def _non_linear_legendre(wcs_dict):
"""Returns a legendre model
Constructs a Legendre1D mathematical model
Parameters
----------
wcs_dict : dict
Dictionary containing all the wcs information decoded from the header and
necessary for constructing the Legendre1D model.
Returns
-------
`~astropy.modeling.Model`
"""
model = models.Legendre1D(degree=wcs_dict['order'] - 1,
domain=[wcs_dict['pmin'], wcs_dict['pmax']], )
new_params = [wcs_dict['fpar'][i] for i in range(wcs_dict['order'])]
model.parameters = new_params
return model
def _non_linear_legendre(self):
"""Set model to Legendre1D
"""
"""Returns a legendre model"""
self.model = models.Legendre1D(degree=self.wcs_dict['order'] - 1,
domain=[self.wcs_dict['pmin'],
self.wcs_dict['pmax']], )
# self.model.parameters[0] = self.wcs_dict['pmin']
for param_index in range(self.wcs_dict['order']):
self.model.parameters[param_index] = self.wcs_dict['fpar'][
param_index]
def test_iraf_non_linear_legendre(remote_data_path):
legendre_model = models.Legendre1D(degree=3, domain=[21, 4048])
legendre_model.c0.value = 5468.67555891
legendre_model.c1.value = 835.332144466
legendre_model.c2.value = -6.02202094803
legendre_model.c3.value = -1.13142953897
wavelength_axis = legendre_model(range(1, 4143)) * u.angstrom
spectrum_1d = Spectrum1D.read(remote_data_path, format='iraf')
assert isinstance(spectrum_1d, Spectrum1D)
assert_allclose(wavelength_axis, spectrum_1d.wavelength)
def test_linear_fitter_1dlegend(self):
"""
1 pset, 1 set 1D x, 1 set 1D y, Legendre 1D polynomial
"""
expected = np.array([[
1925.5000000000011, 3444.7500000000005, 1883.2500000000014,
364.4999999999996
]]).T
leg1 = models.Legendre1D(3)
leg1.parameters = [1, 2, 3, 4]
y1 = leg1(self.x1)
pfit = fitting.LinearLSQFitter()
model = pfit(leg1, self.x1, y1)
assert_allclose(model.param_sets, expected, atol=10**(-12))
p2 = astmodels.Polynomial1D(1)
p3 = astmodels.Polynomial1D(1)
p4 = astmodels.Polynomial1D(1)
m1 = p1 & p2
m2 = p3 & p4
m1.inverse = m2
return m1
test_models = [
astmodels.Identity(2),
astmodels.Polynomial1D(2, c0=1, c1=2, c2=3),
astmodels.Polynomial2D(1, c0_0=1, c0_1=2, c1_0=3),
astmodels.Shift(2.),
astmodels.Hermite1D(2, c0=2, c1=3, c2=0.5),
astmodels.Legendre1D(2, c0=2, c1=3, c2=0.5),
astmodels.Chebyshev1D(2, c0=2, c1=3, c2=0.5),
astmodels.Chebyshev2D(1, 1, c0_0=1, c0_1=2, c1_0=3),
astmodels.Legendre2D(1, 1, c0_0=1, c0_1=2, c1_0=3),
astmodels.Hermite2D(1, 1, c0_0=1, c0_1=2, c1_0=3),
astmodels.Scale(3.4),
astmodels.RotateNative2Celestial(5.63, -72.5, 180),
astmodels.Multiply(3),
astmodels.Multiply(10 * u.m),
astmodels.RotateCelestial2Native(5.63, -72.5, 180),
astmodels.EulerAngleRotation(23, 14, 2.3, axes_order='xzx'),
astmodels.Mapping((0, 1), n_inputs=3),
astmodels.Shift(2. * u.deg),
astmodels.Scale(3.4 * u.deg),
astmodels.RotateNative2Celestial(5.63 * u.deg, -72.5 * u.deg, 180 * u.deg),
astmodels.RotateCelestial2Native(5.63 * u.deg, -72.5 * u.deg, 180 * u.deg),
'Linear1D':
models.Linear1D(1.0, 0.0),
'Const1D':
models.Const1D(0.0),
'Redshift':
models.Redshift(0.0),
'Scale':
models.Scale(1.0),
'Shift':
models.Shift(0.0),
'Sine1D':
models.Sine1D(1.0, 1.0),
'Chebyshev1D':
models.Chebyshev1D(1),
'Legendre1D':
models.Legendre1D(1),
'Polynomial1D':
models.Polynomial1D(1),
}
# this nightmarish way of getting the function name results from the way
# astropy functional models store them. Both their '_name' and 'name'
# attributes are set to None, and a plain, easy to use name is nowhere
# to be seen. And worse, the name coding changed dramatically from
# astropy 0.4 to 1.0.
def getComponentName(function):
if issubclass(function.__class__, Fittable1DModel):
name = function.__class__()
return _getName(name)
elif issubclass(function.__class__, PolynomialModel):
def continuum(x,
y,
output='ratio',
degree=6,
n_iterate=5,
lower_threshold=2,
upper_threshold=3,
verbose=False,
weights=None) -> Union[Iterable, tuple, Callable]:
"""
Builds a polynomial continuum from segments of a spectrum,
given in the form of wl and flux arrays.
