def __init__(self, par=None):
if par is None:
# Set the default parameter set. The guess_redshift,
# stellar_continuum, and emission_lines values can be filled
# by EmissionLineModel._fill_method_par()
par = { 'guess_redshift': None, # The guess redshift for each binned spectrum
'stellar_continuum': None, # The StellarContinuumModel object
'emission_lines': None, # The EmissionLineDB object
'degree': -1, # Additive polynomial order
'mdegree': 10 } # Multiplicative polynomial order
EmissionLineFit.__init__(self, 'XJMC', None, par=par)
def fit(self, binned_spectra, par=None, loggers=None, quiet=False):
if par is not None:
self.par = par
# Check the parameter keys
required_keys = [ 'guess_redshift', 'stellar_continuum', 'emission_lines', 'degree',
'mdegree' ]
if numpy.any([ reqk not in self.par.keys() for reqk in required_keys ]):
raise ValueError('Parameter dictionary does not have all the required keys.')
# Wavelengths are in vacuum
wave0 = binned_spectra['WAVE'].data.copy()
# Velocity step per pixel
velscale = spectrum_velocity_scale(wave0)
# Get the best-fitting stellar kinematics for the binned spectra
# And correct sigmas with instrumental resolutions (if convolved templates are to be used)
if self.par['stellar_continuum'] is None \
or not isinstance(self.par['stellar_continuum'], StellarContinuumModel):
raise ValueError('Must provide StellarContinuumModel object as the '
'\'stellar_continuum\' item in the parameter dictionary')
stars_vel, stars_sig = self.par['stellar_continuum'].matched_guess_kinematics(
binned_spectra, cz=True, corrected=True, nearest=True)
# Convert the input stellar velocity from redshift (c*z) to ppxf velocity (c*log(1+z))
stars_vel = PPXFFit.revert_velocity(stars_vel,0)[0]
# Get the stellar templates and the template resolution;
# shape is (Nstartpl, Ntplwave)
stars_templates = self.par['stellar_continuum'].get_template_library(
velocity_offset=numpy.median(stars_vel), match_to_drp_resolution=True)
stars_templates_wave = stars_templates['WAVE'].data.copy()
template_sres = stars_templates['SPECRES'].data[0,:]
stars_templates = stars_templates['FLUX'].data.copy()
velscale_ratio = self.par['stellar_continuum'].method['fitpar']['velscale_ratio']
# Set mask for the galaxy spectra according to the templates wave range
mask = PPXFFit.fitting_mask(tpl_wave=stars_templates_wave, obj_wave=wave0,
velscale=velscale, velscale_ratio=velscale_ratio,
velocity_offset=numpy.median(stars_vel))[0]
wave = wave0[mask]
# Calculate the velocity offset between the masked spectra and the tempaltes
dv = -PPXFFit.ppxf_tpl_obj_voff(stars_templates_wave, wave, velscale,
velscale_ratio=velscale_ratio)
# UNBINNED DATA:
# Flux and noise masked arrays; shape is (Nspaxel,Nwave) where
# Nspaxel is Nx*Ny
# Bin ID from VOR10 reference file used to mask buffer spaxels
#binid0 = binned_spectra['BINID'].data
# Create mask for parsing the input data
#binid = binid0.reshape(-1)
#mask_spaxel = ~(binid==-1)
# pPXF would run into problems if dircetly masked arrays were used
# So both fluxes and their masks are to be provided as input
# flux00 = binned_spectra.drpf.copy_to_masked_array(flag=['DONOTUSE', 'FORESTAR'])
# mask_drp0 = ~flux00.mask
# flux = binned_spectra.drpf.copy_to_array(ext='FLUX')
# ivar = binned_spectra.drpf.copy_to_array(ext='IVAR')
# flux0, ivar0 = binned_spectra.galext.apply(flux, ivar=ivar, deredden=True)
# noise0 = numpy.power(ivar0.data, -0.5)
# mask_drp = mask_drp0.reshape(-1, mask_drp0.shape[-1])[:,mask]
# flux = flux0.data[:,mask]
# noise = noise0[:,mask]
flux0 = binned_spectra.drpf.copy_to_masked_array(flag=['DONOTUSE', 'FORESTAR'])
ivar0 = binned_spectra.drpf.copy_to_masked_array(ext='IVAR', flag=['DONOTUSE', 'FORESTAR'])
flux0, ivar0 = binned_spectra.galext.apply(flux0, ivar=ivar0, deredden=True)
noise = numpy.ma.power(ivar0, -0.5)
noise[numpy.invert(noise > 0)] = numpy.ma.masked
mask_drp = numpy.invert(flux0.mask | noise.mask)[:,mask]
flux = flux0.data[:,mask]
noise = noise.filled(0.0)[:,mask]
# stack_sres sets whether or not the spectral resolution is
# determined on a per-spaxel basis or with a single vector
sres = binned_spectra.drpf.spectral_resolution(toarray=True, fill=True) \
if binned_spectra.method['stackpar']['stack_sres'] else \
binned_spectra.drpf.spectral_resolution(ext='SPECRES', toarray=True, fill=True)
sres = sres[:,mask]
# Spaxel coordinates; shape is (Nspaxel,)
x = binned_spectra.rdxqa['SPECTRUM'].data['SKY_COO'][:,0]
y = binned_spectra.rdxqa['SPECTRUM'].data['SKY_COO'][:,1]
# BINNED DATA:
# Binned flux and binned noise masked arrays; shape is (Nbin,Nwave)
# flux_binned00 = binned_spectra.copy_to_masked_array(flag=binned_spectra.do_not_fit_flags())
# mask_binned0 = ~flux_binned00.mask
# flux_binned0 = binned_spectra.copy_to_array(ext='FLUX')
# noise_binned0 = numpy.power(binned_spectra.copy_to_array(ext='IVAR'), -0.5)
# mask_binned = mask_binned0[:,mask]
# flux_binned = flux_binned0[:,mask]
# noise_binned = noise_binned0[:,mask]
# sres_binned0 = binned_spectra.copy_to_array(ext='SPECRES')
# sres_binned = sres_binned0[:,mask]
flux_binned = binned_spectra.copy_to_masked_array(flag=binned_spectra.do_not_fit_flags())
