Skip to content

Commit

Permalink
initial effort to use chi2 scan for vdisp
Browse files Browse the repository at this point in the history
  • Loading branch information
@moustakas
moustakas committed Nov 8, 2024
1 parent 9ecd526 commit c3b4047
Showing 1 changed file with 99 additions and 29 deletions.
128 changes: 99 additions & 29 deletions py/fastspecfit/continuum.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -23,8 +23,8 @@ class ContinuumTools(object):
"""
def __init__(self, data, templates, phot, igm, tauv_guess=0.1,
vdisp_guess=250., tauv_bounds=(0., 2.),
vdisp_bounds=(75., 500.), fastphot=False,
constrain_age=False):
vdisp_bounds=(75., 500.), vdisp_nbin=25,
fastphot=False, constrain_age=False):

self.phot = phot
self.templates = templates
Expand All @@ -35,6 +35,7 @@ def __init__(self, data, templates, phot, igm, tauv_guess=0.1,
self.vdisp_guess = vdisp_guess
self.tauv_bounds = tauv_bounds
self.vdisp_bounds = vdisp_bounds
self.vdisp_grid = np.linspace(vdisp_bounds[0], vdisp_bounds[1], vdisp_nbin)

# Cache the redshift-dependent factors (incl. IGM attenuation),
redshift = data['redshift']
Expand Down Expand Up @@ -1087,13 +1088,14 @@ def continuum_fastphot(redshift, objflam, objflamivar, CTools,
sedmodel_nolines_monte)


def continuum_fastspec(redshift, objflam, objflamivar, CTools,
nmonte=50, rng=None, uniqueid=0,
no_smooth_continuum=False, debug_plots=False,
fit_for_vdisp=False):
def continuum_fastspec(redshift, objflam, objflamivar, CTools, nmonte=50,
rng=None, uniqueid=0, no_smooth_continuum=False,
vdisp_chi2scan=True, debug_plots=False):
"""Jointly model the spectroscopy and broadband photometry.
"""
from fastspecfit.util import find_minima, minfit

data = CTools.data
phot = CTools.phot
templates = CTools.templates
Expand Down Expand Up @@ -1158,29 +1160,79 @@ def continuum_fastspec(redshift, objflam, objflamivar, CTools,
# should be all zeros
objflam_monte = np.repeat(objflam[np.newaxis, :], nmonte, axis=1)

# Perform an initial fit to the data using the cached templates with
# nominal velocity dispersion. Reduce the prior bound on tauv to minimize
# the age-dust degeneracy, which is especially important when trying to
# estimate the velocity dispersion in the next call to
# fit_stellar_continuum. When we fit the full spectrum+photometry (at fixed
# velocity dispersion), we revert to the larger (default) tauv prior, since
# the IR fluxes put an additional constraint on the dust content.
init_tauv_bounds = (0., 1.)
tauv_nomvdisp, _, coeff_nomvdisp, resid_nomvdisp = CTools.fit_stellar_continuum(
templates.flux_nolines_nomvdisp[agekeep, :], fit_vdisp=False, conv_pre=None,
tauv_bounds=init_tauv_bounds, specflux=specflux, specistd=specistd,
dust_emission=False, synthspec=True)

# Next, attempt to solve for the velocity dispersion.
compute_vdisp, _ = can_compute_vdisp(redshift, specwave)

# If the initial coefficients are all zero, override estimating vdisp.
if np.all(coeff_nomvdisp == 0.):
compute_vdisp = False

if compute_vdisp:
t0 = time.time()

init_tauv_bounds = (0., 2.)

# find vdisp via a chi2 scan
chi2grid = np.zeros(len(CTools.vdisp_grid))
for iv, vdisp1 in enumerate(CTools.vdisp_grid):
# convolve the templates at the derived vdisp and fit
input_templateflux_nolines = templates.convolve_vdisp(templates.flux_nolines[agekeep, :], vdisp1)
tauv1, _, coeff1, resid1 = CTools.fit_stellar_continuum(
input_templateflux_nolines, fit_vdisp=False, conv_pre=None,
tauv_bounds=init_tauv_bounds, specflux=specflux, specistd=specistd,
dust_emission=False, synthspec=True)
chi2grid[iv] = resid1.dot(resid1)
print(vdisp1, tauv1)
if vdisp1 > 360.:
model1 = CTools.continuum_to_spectroscopy(CTools.optimizer_saved_contmodel)
I = specivar > 0.
import matplotlib.pyplot as plt
plt.clf()
plt.plot(specwave[I], specflux[I])
plt.plot(specwave[I], model1[I])
plt.savefig('junk.png')
import pdb ; pdb.set_trace()


try:
imin = find_minima(chi2grid)[0]
vdisp, vdisp_sigma, chi2min, warn, (a, b, c) = minfit(
CTools.vdisp_grid[imin-1:imin+2], chi2grid[imin-1:imin+2], return_coeff=True)
except:
errmsg = 'A problem was encountered minimizing chi2.'
log.warning(errmsg)
imin, vdisp, vdisp_sigma, chi2min, warn, (a, b, c) = \
0, templates.vdisp_nominal, 0., 0., 1, (0., 0., 0.)

