Science/psdlag-agn
1
0
Fork 0
mirror of https://asciireactor.com/otho/psdlag-agn.git synced 2024-11-24 06:45:06 +00:00
Code Issues Packages Projects Releases Wiki Activity
7f9157eaa0
psdlag-agn/scripts/spec.py

5688 lines
189 KiB
Python
Executable File
Raw Blame History

"""
Suite to reduce spectroscopic data.
subfunctions:
calibrate
setheaders -- exptime, gain, readnoise, etc.
makeflat -- make median flat and noisy pixel map
makedark -- make median dark, and estimate noise in each pixel.
clean -- clean and replace bad pixels
extract
trace -- trace spectral orders
makeprofile -- compute mean spectral PSF (a spline) for an order
fitprofile -- fit given spline-PSF to a spectral cross-section
Utilities:
pickloc
fitPSF
"""
# 2010-07-02 10:56 IJC: Began the great endeavor.
try:
from astropy.io import fits as pyfits
except:
import pyfits
import matplotlib.pyplot as plt
import numpy as np
from scipy import optimize, interpolate
import pdb
[docs]
class baseObject:
"""Empty object container.
"""
def __init__(self):
return
########
# Utilities to add in a separate cohesive package at a later date:
from analysis import polyfitr, stdr, binarray, gaussian, egaussian
import analysis as an
from nsdata import bfixpix
########
# Parameters to put in a GUI:
gain = 5 # e-/ADU
readnoise = 25 # e-
#########
[docs]
def message(text):
"""Display a message; for now, with text."""
from sys import stdout
print text
stdout.flush()
[docs]
def pickloc(ax=None, zoom=10):
"""
:INPUTS:
ax : (axes instance) -- axes in which to pick a location
zoom : int -- zoom radius for target confirmation
: 2-tuple -- (x,y) radii for zoom confirmation.
"""
# 2011-04-29 19:26 IJC:
# 2011-09-03 20:59 IJMC: Zoom can now be a tuple; x,y not cast as int.
pickedloc = False
if ax is None:
ax = plt.gca()
axlimits = ax.axis()
if hasattr(zoom, '__iter__') and len(zoom)>1:
xzoom, yzoom = zoom
else:
xzoom = zoom
yzoom = zoom
while not pickedloc:
ax.set_title("click to select location")
ax.axis(axlimits)
x = None
while x is None:
selectevent = plt.ginput(n=1,show_clicks=False)
if len(selectevent)>0: # Prevent user from cancelling out.
x,y = selectevent[0]
#x = x.astype(int)
#y = y.astype(int)
if zoom is not None:
ax.axis([x-xzoom,x+xzoom,y-yzoom,y+yzoom])
ax.set_title("you selected xy=(%i,%i)\nclick again to confirm, or press Enter/Return to try again" %(x,y) )
plt.draw()
confirmevent = plt.ginput(n=1,show_clicks=False)
if len(confirmevent)>0:
pickedloc = True
loc = confirmevent[0]
return loc
[docs]
def fitTophat(vec, err=None, verbose=False, guess=None):
"""Fit a 1D tophat function to an input data vector.
Return the fit, and uncertainty estimates on that fit.
SEE ALSO: :func:`analysis.gaussian`"""
xtemp = np.arange(1.0*len(vec))
if guess is None: # Make some educated guesses as to the parameters:
pedestal = 0.5 * (0.8*np.median(vec) + 0.2*(vec[0]+vec[1]))
area = (vec-pedestal).sum()
centroid = (vec*xtemp).sum()/vec.sum()
if centroid<0:
centroid = 1.
elif centroid>len(vec):
centroid = len(vec)-2.
sigma = area/vec[int(centroid)]/np.sqrt(2*np.pi)
if sigma<=0:
sigma = 0.01
guess = [area,sigma,centroid,pedestal]
if verbose:
print 'Gaussian guess parameters>>', guess
if err is None:
fit, fitcov = optimize.leastsq(egaussian, guess, args=(xtemp, vec), full_output=True)[0:2]
pc.resfunc
fit, fitcov = optimize.leastsq(egaussian, guess, args=(xtemp, vec), full_output=True)[0:2]
else:
fit, fitcov = optimize.leastsq(egaussian, guess, args=(xtemp, vec, err), full_output=True)[0:2]
if fitcov is None: # The fitting was really bad!
fiterr = np.abs(fit)
else:
fiterr = np.sqrt(np.diag(fitcov))
if verbose:
print 'Best-fit parameters>>', fit
f = plt.figure()
ax = plt.axes()
plt.plot(xtemp, vec, 'o', \
xtemp, gaussian(fit, xtemp), '-', \
xtemp, gaussian(guess, xtemp), '--')
return fit, fiterr
[docs]
def fitGaussian(vec, err=None, verbose=False, guess=None):
"""Fit a Gaussian function to an input data vector.