noise_binned = numpy.ma.power(binned_spectra.copy_to_masked_array(ext='IVAR',
flag=binned_spectra.do_not_fit_flags()), -0.5)
noise_binned[numpy.invert(noise_binned > 0)] = numpy.ma.masked
mask_binned = numpy.invert(flux_binned.mask | noise_binned.mask)[:,mask]
flux_binned = flux_binned.data[:,mask]
noise_binned = noise_binned.filled(0.0)[:,mask]
sres_binned = binned_spectra.copy_to_array(ext='SPECRES')[:,mask]
# Bin coordinates; shape is (Nbin,)
x_binned = binned_spectra['BINS'].data['SKY_COO'][:,0]
y_binned = binned_spectra['BINS'].data['SKY_COO'][:,1]
# Set initial guesses for the velocity and velocity dispersion
if self.par['guess_redshift'] is not None:
# Use guess_redshift if provided
guess_vel0 = self.par['guess_redshift'] * astropy.constants.c.to('km/s').value
guess_vel = PPXFFit.revert_velocity(guess_vel0,0)[0]
# And set default velocity dispersion to 100 km/s
guess_sig = numpy.full(guess_vel.size, 100, dtype=float)
elif self.par['stellar_continuum'] is not None:
# Otherwise use the stellar-continuum result
guess_vel, guess_sig = stars_vel.copy(), stars_sig.copy()
else:
raise ValueError('Cannot set guess kinematics; must provide either \'guess_redshift\' '
'or \'stellar_continuum\' in input parameter dictionary.')
# Construct gas templates; shape is (Ngastpl, Ntplwave).
# Template resolution matched between the stellar and gas
# templates?
# Decide whether to use convolved gas templates
# Set the wavelength of lines to be in vacuum
FWHM = wave/numpy.max(sres, axis=0)
FWHM_binned = wave/sres_binned[0,:]
def fwhm_drp(wave_len0):
wave_len = wave_len0*(1 + numpy.median(self.par['guess_redshift']))
index = numpy.argmin(abs(wave-wave_len[:,None]),axis=1) \
if numpy.asarray(wave_len) is wave_len \
else numpy.argmin(abs(wave-wave_len))
return FWHM[index]
def fwhm_binned(wave_len0):
wave_len = wave_len0*(1 + numpy.median(self.par['guess_redshift']))
index = numpy.argmin(abs(wave-wave_len[:,None]),axis=1) \
if numpy.asarray(wave_len) is wave_len \
else numpy.argmin(abs(wave-wave_len))
return FWHM_binned[index]
lam_range_gal = numpy.array([numpy.min(wave), numpy.max(wave)]) \
/ (1 + numpy.median(self.par['guess_redshift']))
gas_templates, gas_names, gas_wave = \
ppxf_util.emission_lines(numpy.log(stars_templates_wave), lam_range_gal, fwhm_drp)
gas_templates_binned, gas_names, gas_wave = \
ppxf_util.emission_lines(numpy.log(stars_templates_wave), lam_range_gal, fwhm_binned)
# Default polynomial orders
degree = -1 if self.par['degree'] is None else self.par['degree']
mdegree = 10 if self.par['mdegree'] is None else self.par['mdegree']
# --------------------------------------------------------------
# CALL TO EMLINE_FITTER_WITH_PPXF:
# Input is:
# - wave: wavelength vector; shape is (Nwave,)
# - flux: observed, unbinned flux; masked array with shape
# (Nspaxel,Nwave)
# - noise: error in observed, unbinned flux; masked array with
# shape (Nspaxel,Nwave)
# - sres: spectral resolution (R=lambda/delta lambda) as a
# function of wavelength for each unbinned spectrum; shape
# is (Nspaxel,Nwave)
# - flux_binned: binned flux; masked array with shape
# (Nbin,Nwave)
# - noise_binned: noise in binned flux; masked array with
# shape (Nbin,Nwave)
# - sres_binned: spectral resolution (R=lambda/delta lambda)
# as a function of wavelength for each binned spectrum;
# shape is (Nbin,Nwave)
# - velscale: Velocity step per pixel
# - velscale_ratio: Ratio of velocity step per pixel in the
# observed data versus in the template data
# - dv: Velocity offset between the galaxy and template data
# due to the difference in the initial wavelength of the
# spectra
# - stars_vel: Velocity of the stellar component; shape is
# (Nbins,)
# - stars_sig: Velocity dispersion of the stellar component;
# shape is (Nbins,)
# - stars_templates: Stellar templates; shape is (Nstartpl,
# Ntplwave)
# - guess_vel: Initial guess velocity for the gas components;
# shape is (Nbins,)
# - guess_sig: Initial guess velocity dispersion for the gas
# components; shape is (Nbins,)
# - gas_templates: Gas template flux; shape is (Ngastpl,
# Ntplwave)
# - gas_names: Name of the gas templats; shape is (Ngastpl,)
# - template_sres: spectral resolution (R=lambda/delta
# lambda) as a function of wavelength for all the templates
# templates; shape is (Ntplwave,)
# - degree: Additive polynomial order
# - mdegree: Multiplicative polynomial order
# - x: On-sky spaxel x coordinates; shape is (Nspaxel,)
# - y: On-sky spaxel y coordinates; shape is (Nspaxel,)
# - x_binned: On-sky bin x coordinate; shape is (Nbin,)
# - y_binned: On-sky bin y coordinate; shape is (Nbin,)
model_flux0, model_eml_flux0, model_mask0, model_binid, eml_flux, eml_fluxerr, \
eml_kin, eml_kinerr, eml_sigmacorr \