Parameters
----------
x : array-like
Independent variable
y : array-like
y = f(x)
output: string
Specifies what will be returned by the function
'ratio' = ratio between fitted continuum and the spectrum
'difference' = difference between fitted continuum and the spectrum
'function' = continuum function evaluated at x
degree : integer
Degree of polynomial for the fit
n_iterate : integer
Number of rejection iterations
lower_threshold : float
Lower threshold for point rejection in units of standard
deviation of the residuals
upper_threshold : float
Upper threshold for point rejection in units of standard
deviation of the residuals
verbose : boolean
Prints information about the fitting
weights : array-like
Weights for continuum fitting. Must be the shape of x and y.
Returns
-------
c : tuple
c[0]: numpy.ndarray
Input x coordinates
c[1]: numpy.ndarray
See parameter "output".
"""
assert not np.isnan(x).all(), 'All x values are NaN.'
assert not np.isnan(y).all(), 'All y values are NaN.'
x_full = copy.deepcopy(x)
# NOTE: For now, interp1d skips interpolation of NaNs.
s = interp1d(x, y)
if weights is None:
weights = np.ones_like(x)
if np.isnan(y).any():
nan_mask = np.isnan(s(x))
x = x[~nan_mask]
weights = copy.deepcopy(weights)[~nan_mask]
warnings.warn(
'NaN values found in data! Removed {:d} out of {:d} data points.'.
format(np.count_nonzero(nan_mask), len(x_full)),
category=RuntimeWarning,
)
model = models.Legendre1D(degree=degree)
fitter = fitting.LinearLSQFitter()
for i in range(n_iterate):
f = fitter(model, x, s(x), weights=weights)
res = s(x) - f(x)
sig = np.std(res)
rej_cond = ((res < upper_threshold * sig) &
(res > -lower_threshold * sig))
if np.sum(rej_cond) <= degree:
if verbose:
warnings.warn(
'Not enough fitting points. Stopped at iteration {:d}. sig={:.2e}'
.format(i, sig))
break
if np.sum(weights == 0.0) >= degree:
if verbose:
warnings.warn(
'Number of non-zero values in weights vector is lower than the polynomial degree. '
'Stopped at iteration {:d}. sig={:.2e}'.format(i, sig))
break
x = x[rej_cond]
weights = weights[rej_cond]
if verbose:
print('Final number of points used in the fit: {:d}'.format(len(x)))
print('Rejection ratio: {:.2f}'.format(1. - float(len(x)) /
float(len(x_full))))
p = fitter(model, x, s(x), weights=weights)
out = dict(
ratio=(x_full, s(x_full) / p(x_full)),
difference=(x_full, s(x_full) - p(x_full)),
function=(x_full, p(x_full)),
polynomial=p,
)
return out[output]
def rb_iter_contfit(wave, flux, error, **kwargs):
'''
Iterative continuum fitter using Legendre polynomials
Input: -
wavelength array
flux array
error array
optional input:
maxiter :- maximum iteration [25 default]
order :- polynomial order of fit [4 default]
output: -
fit_final : Final fitted continuum array
resid_final : residual error array
fit_error : error on the fit [standard deviation of the residual]
Written by: Rongmon Bordoloi
Tested on Python 3.7 Sep 4 2019
--------------------------
example :
from IGM import rb_iter_contfit as r
out= r.rb_iter_contfit(wave,flux,error,order=5)
out[0] = fitted continuum
'''
if 'maxiter' in kwargs:
maxiter = kwargs['maxiter']
else:
maxiter = 25
if 'order' in kwargs:
order = kwargs['order']
else:
order = 4
#Initialize a mask
mask = np.ones((np.size(wave), ))
# Looking for chip gaps
chip_gap = np.where(((error == 0) & (flux == 0)) | (error - flux == 0))
if (np.size(chip_gap) > 0):
#error[chip_gap]=1;
#flux[chip_gap]=1;
mask[chip_gap] = 0
# Now get rid of negative error values
qq = np.where(error <= 0)
if (np.size(qq) > 0):
error[qq] = np.median(error)
#Do a sanity check to avoid bad flux values
q = np.where(flux <= 0)
if (np.size(q) > 0):
flux[q] = error[q]
w = 1 / error**2. # Weight
index = 0 #Initialize the counter
outside_chip_gap = np.where(((error != 0) & (flux != 0))
| (error - flux != 0))