if vdisp_sigma > TINY:
vdisp_ivar = 1. / vdisp_sigma**2
else:
vdisp_ivar = 0.

if debug_plots:
import matplotlib.pyplot as plt
import seaborn as sns
sns.set(context='talk', style='ticks', font_scale=0.8)

pngfile = f'qa-vdisp-{uniqueid}.png'

if vdisp_ivar > 0:
txt = r'$\sigma_{star}$='+f'{vdisp:.0f}'+r'$\pm$'+f'{vdisp_sigma:.0f} km/s'
else:
txt = r'$\sigma_{star}$='+f'{vdisp:.0f}'

fig, ax = plt.subplots(figsize=(8, 6))
ax.scatter(CTools.vdisp_grid, chi2grid-chi2min, marker='s', s=50, color='gray', edgecolor='k')
if not np.all(np.array([a, b, c]) == 0.):
yquad = np.polyval([a, b, c], CTools.vdisp_grid)
ax.plot(CTools.vdisp_grid[imin-3:imin+4], yquad[imin-3:imin+4]-chi2min, lw=2, ls='--')
#ax.set_ylim(0., np.min((np.max(yquad), np.max(chi2grid)))-chi2min)
ax.set_xlabel(r'$\sigma_{star}$ (km/s)')
ax.text(0.9, 0.9, txt, ha='right', va='center', transform=ax.transAxes)
ax.set_ylabel(r'$\Delta\chi^2$')
fig.tight_layout()
fig.savefig(pngfile)#, bbox_inches='tight')
plt.close()
log.info(f'Wrote {pngfile}')

import pdb ; pdb.set_trace()

# copy the model before it is overwritten...
contmodel_nomvdisp = CTools.optimizer_saved_contmodel.copy() # copy needed??

Expand All @@ -1194,6 +1246,10 @@ def continuum_fastspec(redshift, objflam, objflamivar, CTools,
specistd=specistd, dust_emission=False, synthspec=True)
contmodel_withvdisp = CTools.optimizer_saved_contmodel.copy() # copy needed??

## If the initial coefficients are all zero, override estimating vdisp.
#if np.all(coeff_nomvdisp == 0.):
# compute_vdisp = False

# ToDo: use a delta-chi2 test to determine if solving for the velocity
# dispersion improves the fit in a statistically significant way.
chi2_withvdisp = resid_withvdisp.dot(resid_withvdisp)
Expand Down Expand Up @@ -1228,12 +1284,9 @@ def continuum_fastspec(redshift, objflam, objflamivar, CTools,

for imonte in range(nmonte):
tauv1, vdisp1, coeff1, _ = CTools.fit_stellar_continuum(
templates.flux_nolines[agekeep, :], fit_vdisp=True,
conv_pre=input_conv_pre_nolines,
vdisp_guess=vdisp_guess[imonte],
tauv_guess=tauv_guess[imonte],
tauv_bounds=init_tauv_bounds,
specflux=specflux_monte[imonte, :],
templates.flux_nolines, fit_vdisp=True, conv_pre=input_conv_pre_nolines,
vdisp_guess=vdisp_guess[imonte], tauv_guess=tauv_guess[imonte],
tauv_bounds=init_tauv_bounds, specflux=specflux_monte[imonte, :],
specistd=specistd, dust_emission=False, synthspec=True)
coeff_monte[imonte, :] = coeff1
tauv_monte[imonte] = tauv1
Expand Down Expand Up @@ -1310,13 +1363,30 @@ def continuum_fastspec(redshift, objflam, objflamivar, CTools,
log.info(f'Wrote {pngfile}')
else:
log.info('Insufficient wavelength coverage to compute velocity dispersion.')

# Perform an initial fit to the data using the cached templates with
# nominal velocity dispersion. Reduce the prior bound on tauv to minimize
# the age-dust degeneracy, which is especially important when trying to
# estimate the velocity dispersion in the next call to
# fit_stellar_continuum. When we fit the full spectrum+photometry (at fixed
# velocity dispersion), we revert to the larger (default) tauv prior, since
# the IR fluxes put an additional constraint on the dust content.
init_tauv_bounds = (0., 1.)
tauv_nomvdisp, _, coeff_nomvdisp, resid_nomvdisp = CTools.fit_stellar_continuum(
templates.flux_nolines_nomvdisp[agekeep, :], fit_vdisp=False, conv_pre=None,
tauv_bounds=init_tauv_bounds, specflux=specflux, specistd=specistd,
dust_emission=False, synthspec=True)


print('OLD CODE HERE')
tauv = tauv_nomvdisp
coeff = coeff_nomvdisp
vdisp = templates.vdisp_nominal
input_templateflux = templates.flux_nomvdisp[agekeep, :]
input_templateflux_nolines = templates.flux_nolines_nomvdisp[agekeep, :]
contmodel = CTools.optimizer_saved_contmodel.copy()


# Next, estimate the aperture correction.
apercorrs = np.ones(len(phot.synth_bands))
median_apercorr = 1.
Expand Down

0 comments on commit c3b4047

Please sign in to comment.