= emline_fitter_with_ppxf(wave, flux, noise, sres, flux_binned,
noise_binned, velscale, velscale_ratio, dv,
stars_vel, stars_sig, stars_templates,
guess_vel, guess_sig, gas_templates,
gas_templates_binned, gas_names, template_sres,
degree, mdegree, x, y, x_binned, y_binned,
mask_binned, mask_drp,
numpy.median(self.par['guess_redshift']))
#, debug=True)
# Output is:
# - model_flux: stellar-continuum + emission-line model; shape
# is (Nmod, Nwave); first axis is ordered by model ID number
# - model_eml_flux: model emission-line flux only; shape is
# (Nmod, Nwave); first axis is ordered by model ID number
# - model_mask: boolean or bit mask for fitted models; shape
# is (Nmod, Nwave); first axis is ordered by model ID number
# - model_binid: ID numbers assigned to each spaxel with a
# fitted model; any spaxel without a model should have
# model_binid = -1; the number of >-1 IDs must be Nmod;
# shape is (Nx,Ny) which is equivalent to:
# flux[:,0].reshape((numpy.sqrt(Nspaxel).astype(int),)*2).shape
# - eml_flux: Flux of each emission line; shape is (Nmod,Neml)
# - eml_fluxerr: Error in emission-line fluxes; shape is
# (Nmod, Neml)
# - eml_kin: Kinematics (velocity and velocity dispersion) of
# each emission line; shape is (Nmod,Neml,Nkin)
# - eml_kinerr: Error in the kinematics of each emission line
# - eml_sigmacorr: Quadrature corrections required to obtain
# the astrophysical velocity dispersion; shape is
# (Nmod,Neml); corrections are expected to be applied as
# follows:
# sigma = numpy.ma.sqrt( numpy.square(eml_kin[:,:,1])
# - numpy.square(eml_sigmacorr))
# --------------------------------------------------------------
# Convert the output velocity back from ppxf velocity (c*log(1+z)) to redshift (c*z)
eml_kin[:,:,0] = PPXFFit.convert_velocity(eml_kin[:,:,0],0)[0]
# Mask output data according to model_binid
model_binid = numpy.asarray(model_binid, dtype=numpy.int16)
mask_id = (model_binid > -1)
model_flux0 = model_flux0[mask_id]
model_eml_flux0 = model_eml_flux0[mask_id]
model_mask0 = model_mask0[mask_id]
eml_flux = eml_flux[mask_id]
eml_fluxerr = eml_fluxerr[mask_id]
eml_kin = eml_kin[mask_id]
eml_kinerr = eml_kinerr[mask_id]
eml_sigmacorr = eml_sigmacorr[mask_id]
#cube_binid = numpy.full_like(mask_spaxel, -1, dtype=numpy.int16)
#cube_binid[mask_spaxel] = model_binid
Nspaxel = x.shape[0]
model_binid = model_binid.reshape((numpy.sqrt(Nspaxel).astype(int),)*2)
# The ordered indices in the flatted bin ID map with/for each model
model_srt = numpy.argsort(model_binid.ravel())[model_binid.ravel() > -1]
# Construct the output emission-line database. The data type
# defined by EmissionLineFit._per_emission_line_dtype(); shape
# is (Nmod,); parameters must be ordered by model ID number
nmod = len(model_srt)
neml = eml_flux.shape[1]
nkin = eml_kin.shape[-1]
model_eml_par = init_record_array(nmod,
EmissionLineFit._per_emission_line_dtype(neml, nkin, numpy.int16))
model_eml_par['BINID'] = model_binid.ravel()[model_srt]
model_eml_par['BINID_INDEX'] = numpy.arange(nmod)
model_eml_par['MASK'][:,:] = 0
model_eml_par['FLUX'] = eml_flux
model_eml_par['FLUXERR'] = eml_fluxerr
model_eml_par['KIN'] = eml_kin
model_eml_par['KINERR'] = eml_kinerr
model_eml_par['SIGMACORR'] = eml_sigmacorr
# Include the equivalent width measurements
if self.par['emission_lines'] is not None:
EmissionLineFit.measure_equivalent_width(wave, flux[model_srt,:],
par['emission_lines'], model_eml_par)
# Change back the wavelength range of models to match that of the galaxy's
model_flux = numpy.zeros(flux0[mask_id].shape)
model_eml_flux = numpy.zeros(flux0[mask_id].shape)
model_mask = numpy.full_like(flux0[mask_id], 0, dtype=bool)
model_flux[:,mask] = model_flux0
model_eml_flux[:,mask] = model_eml_flux0
model_mask[:,mask] = model_mask0
# Calculate the "emission-line baseline" as the difference
# between the stellar continuum model determined for the
# kinematics and the one determined by the optimized
# stellar-continuum + emission-line fit:
if self.par['stellar_continuum'] is not None:
# Construct the full 3D cube for the stellar continuum
# models
sc_model_flux, sc_model_mask \
= DAPFitsUtil.reconstruct_cube(binned_spectra.drpf.shape,
self.par['stellar_continuum']['BINID'].data.ravel(),
[ self.par['stellar_continuum']['FLUX'].data,
self.par['stellar_continuum']['MASK'].data ])
# Set any masked pixels to 0
sc_model_flux[sc_model_mask>0] = 0.0
# Construct the full 3D cube of the new stellar continuum
# from the combined stellar-continuum + emission-line fit
el_continuum = DAPFitsUtil.reconstruct_cube(binned_spectra.drpf.shape,
model_binid.ravel(),
model_flux - model_eml_flux)
# Get the difference, restructure it to match the shape
# of the emission-line models, and zero any masked pixels
model_eml_base = (el_continuum - sc_model_flux).reshape(-1,wave0.size)[model_srt,:]
if model_mask is not None:
model_eml_base[model_mask==0] = 0.0
else:
model_eml_base = numpy.zeros(model_flux.shape, dtype=float)
