry of the mini_mocha mocks
photo, _ = Data.Photometry(sim='lgal',
noise='legacy',
lib='bc03',
sample='mini_mocha')
print('Data.Photometry takes %.3f sec' % (time.time() - t0)) # 0.019sec
photo_obs = photo['flux'][0, :5]
photo_ivar_obs = photo['ivar'][0, :5]
# initial fitter
ifsps = Fitters.iFSPS(model_name='vanilla', prior=None)
# middle of prior theta value
theta_test = np.average(np.array(ifsps.priors), axis=1)
# get mask
t0 = time.time()
_mask = ifsps._check_mask('emline', w_obs, spec_ivar_obs,
meta['redshift'][0])
print('iFSPS._check_mask takes %.3f sec' % (time.time() - t0)) # <0.001sec
ifsps._lnPost_spec(theta_test,
w_obs,
spec_obs,
spec_ivar_obs,
meta['redshift'][0],
mask=_mask,
def fit_photometry(igal, noise='none', nwalkers=100, burnin=100, niter=1000, overwrite=False, justplot=False):
''' Fit Lgal photometry. `noise` specifies whether to fit spectra without noise or
with legacy-like noise. `dust` specifies whether to if spectra w/ dust or not.
Produces an MCMC chain and, if not on nersc, a corner plot of the posterior.
:param igal:
index of Lgal galaxy within the spectral_challenge
:param noise:
If 'none', fit noiseless photometry.
If 'legacy', fit Legacy-like photometry. (default: 'none')
:param justplot:
If True, skip the fitting and plot the best-fit. This is mainly implemented
because I'm having issues plotting in NERSC. (default: False)
'''
# read Lgal photometry of the mini_mocha mocks
photo, meta = Data.Photometry(sim='lgal', noise=noise, lib='bc03', sample='mini_mocha')
model = 'vanilla'
photo_obs = photo['flux'][igal,:5]
if noise != 'none': ivar_obs = photo['ivar'][igal,:5]
truths = [meta['logM_total'][igal], np.log10(meta['Z_MW'][igal]), meta['t_age_MW'][igal], None, None]
labels = ['$\log M_*$', '$\log Z$', r'$t_{\rm age}$', 'dust2', r'$\tau$']
if noise == 'none': # no noise
ivar_obs = np.ones(photo_obs.shape[0])
print('--- input ---')
print('z = %f' % meta['redshift'][igal])
print('log M* total = %f' % meta['logM_total'][igal])
print('MW Z = %f' % meta['Z_MW'][igal])
print('MW tage = %f' % meta['t_age_MW'][igal])
f_bf = os.path.join(UT.dat_dir(), 'mini_mocha', 'ifsps', 'lgal.photo.noise_%s.%s.%i.hdf5' % (noise, model, igal))
if not justplot:
if os.path.isfile(f_bf):
if not overwrite:
print("** CAUTION: %s already exists **" % os.path.basename(f_bf))
return None
# initiate fitting
ifsps = Fitters.iFSPS(model_name=model, prior=None)
bestfit = ifsps.MCMC_photo(
photo_obs,
ivar_obs,
meta['redshift'][igal],
bands='desi',
nwalkers=nwalkers,
burnin=burnin,
niter=niter,
writeout=f_bf,
silent=False)
else:
# read in best-fit file with mcmc chain
fbestfit = h5py.File(f_bf, 'r')
bestfit = {}
for k in fbestfit.keys():
bestfit[k] = fbestfit[k][...]
print('--- bestfit ---')
print('written to %s ---' % f_bf)
print('log M* = %f' % bestfit['theta_med'][0])
print('log Z = %f' % bestfit['theta_med'][1])
print('---------------')
try:
# plotting on nersc never works.
if os.environ['NERSC_HOST'] == 'cori': return None
except KeyError:
fig = DFM.corner(bestfit['mcmc_chain'], range=bestfit['priors'], quantiles=[0.16, 0.5, 0.84],
levels=[0.68, 0.95], nbin=40, smooth=True,
truths=truths, labels=labels, label_kwargs={'fontsize': 20})
fig.savefig(f_bf.replace('.hdf5', '.png'), bbox_inches='tight')
return None