Return the fit, and uncertainty estimates on that fit.
SEE ALSO: :func:`analysis.gaussian`"""
# 2012-12-20 13:28 IJMC: Make a more robust guess for the centroid.
xtemp = np.arange(1.0*len(vec))
if guess is None: # Make some educated guesses as to the parameters:
pedestal = (0.8*np.median(vec) + 0.2*(vec[0]+vec[1]))
area = (vec-pedestal).sum()
centroid = ((vec-pedestal)**2*xtemp).sum()/((vec-pedestal)**2).sum()
if centroid<0:
centroid = 1.
elif centroid>len(vec):
centroid = len(vec)-2.
#pdb.set_trace()
sigma = area/vec[int(centroid)]/np.sqrt(2*np.pi)
if sigma<=0:
sigma = .01
guess = [area,sigma,centroid,pedestal]
if err is None:
err = np.ones(vec.shape, dtype=float)
badvals = True - (np.isfinite(xtemp) * np.isfinite(err) * np.isfinite(vec))
vec[badvals] = np.median(vec[True - badvals])
err[badvals] = vec[True - badvals].max() * 1e9
if verbose:
print 'Gaussian guess parameters>>', guess
if not np.isfinite(xtemp).all():
pdb.set_trace()
if not np.isfinite(vec).all():
pdb.set_trace()
if not np.isfinite(err).all():
pdb.set_trace()
try:
fit, fitcov = optimize.leastsq(egaussian, guess, args=(xtemp, vec, err), full_output=True)[0:2]
except:
pdb.set_trace()
if fitcov is None: # The fitting was really bad!
fiterr = np.abs(fit)
else:
fiterr = np.sqrt(np.diag(fitcov))
#pdb.set_trace()
if verbose:
print 'Best-fit parameters>>', fit
f = plt.figure()
ax = plt.axes()
plt.plot(xtemp, vec, 'o', \
xtemp, gaussian(fit, xtemp), '-', \
xtemp, gaussian(guess, xtemp), '--')
return fit, fiterr
[docs]
def fitGaussiann(vec, err=None, verbose=False, guess=None, holdfixed=None):
"""Fit a Gaussian function to an input data vector.
Return the fit, and uncertainty estimates on that fit.
SEE ALSO: :func:`analysis.gaussian`"""
from phasecurves import errfunc
from analysis import fmin
xtemp = np.arange(1.0*len(vec))
if guess is None: # Make some educated guesses as to the parameters:
holdfixed = None
pedestal = 0.5 * (0.8*np.median(vec) + 0.2*(vec[0]+vec[1]))
area = (vec-pedestal).sum()
centroid = (vec*xtemp).sum()/vec.sum()
if centroid<0:
centroid = 1.
elif centroid>len(vec):
centroid = len(vec)-2.
sigma = area/vec[int(centroid)]/np.sqrt(2*np.pi)
if sigma<=0:
sigma = 0.01
guess = [area,sigma,centroid,pedestal]
if err is None:
err = np.ones(vec.shape, dtype=float)
badvals = True - (np.isfinite(xtemp) * np.isfinite(err) * np.isfinite(vec))
vec[badvals] = np.median(vec[True - badvals])
err[badvals] = vec[True - badvals].max() * 1e9
if verbose:
print 'Gaussian guess parameters>>', guess
if not np.isfinite(xtemp).all():
pdb.set_trace()
if not np.isfinite(vec).all():
pdb.set_trace()
if not np.isfinite(err).all():
pdb.set_trace()
try:
#fit, fitcov = optimize.leastsq(egaussian, guess, args=(xtemp, vec, err), full_output=True)[0:2]
fitargs = (gaussian, xtemp, vec, 1./err**2)
fit = fmin(errfunc, guess, args=fitargs, full_output=True, disp=False, holdfixed=holdfixed)[0]
fitcov = None
except:
pdb.set_trace()
if fitcov is None: # The fitting was really bad!
fiterr = np.abs(fit)
else:
fiterr = np.sqrt(np.diag(fitcov))
if verbose:
print 'Best-fit parameters>>', fit
f = plt.figure()
ax = plt.axes()
plt.plot(xtemp, vec, 'o', \
xtemp, gaussian(fit, xtemp), '-', \
xtemp, gaussian(guess, xtemp), '--')
return fit, fiterr
[docs]
def fit2Gaussian(vec, err=None, verbose=False, guess=None, holdfixed=None):
"""Fit two Gaussians simultaneously to an input data vector.
:INPUTS:
vec : sequence
1D array or list of values to fit to
err : sequence
uncertainties on vec
guess : sequence
guess parameters: [area1, sigma1, cen1, area2, sig2, cen2, constant].