# Returned arrays are:
# - model_eml_flux: model emission-line flux only; shape is
# (Nmod, Nwave); first axis is ordered by model ID number
# - model_eml_base: difference between the combined fit and the
# stars-only fit; shape is (Nmod, Nwave); first axis is
# ordered by model ID number
# - model_mask: boolean or bit mask for fitted models; shape
# is (Nmod, Nwave); first axis is ordered by model ID number
# - model_fit_par: This provides the results of each fit;
# TODO: The is set to None. Provide metrics of the ppxf fit
# to each spectrum?
# - model_eml_par: output model parameters; data type must be
# EmissionLineFit._per_emission_line_dtype(); shape is
# (Nmod,); parameters must be ordered by model ID number
# - model_binid: ID numbers assigned to each spaxel with a
# fitted model; any spaxel with a model should have
# model_binid = -1; the number of >-1 IDs must be Nmod;
# shape is (Nx,Ny)
return model_eml_flux, model_eml_base, model_mask, None, model_eml_par, model_binid
def main():
t = time.perf_counter()
arg = parse_args()
if not os.path.isfile(arg.inp):
raise FileNotFoundError('No file: {0}'.format(arg.inp))
directory_path = os.getcwd(
) if arg.output_root is None else os.path.abspath(arg.output_root)
if not os.path.isdir(directory_path):
os.makedirs(directory_path)
data_file = os.path.abspath(arg.inp)
fit_file = os.path.join(directory_path, arg.out)
flag_db = None if arg.spec_flags is None else os.path.abspath(
arg.spec_flags)
# Read the data
spectral_step = 1e-4
wave, flux, ferr, sres, redshift, fit_spectrum = object_data(
data_file, flag_db)
nspec, npix = flux.shape
dispersion = numpy.full(nspec, 100., dtype=numpy.float)
# fit_spectrum[:] = False
# fit_spectrum[0] = True
# fit_spectrum[171] = True
# fit_spectrum[791] = True
# Mask spectra that should not be fit
indx = numpy.any(numpy.logical_not(numpy.ma.getmaskarray(flux)),
axis=1) & fit_spectrum
flux[numpy.logical_not(indx), :] = numpy.ma.masked
print('Read: {0}'.format(arg.inp))
print('Contains {0} spectra'.format(nspec))
print(' each with {0} pixels'.format(npix))
print('Fitting {0} spectra.'.format(numpy.sum(fit_spectrum)))
#-------------------------------------------------------------------
#-------------------------------------------------------------------
# Fit the stellar continuum
# Construct the template library
sc_tpl = TemplateLibrary(arg.sc_tpl,
match_resolution=False,
velscale_ratio=arg.sc_vsr,
spectral_step=spectral_step,
log=True,
hardcopy=False)
# Set the spectral resolution
sc_tpl_sres = numpy.mean(sc_tpl['SPECRES'].data, axis=0).ravel()
# Set the pixel mask
sc_pixel_mask = SpectralPixelMask(artdb=ArtifactDB.from_key('BADSKY'),
emldb=EmissionLineDB.from_key('ELPMPL8'))
# Instantiate the fitting class
ppxf = PPXFFit(StellarContinuumModelBitMask())
# The following call performs the fit to the spectrum. Specifically
# note that the code only fits the first two moments, uses an
# 8th-order additive polynomial, and uses the 'no_global_wrej'
# iteration mode. See
# https://sdss-mangadap.readthedocs.io/en/latest/api/mangadap.proc.ppxffit.html#mangadap.proc.ppxffit.PPXFFit.fit
cont_wave, cont_flux, cont_mask, cont_par \
= ppxf.fit(sc_tpl['WAVE'].data.copy(), sc_tpl['FLUX'].data.copy(), wave, flux, ferr,
redshift, dispersion, iteration_mode='no_global_wrej', reject_boxcar=100,
ensemble=False, velscale_ratio=arg.sc_vsr, mask=sc_pixel_mask,
matched_resolution=False, tpl_sres=sc_tpl_sres, obj_sres=sres,
degree=arg.sc_deg, moments=2) #, plot=True)
if arg.sc_only:
write(fit_file, wave, cont_flux, cont_mask, cont_par)
print('Elapsed time: {0} seconds'.format(time.perf_counter() - t))
return
# if numpy.any(cont_par['KIN'][:,1] < 0):
# embed()
# exit()
#-------------------------------------------------------------------
#-------------------------------------------------------------------
#-------------------------------------------------------------------
# Measure the emission-line moments
# # Remask the continuum fit
# sc_continuum = StellarContinuumModel.reset_continuum_mask_window(
# numpy.ma.MaskedArray(cont_flux, mask=cont_mask>0))
# # Read the database that define the emission lines and passbands
# momdb = EmissionMomentsDB.from_key(arg.el_band)
# # Measure the moments
# elmom = EmissionLineMoments.measure_moments(momdb, wave, flux, continuum=sc_continuum,
# redshift=redshift)
#-------------------------------------------------------------------
#-------------------------------------------------------------------
# Fit the emission-line model
# Set the emission-line continuum templates if different from those
# used for the stellar continuum
if arg.sc_tpl == arg.el_tpl:
# If the keywords are the same, just copy over the previous
# library and the best fitting stellar kinematics
el_tpl = sc_tpl
el_tpl_sres = sc_tpl_sres
stellar_kinematics = cont_par['KIN'].copy()
else:
# If the template sets are different, we need to match the
# spectral resolution to the galaxy data and use the corrected
# velocity dispersions.
_sres = SpectralResolution(wave, sres[0, :], log10=True)
el_tpl = TemplateLibrary(arg.el_tpl,
sres=_sres,
velscale_ratio=arg.el_vsr,
spectral_step=spectral_step,
log=True,
hardcopy=False)
el_tpl_sres = numpy.mean(el_tpl['SPECRES'].data, axis=0).ravel()
stellar_kinematics = cont_par['KIN'].copy()
stellar_kinematics[:, 1] = numpy.ma.sqrt(
numpy.square(cont_par['KIN'][:, 1]) -
numpy.square(cont_par['SIGMACORR_SRES'])).filled(0.0)
# if numpy.any(cont_par['KIN'][:,1] < 0):
# embed()
# exit()
#
# if numpy.any(stellar_kinematics[:,1] < 0):
# embed()
# exit()
# Mask the 5577 sky line
el_pixel_mask = SpectralPixelMask(artdb=ArtifactDB.from_key('BADSKY'))
# Read the emission line fitting database
emldb = EmissionLineDB.from_key(arg.el_list)
# Instantiate the fitting class
emlfit = Sasuke(EmissionLineModelBitMask())
# TODO: Improve the initial velocity guess using the first moment...
# Perform the fit
elfit_time = time.perf_counter()
model_wave, model_flux, eml_flux, model_mask, eml_fit_par, eml_eml_par \
= emlfit.fit(emldb, wave, flux, obj_ferr=ferr, obj_mask=el_pixel_mask, obj_sres=sres,
guess_redshift=redshift, guess_dispersion=dispersion, reject_boxcar=101,
stpl_wave=el_tpl['WAVE'].data, stpl_flux=el_tpl['FLUX'].data,
stpl_sres=el_tpl_sres, stellar_kinematics=stellar_kinematics,
etpl_sinst_mode='offset', etpl_sinst_min=10., velscale_ratio=arg.el_vsr,
matched_resolution=False, mdegree=arg.el_deg, ensemble=False)#, plot=True)
print('EML FIT TIME: ', time.perf_counter() - elfit_time)
# Line-fit metrics (should this be done in the fit method?)
eml_eml_par = EmissionLineFit.line_metrics(emldb,
wave,
flux,
ferr,
model_flux,
eml_eml_par,
model_mask=model_mask,
bitmask=emlfit.bitmask)
# Equivalent widths
EmissionLineFit.measure_equivalent_width(wave,
flux,
emldb,
eml_eml_par,
bitmask=emlfit.bitmask,
checkdb=False)
# Measure the emission-line moments
# - Model continuum
continuum = StellarContinuumModel.reset_continuum_mask_window(model_flux -
eml_flux)
# - Updated redshifts
fit_redshift = eml_eml_par['KIN'][:,numpy.where(emldb['name'] == 'Ha')[0][0],0] \
/ astropy.constants.c.to('km/s').value
# - Set the moment database
momdb = EmissionMomentsDB.from_key(arg.el_band)
# - Set the moment bitmask
mombm = EmissionLineMomentsBitMask()
# - Measure the moments
elmom = EmissionLineMoments.measure_moments(momdb,
wave,
flux,
ivar=numpy.ma.power(ferr, -2),
continuum=continuum,
redshift=fit_redshift,
bitmask=mombm)
# - Select the bands that are valid
include_band = numpy.array([numpy.logical_not(momdb.dummy)]*nspec) \
& numpy.logical_not(mombm.flagged(elmom['MASK'],
flag=['BLUE_EMPTY', 'RED_EMPTY']))
# - Set the line center at the center of the primary passband
line_center = (1.0 + fit_redshift)[:, None] * momdb['restwave'][None, :]
elmom['BMED'], elmom['RMED'], pos, elmom['EWCONT'], elmom['EW'], elmom['EWERR'] \
= emission_line_equivalent_width(wave, flux, momdb['blueside'], momdb['redside'],
line_center, elmom['FLUX'], redshift=fit_redshift,
line_flux_err=elmom['FLUXERR'],
include_band=include_band)
# - Flag non-positive measurements
indx = include_band & numpy.logical_not(pos)
elmom['MASK'][indx] = mombm.turn_on(elmom['MASK'][indx],
'NON_POSITIVE_CONTINUUM')
# - Set the binids
elmom['BINID'] = numpy.arange(nspec)
elmom['BINID_INDEX'] = numpy.arange(nspec)
write(fit_file,
wave,
cont_flux,
cont_mask,
cont_par,
model_flux=model_flux,
model_mask=model_mask,
eml_flux=eml_flux,
eml_fit_par=eml_fit_par,
eml_eml_par=eml_eml_par,
elmom=elmom)
print('Elapsed time: {0} seconds'.format(time.perf_counter() - t))
# kinematics from the stacked spectrum
eml_wave, model_flux, eml_flux, eml_mask, eml_fit_par, eml_eml_par \
= emlfit.fit(emldb, wave_binned, flux_binned, obj_ferr=ferr_binned,
obj_mask=el_pixel_mask, obj_sres=sres_binned,
guess_redshift=z_binned, guess_dispersion=dispersion_binned,
reject_boxcar=101, stpl_wave=el_tpl['WAVE'].data,
stpl_flux=el_tpl['FLUX'].data, stpl_sres=el_tpl_sres,
stellar_kinematics=stellar_kinematics, etpl_sinst_mode='offset',
etpl_sinst_min=10., velscale_ratio=velscale_ratio,
matched_resolution=False, mdegree=8, plot=fit_plots,
remapid=binid, remap_flux=flux, remap_ferr=ferr,
remap_mask=el_pixel_mask, remap_sres=sres, remap_skyx=x, remap_skyy=y,
obj_skyx=x_binned, obj_skyy=y_binned)
# Line-fit metrics
eml_eml_par = EmissionLineFit.line_metrics(emldb, wave, flux, ferr, model_flux, eml_eml_par,
model_mask=eml_mask, bitmask=emlfit.bitmask)
# Get the stellar continuum that was fit for the emission lines
elcmask = eml_mask > 0
edges = numpy.ma.notmasked_edges(numpy.ma.MaskedArray(model_flux, mask=elcmask), axis=1)
for i,s,e in zip(edges[0][0],edges[0][1],edges[1][1]):
elcmask[i,s:e+1] = False
el_continuum = numpy.ma.MaskedArray(model_flux - eml_flux, mask=elcmask)
# Plot the result
if usr_plots:
for i in range(flux.shape[0]):
pyplot.plot(wave, flux[i,:], label='Data')
pyplot.plot(wave, model_flux[i,:], label='Model')
pyplot.plot(wave, el_continuum[i,:], label='EL Cont.')