med_err = np.median(error[outside_chip_gap])
med_flux = np.median(flux[outside_chip_gap])
# Now take out any possible emission or absorption features
#%Flux values less than the median error are 1sigma within zero, so should be masked
#%Try to exclude emission. anything above 2sigma over the median flux should be masked
bd = np.where((flux < med_err)) #| (flux > (med_flux+3*med_err)))
nbd = len(bd[0])
if (nbd > 0):
mask[bd] = 0
mask = np.array(mask)
qq = np.where(mask == 1)
#Here I'm cutting out any chip gap or any other absorption feature. These are permanently excluded from the fit
flux_new = flux[qq]
wave_new = wave[qq]
weights = np.array(w[qq])
# Fit the data using Legedre Polynomials
g_init = models.Legendre1D(order)
# initialize fitters
fit = fitting.LevMarLSQFitter()
with warnings.catch_warnings():
# Ignore model linearity warning from the fitter
warnings.simplefilter('ignore')
new_fit = fitting.FittingWithOutlierRemoval(fit,
sigma_clip,
niter=maxiter,
sigma=3.0)
filtered_fit, filtered_data = new_fit(g_init, wave_new,
flux_new) #,weights=weights)
''''
while (index < maxiter):
print("Index="+np.str(index))
index=index+1
oldmask=mask
# Fit the data using Legedre Polynomials
g_init = models.Legendre1D(order)
# initialize fitters
fit = fitting.LevMarLSQFitter()
new_fit = fitting.FittingWithOutlierRemoval(fit, sigma_clip,niter=maxiter, sigma=3.0)
filtered_data, new_fit = new_fit(g_init, wave_new,flux_new,weights=weights)
#fit_g = fitting.LevMarLSQFitter()
#new_fit = fit_g(g_init, wave_new,flux_new) # Learn how to use the weights....
cont=new_fit(wave_new)
resid=flux_new/np.double(cont)
#mederr=np.sqrt(scipy.stats.moment(resid,2)) #;;Use the median error to set "sigma" for the clipping. If you median is skewed, i think you're pretty much hosed
mederr=np.std(resid)
# Do some sigma clipping here
median_val=np.median(resid)
kappa=3.; # 3 sigma clipping
bd = np.where( (resid > (median_val+kappa*mederr)) | (resid < (median_val - mederr*kappa)))
gd= np.where( (resid <= (median_val+kappa*mederr)) & (resid >= (median_val - mederr*kappa)))
print np.size(gd)
print np.size(bd)
if (np.size(gd) > 0):
mask[gd] = 1 # Allowing previously rejected points to reenter
if (np.size(bd) > 0):
mask[bd] = 0
if (index >0):
diff=oldmask-mask
qq = np.where(diff != 0.)
if (np.size(qq)==0):
print index
print "No more Clipping"
index=maxiter # ;;mask and oldmask are identical if all elements match
# Updating everything
qq=np.where(mask==1)
flux_new=flux[qq]
wave_new=wave[qq]
weights=w[qq]
'''
#pdb.set_trace()
fit_final = filtered_fit(wave)
resid_final = flux / fit_final
fit_error = np.std(
resid_final) #np.sqrt(scipy.stats.moment(resid_final,2))
return fit_final, resid_final, fit_error
# Get aperture lines
linecat = None
for ap in gdap:
linecat1 = getaplines(idlines,apinfo,ap,diag=diag)
print(ap,len(linecat1))
if linecat is None:
linecat = linecat1
else:
linecat = np.hstack((linecat,linecat1))
nlinecat = len(linecat)
# Fit 4th or 5th order polynomial plus a separate offset for each aperture
# 1) Initial polynomial fit to all lines in all fibers
gline, = np.where(linecat['match']==1)
func_init = models.Legendre1D(degree=order,domain=[0,npix])
fitter = fitting.LinearLSQFitter()
func1 = fitter(func_init,linecat['center'][gline],linecat['wave'][gline])
dwave1 = func1(linecat['center'][gline]) - linecat['wave'][gline]
#coef1 = dln.poly_fit(linecat['center'][gline],linecat['wave'][gline],4)
#dwave1 = dln.poly(linecat['center'][gline],coef1) - linecat['wave'][gline]
sigwave1 = dln.mad(dwave1)
print('Initial fit coefficients ',func1.parameters)
#print('Initial fit coefficients ',coef1)
print('Initial sigma = %f6.2' % sigwave1)
#xx = linecat['center'][gline]
#yy = linecat['wave'][gline]
#si = np.argsort(xx)
#spl = UnivariateSpline(xx[si],yy[si],k=3,s=10)
#dwave1 = spl(xx) - linecat['wave'][gline]