Note that parameters are in pixel units.
holdfixed : sequence
parameters to hold fixed in analysis, _IF_ guess is passed in.
SEE ALSO: :func:`analysis.gaussian`, :func:`fitGaussian`"""
# 2012-03-14 14:33 IJMC: Created
from tools import sumfunc
from phasecurves import errfunc
from analysis import fmin # Don't use SciPy, because I want keywords!
xtemp = np.arange(1.0*len(vec))
if err is None:
err = np.ones(xtemp.shape)
else:
err = np.array(err, copy=False)
if guess is None: # Make some semi-educated guesses as to the parameters:
holdfixed = None
pedestal = 0.5 * (0.8*np.median(vec) + 0.2*(vec[0]+vec[1]))
area = (vec-pedestal).sum()
centroid = (vec*xtemp).sum()/vec.sum()
if centroid<0:
centroid = 1.
elif centroid>len(vec):
centroid = len(vec)-2.
sigma = area/vec[int(centroid)]/np.sqrt(2*np.pi)
if sigma<=0:
sigma = 0.01
guess1 = [area/2.,sigma/1.4,centroid+xtemp.size/5.]
guess2 = [area/2.,sigma/1.4,centroid-xtemp.size/5.]
guess = guess1 + guess2 + [pedestal]
## Testing:
#mod = sumfunc(guess, gaussian, gaussian, 3, 4, args1=(xtemp,), args2=(xtemp,))
#fitargs = (sumfunc, gaussian, gaussian, 3, 4, (xtemp,), (xtemp,), None, vec, 1./err**2)
#fitkw = dict(useindepvar=False, testfinite=False)
#chisq = errfunc(guess, *fitargs, **fitkw)
#thisfit = fmin(errfunc, guess, args=fitargs, kw=fitkw, full_output=True)
#mod2 = sumfunc(thisfit[0], gaussian, gaussian, 3, 4, args1=(xtemp,), args2=(xtemp,))
if verbose:
print '2-Gaussian guess parameters>>', guess
fitargs = (sumfunc, gaussian, gaussian, 3, 4, (xtemp,), (xtemp,), vec, 1./err**2)
fitkw = dict(useindepvar=False, testfinite=False)
fit = fmin(errfunc, guess, args=fitargs, kw=fitkw, full_output=True, disp=False, holdfixed=holdfixed)
if verbose:
model = sumfunc(fit[0], gaussian, gaussian, 3, 4, args1=(xtemp,), args2=(xtemp,))
model1 = gaussian(fit[0][0:3], xtemp)
model2 = gaussian(fit[0][3:6], xtemp)
print 'Best-fit parameters>>', fit[0]
f = plt.figure()
ax = plt.axes()
plt.plot(xtemp, vec, 'o', \
xtemp, model, '--')
plt.plot(xtemp, model1+fit[0][6], ':')
plt.plot(xtemp, model2+fit[0][6], ':')
return fit[0]
[docs]
def fitPSF(ec, guessLoc, fitwidth=20, verbose=False, sigma=5, medwidth=6, err_ec=None):
"""
Helper function to fit 1D PSF near a given region. Assumes
spectrum runs horizontally across the frame!
ec : 2D numpy array
echellogram array, with horizontal dispersion direction
guessLoc : 2-tuple
A slight misnomer for this (x,y) tuple: y is a guess and will
be fit, but x is the coordinate at which the fitting takes
place.
fitwidth : int
width of cross-dispersion direction to use in fitting
medwidth : int
number of columns to average over when fitting a profile
verbose : bool
verbosity/debugging printout flag
sigma : scalar
sigma scale for clipping bad values
"""