pyplot.plot(wave, sc_continuum[binid[i],:], label='SC Cont.')
def main():
t = time.perf_counter()
#-------------------------------------------------------------------
# Read spectra to fit. The following reads a single MaNGA spectrum.
# This is where you should read in your own spectrum to fit.
# Plate-IFU to use
plt = 7815
ifu = 3702
# Spaxel coordinates
x = 25 #30
y = 25 #37
# Where to find the relevant datacube. This example accesses the test data
# that can be downloaded by executing the script here:
# https://github.com/sdss/mangadap/blob/master/download_test_data.py
directory_path = defaults.dap_source_dir() / 'data' / 'remote'
# Read a spectrum
wave, flux, ivar, sres = get_spectra(plt, ifu, x, y, directory_path=directory_path)
# In general, the DAP fitting functions expect data to be in 2D
# arrays with shape (N-spectra,N-wave). So if you only have one
# spectrum, you need to expand the dimensions:
flux = flux.reshape(1,-1)
ivar = ivar.reshape(1,-1)
ferr = numpy.ma.power(ivar, -0.5)
sres = sres.reshape(1,-1)
# The majority (if not all) of the DAP methods expect that your
# spectra are binned logarithmically in wavelength (primarily
# because this is what pPXF expects). You can either have the DAP
# function determine this value (commented line below) or set it
# directly. The value is used to resample the template spectra to
# match the sampling of the spectra to fit (up to some integer; see
# velscale_ratio).
# spectral_step = spectral_coordinate_step(wave, log=True)
spectral_step = 1e-4
# Hereafter, the methods expect a wavelength vector, a flux array
# with the spectra to fit, an ferr array with the 1-sigma errors in
# the flux, and sres with the wavelength-dependent spectral
# resolution, R = lambda / Dlambda
#-------------------------------------------------------------------
#-------------------------------------------------------------------
# The DAP needs a reasonable guess of the redshift of the spectrum
# (within +/- 2000 km/s). In this example, I'm pulling the redshift
# from the DRPall file. There must be one redshift estimate per
# spectrum to fit. Here that means it's a single element array
# This example accesses the test data
# that can be downloaded by executing the script here:
# https://github.com/sdss/mangadap/blob/master/download_test_data.py
drpall_file = directory_path / f'drpall-{drp_test_version}.fits'
z = numpy.array([get_redshift(plt, ifu, drpall_file)])
print('Redshift: {0}'.format(z[0]))
# The DAP also requires an initial guess for the velocity
# dispersion. A guess of 100 km/s is usually robust, but this may
# depend on your spectral resolution.
dispersion = numpy.array([100.])
#-------------------------------------------------------------------
#-------------------------------------------------------------------
# The following sets the keyword for the template spectra to use
# during the fit. You can specify different template sets to use
# during the stellar-continuum (stellar kinematics) fit and the
# emission-line modeling.
# Templates used in the stellar continuum fits
sc_tpl_key = 'MILESHC'
# Templates used in the emission-line modeling
el_tpl_key = 'MASTARSSP'
# You also need to specify the sampling for the template spectra.
# The templates must be sampled with the same pixel step as the
# spectra to be fit, up to an integer factor. The critical thing
# for the sampling is that you do not want to undersample the
# spectral resolution element of the template spectra. Here, I set
# the sampling for the MILES templates to be a factor of 4 smaller
# than the MaNGA spectrum to be fit (which is a bit of overkill
# given the resolution difference). I set the sampling of the
# MaStar templates to be the same as the galaxy data.
# Template pixel scale a factor of 4 smaller than galaxy data
sc_velscale_ratio = 4
# Template sampling is the same as the galaxy data
el_velscale_ratio = 1
# You then need to identify the database that defines the
# emission-line passbands (elmom_key) for the non-parametric
# emission-line moment calculations, and the emission-line
# parameters (elfit_key) for the Gaussian emission-line modeling.
# See
# https://sdss-mangadap.readthedocs.io/en/latest/emissionlines.html.
elmom_key = 'ELBMPL9'
elfit_key = 'ELPMPL11'
# If you want to also calculate the spectral indices, you can
# provide a keyword that indicates the database with the passband
# definitions for both the absorption-line and bandhead/color
# indices to measure. The script allows these to be None, if you
# don't want to calculate the spectral indices. See
# https://sdss-mangadap.readthedocs.io/en/latest/spectralindices.html
absindx_key = 'EXTINDX'
bhdindx_key = 'BHBASIC'
# Now we want to construct a pixel mask that excludes regions with
# known artifacts and emission lines. The 'BADSKY' artifact
# database only masks the 5577, which can have strong left-over
# residuals after sky-subtraction. The list of emission lines (set
# by the ELPMPL8 keyword) can be different from the list of
# emission lines fit below.
sc_pixel_mask = SpectralPixelMask(artdb=ArtifactDB.from_key('BADSKY'),
emldb=EmissionLineDB.from_key('ELPMPL11'))
# Mask the 5577 sky line
el_pixel_mask = SpectralPixelMask(artdb=ArtifactDB.from_key('BADSKY'))
