def pixels_cannon(*args,**kwargs):
"""
NAME:
pixels_cannon
PURPOSE:
determine continuum pixels using a Cannon-like technique (Ness et al. 2015)
INPUT:
Either:
a) Input for running the apogee.spec.cannon:
spec - spectra to fit (nspec,nlambda)
specerrs - errors on the spectra (nspec,nlambda); assume no covariances
label1, label2, ... - labels (nspec); best to subtract reference values before running this
type= ('lin') type of Cannon to run:
'lin' - linear Cannon
'quad' - quadratic Cannon
b) Output from a previous Cannon run:
coefficients - coefficients from the fit (ncoeffs,nlambda)
scatter - scatter from the fit (nlambda)
KEYWORDS:
baseline_dev= (0.015) maximum deviation from baseline
label1_max= (10.**-5.) maximum deviation in first linear coefficient
label2_max= (0.006) similar for the second
label3_max= (0.012) similar for the third
labelN_max= same with default 0.03
...
scatter_max= (0.015) maximum scatter of residuals
OUTPUT:
Boolean index into the wavelength range with True for continuum pixels
HISTORY:
2015-02-05 - Written - Bovy (IAS@KITP)
"""
# Grab kwargs
type= kwargs.pop('type','lin')
# Parse input
if len(args) == 0: # Use default fit
from apogee.spec._train_cannon import load_fit
coeffs, scatter, baseline_labels= load_fit()
type= 'quad'
else:
spec= args[0]
specerr= args[1]
# Determine the type of input
if len(specerr.shape) == 2:
# Run the Cannon
if type.lower() == 'lin':
coeffs, scatter= cannon.linfit(*args)
elif type.lower() == 'quad':
coeffs, scatter= cannon.quadfit(*args)
else:
coeffs= spec
scatter= specerr
ncoeffs= coeffs.shape[0]
if type.lower() == 'lin':
nlabels= ncoeffs-1
elif type.lower() == 'quad':
nlabels= int((-3+numpy.sqrt(9+8*(ncoeffs-1))))//2
# Determine continuum pixels
out= numpy.ones(len(scatter),dtype='bool')
# Deviation from baseline
out[numpy.fabs(coeffs[0]-1.) > kwargs.get('baseline_dev',0.015)]= False
# Large dependence on labels
maxs= numpy.zeros(nlabels)
maxs[0]= kwargs.get('label1_max',10.**-5.)
maxs[1]= kwargs.get('label2_max',0.006)
maxs[2]= kwargs.get('label3_max',0.012)
for ii in range(nlabels-3):
maxs[ii+3]= kwargs.get('label%i_max' % (ii+4),0.03)
for ii in range(1,nlabels+1):
out[numpy.fabs(coeffs[ii]) > maxs[ii-1]]= False
# Large residuals
out[scatter > kwargs.get('scatter_max',0.015)]= False
return out
def pixels_cannon(*args, **kwargs):
"""
NAME:
pixels_cannon
PURPOSE:
determine continuum pixels using a Cannon-like technique (Ness et al. 2015)
INPUT:
Either:
a) Input for running the apogee.spec.cannon:
spec - spectra to fit (nspec,nlambda)
specerrs - errors on the spectra (nspec,nlambda); assume no covariances
label1, label2, ... - labels (nspec); best to subtract reference values before running this
type= ('lin') type of Cannon to run:
'lin' - linear Cannon
'quad' - quadratic Cannon
b) Output from a previous Cannon run:
coefficients - coefficients from the fit (ncoeffs,nlambda)
scatter - scatter from the fit (nlambda)
KEYWORDS:
baseline_dev= (0.015) maximum deviation from baseline
label1_max= (10.**-5.) maximum deviation in first linear coefficient
label2_max= (0.006) similar for the second
label3_max= (0.012) similar for the third
labelN_max= same with default 0.03
...