# 2010-08-24 22:00 IJC: Added sigma option
# 2010-11-29 20:54 IJC: Added medwidth option
# 2011-11-26 23:17 IJMC: Fixed bug in computing "badval"
# 2012-04-27 05:15 IJMC: Now allow error estimates to be passed in.
# 2012-04-28 08:48 IJMC: Added better guessing for initial case.
if verbose<0:
verbose = False
ny, nx = ec.shape
x = guessLoc[0].astype(int)
y = guessLoc[1].astype(int)
# Fit the PSF profile at the initial, selected location:
ymin = max(y-fitwidth/2, 0)
ymax = min(y+fitwidth/2, ny)
xmin = max(x-medwidth/2, 0)
xmax = min(x+medwidth/2, nx)
if verbose:
message("Sampling: ec[%i:%i,%i:%i]"%(ymin,ymax,xmin,xmax))
firstSeg = np.median(ec[ymin:ymax,xmin:xmax],1)
if verbose:
print firstSeg
if err_ec is None:
ey = stdr(firstSeg, sigma)
badval = abs((firstSeg-np.median(firstSeg))/ey) > sigma
err = np.ones(firstSeg.shape, float)
err[badval] = 1e9
else:
err = np.sqrt((err_ec[ymin:ymax,xmin:xmax]**2).mean(1))
err[True - np.isfinite(err)] = err[np.isfinite(err)].max() * 1e9
guessAmp = (an.wmean(firstSeg, 1./err**2) - np.median(firstSeg)) * fitwidth
if not np.isfinite(guessAmp):
pdb.set_trace()
#fit, efit = fitGaussian(firstSeg, verbose=verbose, err=err, guess=[guessAmp[0], 5, fitwidth/2., np.median(firstSeg)])
fit, efit = fitGaussian(firstSeg, verbose=verbose, err=err, guess=None)
newY = ymin+fit[2]
err_newY = efit[2]
if verbose:
message("Initial position: (%3.2f,%3.2f)"%(x,newY))
return x, newY, err_newY
[docs]
def traceorders(filename, pord=5, dispaxis=0, nord=1, verbose=False, ordlocs=None, stepsize=20, fitwidth=20, plotalot=False, medwidth=6, xylims=None, uncertainties=None, g=gain, rn=readnoise, badpixelmask=None, retsnr=False, retfits=False):
"""
Trace spectral orders for a specified filename.
filename : str OR 2D array
full path and filename to a 2D echelleogram FITS file, _OR_
a 2D numpy array representing such a file.
OPTIONAL INPUTS:
pord : int
polynomial order of spectral order fit
dispaxis : int
set dispersion axis: 0 = horizontal and = vertical
nord : int
number of spectral orders to trace.
ordlocs : (nord x 2) numpy array
Location to "auto-click" and
verbose: int
0,1,2; whether (and how much) to print various verbose debugging text
stepsize : int
number of pixels to step along the spectrum while tracing
fitwidth : int
number of pixels to use when fitting a spectrum's cross-section
medwidth : int
number of columns to average when fitting profiles to echelleograms
plotalot : bool
Show pretty plot? If running in batch mode (using ordlocs)
default is False; if running in interactive mode (ordlocs is
None) default is True.
xylims : 4-sequence
Extract the given subset of the data array: [xmin, xmax, ymin, ymax]
uncertainties : str OR 2D array
full path and filename to a 2D uncertainties FITS file, _OR_
a 2D numpy array representing such a file.
If this is set, 'g' and 'rn' below are ignored. This is
useful if, e.g., you are analyzing data which have already
been sky-subtracted, nodded on slit, or otherwise altered.
But note that this must be the same size as the input data!
g : scalar > 0
Detector gain, electrons per ADU (for setting uncertainties)
rn : scalar > 0
Detector read noise, electrons (for setting uncertainties)
retsnr : bool
If true, also return the computed S/N of the position fit at
each stepped location.
retfits : bool
If true, also return the X,Y positions at each stepped location.
:RETURNS:
(nord, pord) shaped numpy array representing the polynomial
coefficients for each order (suitable for use with np.polyval)
:NOTES:
If tracing fails, a common reason can be that fitwidth is too
small. Try increasing it!
"""