# Finally, you can set whether or not to show a set of plots.
#
# Show the ppxf-generated plots for each fit stage.
fit_plots = False
# Show summary plots
usr_plots = True
#-------------------------------------------------------------------
#-------------------------------------------------------------------
# Fit the stellar continuum
# First, we construct the template library. The keyword that
# selects the template library (sc_tpl_key) is defined above. The
# following call reads in the template library and processes the
# data to have the appropriate pixel sampling. Note that *no*
# matching of the spectral resolution to the galaxy spectrum is
# performed.
sc_tpl = TemplateLibrary(sc_tpl_key, match_resolution=False, velscale_ratio=sc_velscale_ratio,
spectral_step=spectral_step, log=True, hardcopy=False)
# This calculation of the mean spectral resolution is a kludge. The
# template library should provide spectra that are *all* at the
# same spectral resolution. Otherwise, one cannot freely combine
# the spectra to fit the Doppler broadening of the galaxy spectrum
# in a robust (constrained) way (without substantially more
# effort). There should be no difference between what's done below
# and simply taking the spectral resolution to be that of the first
# template spectrum (i.e., sc_tpl['SPECRES'].data[0])
sc_tpl_sres = numpy.mean(sc_tpl['SPECRES'].data, axis=0).ravel()
# Instantiate the fitting class, including the mask that it should
# use to flag the data. [[This mask should just be default...]]
ppxf = PPXFFit(StellarContinuumModelBitMask())
# The following call performs the fit to the spectrum. Specifically
# note that the code only fits the first two moments, uses an
# 8th-order additive polynomial, and uses the 'no_global_wrej'
# iteration mode. See
# https://sdss-mangadap.readthedocs.io/en/latest/api/mangadap.proc.ppxffit.html#mangadap.proc.ppxffit.PPXFFit.fit
cont_wave, cont_flux, cont_mask, cont_par \
= ppxf.fit(sc_tpl['WAVE'].data.copy(), sc_tpl['FLUX'].data.copy(), wave, flux, ferr,
z, dispersion, iteration_mode='no_global_wrej', reject_boxcar=100,
ensemble=False, velscale_ratio=sc_velscale_ratio, mask=sc_pixel_mask,
matched_resolution=False, tpl_sres=sc_tpl_sres, obj_sres=sres, degree=8,
moments=2, plot=fit_plots)
# The returned objects from the fit are the wavelength, model, and
# mask vectors and the record array with the best-fitting model
# parameters. The datamodel of the best-fitting model parameters is
# set by:
# https://sdss-mangadap.readthedocs.io/en/latest/api/mangadap.proc.spectralfitting.html#mangadap.proc.spectralfitting.StellarKinematicsFit._per_stellar_kinematics_dtype
# Remask the continuum fit
sc_continuum = StellarContinuumModel.reset_continuum_mask_window(
numpy.ma.MaskedArray(cont_flux, mask=cont_mask>0))
# Show the fit and residual
if usr_plots:
pyplot.plot(wave, flux[0,:], label='Data')
pyplot.plot(wave, sc_continuum[0,:], label='Model')
pyplot.plot(wave, flux[0,:] - sc_continuum[0,:], label='Resid')
pyplot.legend()
pyplot.xlabel('Wavelength')
pyplot.ylabel('Flux')
pyplot.show()
#-------------------------------------------------------------------
#-------------------------------------------------------------------
# Get the emission-line moments using the fitted stellar continuum
# Read the database that define the emission lines and passbands
momdb = EmissionMomentsDB.from_key(elmom_key)
# Measure the moments
elmom = EmissionLineMoments.measure_moments(momdb, wave, flux, continuum=sc_continuum,
redshift=z)
#-------------------------------------------------------------------
#-------------------------------------------------------------------
# Fit the emission-line model
# Set the emission-line continuum templates if different from those
# used for the stellar continuum
if sc_tpl_key == el_tpl_key:
# If the keywords are the same, just copy over the previous
# library ...
el_tpl = sc_tpl
el_tpl_sres = sc_tpl_sres
# ... and the best fitting stellar kinematics
stellar_kinematics = cont_par['KIN']
else:
# If the template sets are different, we need to match the
# spectral resolution to the galaxy data ...
_sres = SpectralResolution(wave, sres[0,:], log10=True)
el_tpl = TemplateLibrary(el_tpl_key, sres=_sres, velscale_ratio=el_velscale_ratio,
spectral_step=spectral_step, log=True, hardcopy=False)
el_tpl_sres = numpy.mean(el_tpl['SPECRES'].data, axis=0).ravel()
# ... and use the corrected velocity dispersions.
stellar_kinematics = cont_par['KIN']
stellar_kinematics[:,1] = numpy.ma.sqrt(numpy.square(cont_par['KIN'][:,1]) -
numpy.square(cont_par['SIGMACORR_EMP']))
# Read the emission line fitting database
emldb = EmissionLineDB.from_key(elfit_key)
# Instantiate the fitting class
emlfit = Sasuke(EmissionLineModelBitMask())
# Perform the fit
efit_t = time.perf_counter()
eml_wave, model_flux, eml_flux, eml_mask, eml_fit_par, eml_eml_par \
= emlfit.fit(emldb, wave, flux, obj_ferr=ferr, obj_mask=el_pixel_mask, obj_sres=sres,
guess_redshift=z, guess_dispersion=dispersion, reject_boxcar=101,
stpl_wave=el_tpl['WAVE'].data, stpl_flux=el_tpl['FLUX'].data,
stpl_sres=el_tpl_sres, stellar_kinematics=stellar_kinematics,
etpl_sinst_mode='offset', etpl_sinst_min=10.,
velscale_ratio=el_velscale_ratio, matched_resolution=False, mdegree=8,
plot=fit_plots)
print('TIME: ', time.perf_counter() - efit_t)
# Line-fit metrics
eml_eml_par = EmissionLineFit.line_metrics(emldb, wave, flux, ferr, model_flux, eml_eml_par,
model_mask=eml_mask, bitmask=emlfit.bitmask)
# Get the stellar continuum that was fit for the emission lines
elcmask = eml_mask.ravel() > 0
goodpix = numpy.arange(elcmask.size)[numpy.invert(elcmask)]
start, end = goodpix[0], goodpix[-1]+1
elcmask[start:end] = False
el_continuum = numpy.ma.MaskedArray(model_flux - eml_flux,
mask=elcmask.reshape(model_flux.shape))
# Plot the result
if usr_plots:
pyplot.plot(wave, flux[0,:], label='Data')
pyplot.plot(wave, model_flux[0,:], label='Model')
pyplot.plot(wave, el_continuum[0,:], label='EL Cont.')