scatter_max= (0.015) maximum scatter of residuals
dr= (module-wide default) data release
OUTPUT:
Boolean index into the wavelength range with True for continuum pixels
HISTORY:
2015-02-05 - Written - Bovy (IAS@KITP)
"""
# Grab kwargs
type = kwargs.pop('type', 'lin')
dr = kwargs.pop('dr', path._default_dr())
# Parse input
if len(args) == 0: # Use default fit
from apogee.spec._train_cannon import load_fit
coeffs, scatter, baseline_labels = load_fit()
type = 'quad'
else:
spec = args[0]
specerr = args[1]
# Determine the type of input
if len(specerr.shape) == 2:
# Run the Cannon
if type.lower() == 'lin':
coeffs, scatter = cannon.linfit(*args)
elif type.lower() == 'quad':
coeffs, scatter = cannon.quadfit(*args)
else:
coeffs = spec
scatter = specerr
ncoeffs = coeffs.shape[0]
if type.lower() == 'lin':
nlabels = ncoeffs - 1
elif type.lower() == 'quad':
nlabels = int((-3 + numpy.sqrt(9 + 8 * (ncoeffs - 1)))) // 2
# Determine continuum pixels
out = numpy.ones(len(scatter), dtype='bool')
# Deviation from baseline
out[numpy.fabs(coeffs[0] - 1.) > kwargs.get('baseline_dev', 0.015)] = False
# Large dependence on labels
maxs = numpy.zeros(nlabels)
maxs[0] = kwargs.get('label1_max', 10.**-5.)
maxs[1] = kwargs.get('label2_max', 0.006)
maxs[2] = kwargs.get('label3_max', 0.012)
for ii in range(nlabels - 3):
maxs[ii + 3] = kwargs.get('label%i_max' % (ii + 4), 0.03)
for ii in range(1, nlabels + 1):
out[numpy.fabs(coeffs[ii]) > maxs[ii - 1]] = False
# Large residuals
out[scatter > kwargs.get('scatter_max', 0.015)] = False
_, _, _, aspcapDR12length = _aspcapPixelLimits(dr='12')
if int(dr) > 12 and coeffs.shape[1] == aspcapDR12length:
# Want continuum pixels on >DR12 ASPCAP grid, but using coefficients
# from <= DR12 grid
dr_module = path._default_dr()
path.change_dr(12)
out = toApStarGrid(out)
path.change_dr(dr)
out = toAspcapGrid(out)
path.change_dr(dr_module)
return out
def polylabels(spec,specerr,coeffs=None,scatter=None,poly='lin',
return_cov=False,return_poly=False,baseline_labels=None):
"""
NAME:
polylabels
PURPOSE:
Get the labels using a polynomial fit
INPUT:
spec - spectrum/a to fit (nlambda) or (nspec,nlambda)
specerrs - error/s on the spectra (nlambda) or (nspec,nlambda); assume no covariances
coeffs= array of coefficients from the polynomial fit (ncoeffs,nlambda); if not set the default fit is used
scatter= array of scatter from the polynomial fit (nlambda)); if not set the default fit is used
poly= ('lin') 'lin' or 'quad' currently
return_cov= (False) if True, return the uncertainty covariance matrix for the labels
return_poly= (False) if True, return the best-fit labels, labels-squared, etc.
baseline_labels= baseline to add to the labels (baseline that was subtracted before the fit)
OUTPUT:
Best-fit labels (nspec,nlabels)
if return_poly, the best-fit linear, quadratic, ... terms are returned
HISTORY:
2015-02-23 - Written - Bovy (IAS@KITP)
"""
if len(spec.shape) == 1:
spec= numpy.reshape(spec,(1,len(spec)))
specerr= numpy.reshape(specerr,(1,len(specerr)))
# Load default fit if necessary
if coeffs is None:
from apogee.spec._train_cannon import load_fit
coeffs, scatter, baseline_labels= load_fit()
poly= 'quad'
# Setup output
nspec= spec.shape[0]
ncoeffs= coeffs.shape[0]
if 'lin' in poly:
nlabels= ncoeffs-1
elif 'quad' in poly:
nlabels= int((-3+numpy.sqrt(9+8*(ncoeffs-1))))//2
if return_poly:
nout= ncoeffs-1
else:
nout= nlabels
out= numpy.empty((nspec,nout))
if return_cov:
outcov= numpy.empty((nspec,nout,nout))
# Run through the spectra
for ii in range(nspec):
labels= _polyfit_coeffs(spec[ii]-coeffs[0],specerr[ii],
scatter,coeffs[1:].T,
return_cov=return_cov)
if return_cov:
cov= labels[1]
labels= labels[0]
if return_poly:
out[ii]= labels
else:
out[ii]= labels[:nlabels]
if return_cov and return_poly:
outcov[ii]= cov
elif return_cov:
outcov[ii]= cov[:nlabels,:nlabels]
if not baseline_labels is None:
out+= baseline_labels
if return_cov:
return (out,outcov)
else:
return out