# 2010-08-31 17:00 IJC: If best-fit PSF location goes outside of
# the 'fitwidth' region, nan values are returned that don't
# cause the routine to bomb out quite so often.
# 2010-09-08 13:54 IJC: Updated so that going outside the
# 'fitwidth' region is determined only in the local
# neighborhood, not in relation to the (possibly distant)
# initial guess position.
# 2010-11-29 20:56 IJC: Added medwidth option
# 2010-12-17 09:32 IJC: Changed color scaling options; added xylims option.
# 2012-04-27 04:43 IJMC: Now perform _weighted_ fits to the measured traces!
global gain
global readnoise
if verbose < 0:
verbose = 0
if not g==gain:
message("Setting gain to " + str(g))
gain = g
if not rn==readnoise:
message("Setting readnoise to " + str(rn))
readnoise = rn
if ordlocs is not None:
ordlocs = np.array(ordlocs, copy=False)
if ordlocs.ndim==1:
ordlocs = ordlocs.reshape(1, ordlocs.size)
autopick = True
else:
autopick = False
plotalot = (not autopick) or plotalot
if isinstance(filename, np.ndarray):
ec = filename.copy()
else:
try:
ec = pyfits.getdata(filename)
except:
message("Could not open filename %s" % filename)
return -1
if isinstance(uncertainties, np.ndarray):
err_ec = uncertainties.copy()
else:
try:
err_ec = pyfits.getdata(uncertainties)
except:
err_ec = np.sqrt(ec * gain + readnoise**2)
if dispaxis<>0:
ec = ec.transpose()
err_ec = err_ec.transpose()
if verbose: message("Took transpose of echelleogram to rotate dispersion axis")
else:
pass
if xylims is not None:
try:
ec = ec[xylims[0]:xylims[1], xylims[2]:xylims[3]]
err_ec = err_ec[xylims[0]:xylims[1], xylims[2]:xylims[3]]
except:
message("Could not extract subset: ", xylims)
return -1
if badpixelmask is None:
badpixelmask = np.zeros(ec.shape, dtype=bool)
else:
if not hasattr(badpixelmask, 'shape'):
badpixelmask = pyfits.getdata(badpixelmask)
if xylims is not None:
badpixelmask = badpixelmask[xylims[0]:xylims[1], xylims[2]:xylims[3]]
err_ec[badpixelmask.nonzero()] = err_ec[np.isfinite(err_ec)].max() * 1e9
try:
ny, nx = ec.shape
except:
message("Echellogram file %s does not appear to be 2D: exiting." % filename)
return -1
if plotalot:
f = plt.figure()
ax = plt.axes()
plt.imshow(ec, interpolation='nearest',aspect='auto')
sortedvals = np.sort(ec.ravel())
plt.clim([sortedvals[nx*ny*.01], sortedvals[nx*ny*.99]])
#plt.imshow(np.log10(ec-ec.min()+np.median(ec)),interpolation='nearest',aspect='auto')
ax.axis([0, nx, 0, ny])
orderCoefs = np.zeros((nord, pord+1), float)
position_SNRs = []
xyfits = []
if not autopick:
ordlocs = np.zeros((nord, 2),float)
for ordernumber in range(nord):
message('Selecting %i orders; please click on order %i now.' % (nord, 1+ordernumber))
plt.figure(f.number)
guessLoc = pickloc(ax, zoom=fitwidth)
ordlocs[ordernumber,:] = guessLoc
ax.plot([guessLoc[0]],[guessLoc[1]], '*k')
ax.axis([0, nx, 0, ny])
plt.figure(f.number)
plt.draw()
if verbose:
message("you picked the location: ")
message(guessLoc)
for ordernumber in range(nord):
guessLoc = ordlocs[ordernumber,:]
xInit, yInit, err_yInit = fitPSF(ec, guessLoc, fitwidth=fitwidth,verbose=verbose, medwidth=medwidth, err_ec=err_ec)
if plotalot:
ax.plot([xInit],[yInit], '*k')
ax.axis([0, nx, 0, ny])
plt.figure(f.number)
plt.draw()
if verbose:
message("Initial (fit) position: (%3.2f,%3.2f)"%(xInit,yInit))
# Prepare to fit PSFs at multiple wavelengths.
# Determine the other positions at which to fit:
xAbove = np.arange(1, np.ceil(1.0*(nx-xInit)/stepsize))*stepsize + xInit
xBelow = np.arange(-1,-np.ceil((1.+xInit)/stepsize),-1)*stepsize + xInit
nAbove = len(xAbove)
nBelow = len(xBelow)
nToMeasure = nAbove + nBelow + 1
iInit = nBelow
if verbose:
message("Going to measure PSF at the following %i locations:"%nToMeasure )
message(xAbove)
message(xBelow)
# Measure all positions "above" the initial selection:
yAbove = np.zeros(nAbove,float)
err_yAbove = np.zeros(nAbove,float)
lastY = yInit
for i_meas in range(nAbove):
guessLoc = xAbove[i_meas], lastY
thisx, thisy, err_thisy = fitPSF(ec, guessLoc, fitwidth=fitwidth, verbose=verbose-1, medwidth=medwidth, err_ec=err_ec)
if abs(thisy - yInit)>fitwidth/2:
thisy = yInit
err_thisy = yInit
lastY = yInit
else:
lastY = thisy.astype(int)
yAbove[i_meas] = thisy
err_yAbove[i_meas] = err_thisy
if verbose:
print thisx, thisy
if plotalot and not np.isnan(thisy):
#ax.plot([thisx], [thisy], 'xk')
ax.errorbar([thisx], [thisy], [err_thisy], fmt='xk')
# Measure all positions "below" the initial selection:
yBelow = np.zeros(nBelow,float)
err_yBelow = np.zeros(nBelow,float)
lastY = yInit
for i_meas in range(nBelow):
guessLoc = xBelow[i_meas], lastY
thisx, thisy, err_thisy = fitPSF(ec, guessLoc, fitwidth=fitwidth, verbose=verbose-1, medwidth=medwidth, err_ec=err_ec)
if abs(thisy-lastY)>fitwidth/2:
thisy = np.nan
else:
lastY = thisy.astype(int)
yBelow[i_meas] = thisy
err_yBelow[i_meas] = err_thisy
if verbose:
print thisx, thisy
if plotalot and not np.isnan(thisy):