pyplot.plot(wave, sc_continuum[0,:], label='SC Cont.')
pyplot.legend()
pyplot.xlabel('Wavelength')
pyplot.ylabel('Flux')
pyplot.show()
# Remeasure the emission-line moments with the new continuum
new_elmom = EmissionLineMoments.measure_moments(momdb, wave, flux, continuum=el_continuum,
redshift=z)
# Compare the summed flux and Gaussian-fitted flux for all the
# fitted lines
if usr_plots:
pyplot.scatter(emldb['restwave'], (new_elmom['FLUX']-eml_eml_par['FLUX']).ravel(),
c=eml_eml_par['FLUX'].ravel(), cmap='viridis', marker='.', s=60, lw=0,
zorder=4)
pyplot.grid()
pyplot.xlabel('Wavelength')
pyplot.ylabel('Summed-Gaussian Difference')
pyplot.show()
#-------------------------------------------------------------------
#-------------------------------------------------------------------
# Measure the spectral indices
if absindx_key is None or bhdindx_key is None:
# Neither are defined, so we're done
print('Elapsed time: {0} seconds'.format(time.perf_counter() - t))
return
# Setup the databases that define the indices to measure
absdb = None if absindx_key is None else AbsorptionIndexDB.from_key(absindx_key)
bhddb = None if bhdindx_key is None else BandheadIndexDB.from_key(bhdindx_key)
# Remove the modeled emission lines from the spectra
flux_noeml = flux - eml_flux
redshift = stellar_kinematics[:,0] / astropy.constants.c.to('km/s').value
sp_indices = SpectralIndices.measure_indices(absdb, bhddb, wave, flux_noeml, ivar=ivar,
redshift=redshift)
# Calculate the velocity dispersion corrections
# - Construct versions of the best-fitting model spectra with and without
# the included dispersion
continuum = Sasuke.construct_continuum_models(emldb, el_tpl['WAVE'].data, el_tpl['FLUX'].data,
wave, flux.shape, eml_fit_par)
continuum_dcnvlv = Sasuke.construct_continuum_models(emldb, el_tpl['WAVE'].data,
el_tpl['FLUX'].data, wave, flux.shape,
eml_fit_par, redshift_only=True)
# - Get the dispersion corrections and fill the relevant columns of the
# index table
sp_indices['BCONT_MOD'], sp_indices['BCONT_CORR'], sp_indices['RCONT_MOD'], \
sp_indices['RCONT_CORR'], sp_indices['MCONT_MOD'], sp_indices['MCONT_CORR'], \
sp_indices['AWGT_MOD'], sp_indices['AWGT_CORR'], \
sp_indices['INDX_MOD'], sp_indices['INDX_CORR'], \
sp_indices['INDX_BF_MOD'], sp_indices['INDX_BF_CORR'], \
good_les, good_ang, good_mag, is_abs \
= SpectralIndices.calculate_dispersion_corrections(absdb, bhddb, wave, flux,
continuum, continuum_dcnvlv,
redshift=redshift,
redshift_dcnvlv=redshift)
# Apply the index corrections. This is only done here for the
# Worthey/Trager definition of the indices, as an example
corrected_indices = numpy.zeros(sp_indices['INDX'].shape, dtype=float)
corrected_indices_err = numpy.zeros(sp_indices['INDX'].shape, dtype=float)
# Unitless indices
corrected_indices[good_les], corrected_indices_err[good_les] \
= SpectralIndices.apply_dispersion_corrections(sp_indices['INDX'][good_les],
sp_indices['INDX_CORR'][good_les],
err=sp_indices['INDX_ERR'][good_les])
# Indices in angstroms
corrected_indices[good_ang], corrected_indices_err[good_ang] \
= SpectralIndices.apply_dispersion_corrections(sp_indices['INDX'][good_ang],
sp_indices['INDX_CORR'][good_ang],
err=sp_indices['INDX_ERR'][good_ang],
unit='ang')
# Indices in magnitudes
corrected_indices[good_mag], corrected_indices_err[good_mag] \
= SpectralIndices.apply_dispersion_corrections(sp_indices['INDX'][good_mag],
sp_indices['INDX_CORR'][good_mag],
err=sp_indices['INDX_ERR'][good_mag],
unit='mag')
# Print the results for a few indices
index_names = numpy.append(absdb['name'], bhddb['name'])
print('-'*73)
print(f'{"NAME":<8} {"Raw Index":>12} {"err":>12} {"Index Corr":>12} {"Index":>12} {"err":>12}')
print(f'{"-"*8:<8} {"-"*12:<12} {"-"*12:<12} {"-"*12:<12} {"-"*12:<12} {"-"*12:<12}')
for name in ['Hb', 'HDeltaA', 'Mgb', 'Dn4000']:
i = numpy.where(index_names == name)[0][0]
print(f'{name:<8} {sp_indices["INDX"][0,i]:12.4f} {sp_indices["INDX_ERR"][0,i]:12.4f} '
f'{sp_indices["INDX_CORR"][0,i]:12.4f} {corrected_indices[0,i]:12.4f} '
f'{corrected_indices_err[0,i]:12.4f}')
print('-'*73)
embed()
print('Elapsed time: {0} seconds'.format(time.perf_counter() - t))