ax.errorbar([thisx], [thisy], [err_thisy], fmt='xk')
# Stick all the fit positions together:
yPositions = np.concatenate((yBelow[::-1], [yInit], yAbove))
err_yPositions = np.concatenate((err_yBelow[::-1], [err_yInit], err_yAbove))
xPositions = np.concatenate((xBelow[::-1], [xInit], xAbove))
if verbose:
message("Measured the following y-positions:")
message(yPositions)
theseTraceCoefs = polyfitr(xPositions, yPositions, pord, 3, \
w=1./err_yPositions**2, verbose=verbose)
orderCoefs[ordernumber,:] = theseTraceCoefs
# Plot the traces
if plotalot:
ax.plot(np.arange(nx), np.polyval(theseTraceCoefs,np.arange(nx)), '-k')
ax.plot(np.arange(nx), np.polyval(theseTraceCoefs,np.arange(nx))+fitwidth/2, '--k')
ax.plot(np.arange(nx), np.polyval(theseTraceCoefs,np.arange(nx))-fitwidth/2, '--k')
ax.axis([0, nx, 0, ny])
plt.figure(f.number)
plt.draw()
if retsnr:
position_SNRs.append(yPositions / err_yPositions)
if retfits:
xyfits.append((xPositions, yPositions))
# Prepare for exit and return:
ret = (orderCoefs,)
if retsnr:
ret = ret + (position_SNRs,)
if retfits:
ret = ret + (xyfits,)
if len(ret)==1:
ret = ret[0]
return ret
[docs]
def makeprofile(filename, trace, **kw): #dispaxis=0, fitwidth=20, verbose=False, oversamp=4, nsigma=20, retall=False, neg=False, xylims=None, rn=readnoise, g=gain, extract_radius=10, bkg_radii=[15, 20], bkg_order=0, badpixelmask=None, interp=False):
"""
Make a spatial profile from a spectrum, given its traced location.
We interpolate the PSF at each pixel to a common reference frame,
and then average them.
filename : str _OR_ 2D numy array
2D echellogram
trace : 1D numpy array
set of polynomial coeficients of order (P-1)
dispaxis : int
set dispersion axis: 0 = horizontal and = vertical
fitwidth : int
Total width of extracted spectral sub-block. WIll always be
increased to at least twice the largest value of bkg_radii.
neg : bool scalar
set True for a negative spectral trace
nsigma : scalar
Sigma-clipping cut for bad pixels (beyond read+photon
noise). Set it rather high, and feel free to experiment with
this parameter!
xylims : 4-sequence
Extract the given subset of the data array: [xmin, xmax, ymin, ymax]
retall : bool
Set True to output several additional parameters (see below)
rn : scalar
Read noise (electrons)
g : scalar
Detector gain (electrons per data unit)
bkg_radii : 2-sequence
Inner and outer radius for background computation and removal;
measured in pixels from the center of the profile.
bkg_order : int > 0
Polynomial order of background trend computed in master spectral profile
interp : bool
Whether to (bi-linearly) interpolate each slice to produce a
precisely-centered spectral profile (according to the input
'trace'). If False, slices will only be aligned to the
nearest pixel.
OUTPUT:
if retall:
a spline-function that interpolates pixel locations onto the mean profile
a stack of data slices
estimates of the uncertainties
good pixel flag
list of splines
else:
a spline-function that interpolates pixel locations onto the mean profile
"""
from scipy import signal
#import numpy as np
# 2010-12-17 10:22 IJC: Added xylims option
# 2012-03-15 06:34 IJMC: Updated documentation.
# 2012-04-24 16:43 IJMC: Now properly update gain and readnoise,
# as neccessary. Much better flagging of
# bad pixels. Fixed interpolation on RHS.
# 2012-08-19 17:53 IJMC: Added dispaxis option.
# 2012-09-05 10:17 IJMC: Made keyword options into a dict.
global gain
global readnoise
# Parse inputs:
names = ['dispaxis', 'fitwidth', 'verbose', 'oversamp', 'nsigma', 'retall', 'neg', 'xylims', 'rn', 'g', 'extract_radius', 'bkg_radii', 'bkg_order', 'badpixelmask', 'interp']
defaults = [0, 20, False, 4, 20, False, False, None, readnoise, gain, 10, [15, 20], 0, None, False]
for n,d in zip(names, defaults):
#exec('%s = [d, kw["%s"]][kw.has_key("%s")]' % (n, n, n))
exec('%s = kw["%s"] if kw.has_key("%s") else d' % (n, n, n))
if fitwidth < (bkg_radii[1]*2):
fitwidth = bkg_radii[1]*2
if not g==gain:
message("Setting gain to " + str(g))
gain = g
if not rn==readnoise:
message("Setting readnoise to " + str(rn))
readnoise = rn
if verbose:
f = plt.figure()
ax = plt.axes()
# Check whether we have a filename, or an array:
if isinstance(filename, np.ndarray):
ec = filename.copy()
else:
try:
ec = pyfits.getdata(filename)
except:
message("Could not open filename %s" % filename)
return -1
if neg:
ec *= -1
if xylims is not None:
try:
ec = ec[xylims[0]:xylims[1], xylims[2]:xylims[3]]
if verbose:
message("Extracted subset: " + str(xylims))
except:
message("Could not extract subset: " + str(xylims))
return -1
if badpixelmask is None:
badpixelmask = np.zeros(ec.shape, dtype=bool)
else:
if not hasattr(badpixelmask, 'shape'):
badpixelmask = pyfits.getdata(badpixelmask)
if xylims is not None:
badpixelmask = badpixelmask[xylims[0]:xylims[1], xylims[2]:xylims[3]]
trace = np.array(trace,copy=True)
if len(trace.shape)>1:
if verbose:
message("Multi-order spectrum input...")
#rets = []
rets = [makeprofile(filename, thistrace, dispaxis=dispaxis, fitwidth=fitwidth,verbose=verbose,oversamp=oversamp,nsigma=nsigma,retall=retall, neg=neg, xylims=xylims, rn=rn, g=g, extract_radius=extract_radius, bkg_radii=bkg_radii, bkg_order=bkg_order,interp=interp, badpixelmask=badpixelmask) for thistrace in trace]
return rets
if dispaxis==1:
if verbose: message("Transposing spectra and bad pixel mask...")
ec = ec.transpose()
badpixelmask = badpixelmask.transpose()
if verbose:
message("Making a spatial profile...")
ny, nx = ec.shape
pord = len(trace) - 1
if verbose:
message("Echellogram of size (%i,%i)" % (nx,ny))
message("Polynomial of order %i (i.e., %i coefficients)" % (pord, pord+1))
xPositions = np.arange(nx)
yPositions = np.polyval(trace, xPositions)
yPixels = yPositions.astype(int)
# Placeholder for a more comprehensive workaround:
if (yPixels.max()+fitwidth/2)>=ny:
message("Spectrum may be too close to the upper boundary; try decreasing fitwidth")
if (yPixels.min()-fitwidth/2)<0:
message("Spectrum may be too close to the lower boundary; try decreasing fitwidth")
yProfile = np.linspace(-fitwidth/2, fitwidth/2-1,fitwidth*oversamp)
#yProfile2 = np.linspace(-fitwidth/2, fitwidth/2-1,(fitwidth+1)*oversamp)
profileStack = np.zeros((nx,fitwidth),float)
badpixStack = np.zeros((nx,fitwidth),bool)
profileSplineStack = np.zeros((nx,(fitwidth)*oversamp),float)
xProf = np.arange(fitwidth, dtype=float) - fitwidth/2.
xProf2 = np.arange(ny, dtype=float)
# Loop through to extract the spectral cross-sections at each pixel:
for i_pix in range(nx):
xi = xPositions[i_pix]
yi = yPixels[i_pix]
ymin = max(yi-fitwidth/2, 0)
ymax = min(yi+fitwidth/2, ny)
# Ensure that profile will always be "fitwidth" wide:
if (ymax-ymin)<fitwidth and ymin==0:
ymax = fitwidth
elif (ymax-ymin)<fitwidth and ymax==ny:
ymin = ymax - fitwidth
profile = ec[ymin:ymax,xi]
if interp:
profile = np.interp(xProf + yPositions[i_pix], xProf2[ymin:ymax], profile)
if verbose:
print 'i_pix, xi, yi, y0, ymin, ymax', i_pix, xi, yi, yPositions[i_pix], ymin, ymax
try:
profileStack[i_pix] = profile.copy() * gain
badpixStack[i_pix] = badpixelmask[ymin:ymax, xi]
except:
print "Busted!!!!"
print "profileStack[i_pix].shape, profile.shape, gain", \
profileStack[i_pix].shape, profile.shape, gain
stop
# Use the profile stack to flag bad pixels in the interpolation process:
medianProfile = np.median(profileStack,0)
scaledProfileStack = 0*profileStack
myMatrix = np.linalg.pinv(np.vstack((medianProfile, np.ones(profileStack.shape[1]))).transpose())
for ii in range(profileStack.shape[0]):
these_coefs = np.dot(myMatrix, profileStack[ii])
scaledProfileStack[ii] = (profileStack[ii] - these_coefs[1]) / these_coefs[0]
#import pylab as py
#pdb.set_trace()
#########
# Flag all the bad pixels:
#########
# Hot Pixels should be 1-pixel wide:
nFilt = 3
filteredProfileStack = signal.medfilt2d(profileStack, nFilt)
errorFiltStack = np.sqrt(filteredProfileStack + readnoise**2)
errorFiltStack[badpixStack] = errorFiltStack.max() * 1e9
goodPixels = (True - badpixStack) * np.abs((profileStack - filteredProfileStack) / errorFiltStack) < nsigma
badPixels = True - goodPixels
#cleaned_profileStack = ns.bfixpix(profStack, badpix, retdat=True)
if False: # Another (possibly worse?) way to do it:
stdProfile = stdr(scaledProfileStack, nsigma=nsigma, axis=0) #profileStack.std(0)
goodPixels = (scaledProfileStack-medianProfile.reshape(1,fitwidth)) / \
stdProfile.reshape(1,fitwidth)<=nsigma
goodPixels *= profileStack>=0
goodColumns = np.ones(goodPixels.shape[0],bool)
errorStack = np.sqrt(profileStack + readnoise**2)
errorStack[badPixels] = profileStack[goodPixels].max() * 1e9
# Loop through and fit weighted splines to each cross-section
for i_pix in range(nx):
xi = xPositions[i_pix]
yi = yPixels[i_pix]
ymin = max(yi-fitwidth/2, 0)
ymax = min(yi+fitwidth/2, ny)
# Ensure that profile will always be "fitwidth" wide:
if (ymax-ymin)<fitwidth and ymin==0:
ymax = fitwidth
elif (ymax-ymin)<fitwidth and ymax==ny:
ymin = ymax - fitwidth
profile = profileStack[i_pix]
index = goodPixels[i_pix]
ytemp = np.arange(ymin,ymax)-yPositions[i_pix]
if verbose>1:
print "ii, index.sum()>>",i_pix,index.sum()
print "xi, yi>>",xi,yi
if verbose:
print "ymin, ymax>>",ymin,ymax
print "ytemp.shape, profile.shape, index.shape>>",ytemp.shape, profile.shape, index.shape
if index.sum()>fitwidth/2:
ydata_temp = np.concatenate((ytemp[index], ytemp[index][-1] + np.arange(1,6)))
pdata_temp = np.concatenate((profile[index], [profile[index][-1]]*5))
profileSpline = interpolate.UnivariateSpline(ydata_temp, pdata_temp, k=3.0,s=0.0)
profileSplineStack[i_pix,:] = profileSpline(yProfile)
else:
if verbose: message("not enough good pixels in segment %i" % i_pix)
profileSplineStack[i_pix,:] = 0.0
goodColumns[i_pix] = False
finalYProfile = np.arange(fitwidth)-fitwidth/2
finalSplineProfile = np.median(profileSplineStack[goodColumns,:],0)
## Normalize the absolute scaling of the final Profile
#xProfile = np.arange(-50.*fitwidth, 50*fitwidth)/100.
#profileParam = fitGaussian(finalSplineProfile, verbose=True)
#profileScaling = profileParam[0]*.01
#print 'profileScaling>>', profileScaling
backgroundAperture = (np.abs(yProfile) > bkg_radii[0]) * (np.abs(yProfile) < bkg_radii[1])
#pdb.set_trace()
backgroundFit = np.polyfit(yProfile[backgroundAperture], finalSplineProfile[backgroundAperture], bkg_order)
extractionAperture = (np.abs(yProfile) < extract_radius)
normalizedProfile = finalSplineProfile - np.polyval(backgroundFit, yProfile)
normalizedProfile *= oversamp / normalizedProfile[extractionAperture].sum()
finalSpline = interpolate.UnivariateSpline(yProfile,
normalizedProfile, k=3.0, s=0.0)
if verbose:
ax.plot(yProfile, finalSplineProfile, '--',linewidth=2)
plt.draw()
if retall==True:
ret = finalSpline, profileStack, errorStack, goodPixels,profileSplineStack
else:
ret = finalSpline
return ret
def makeFitModel(param, spc,profile, xtemp=None):
scale, background, shift = param[0:3]
npix = len(spc)
if xtemp is None:
xtemp = np.arange(-npix/2,npix/2,dtype=float)
model = scale*profile(xtemp-shift)+background
return model
[docs]
def profileError(param, spc, profile, w, xaxis=None, retresidual=False):
"Compute the chi-squared error on a spectrum vs. a profile "
# 2012-04-25 12:59 IJMC: Slightly optimised (if retresidual is False)
#scale, background, shift = param[0:3]
#npix = len(spc)
#xtemp = np.arange(-npix/2,npix/2,dtype=float)
interpolatedProfile = makeFitModel(param,spc,profile, xtemp=xaxis) #scale*profile(xtemp-shift)+background
#wresiduals =
#wresiduals = (w*((spc-interpolatedProfile))**2).sum()
if retresidual:
ret = (w**0.5) * (spc - interpolatedProfile)
else:
ret = (w * (spc - interpolatedProfile)**2).sum()
return ret
[docs]
def fitprofile(spc, profile, w=None,verbose=False, guess=None, retall=False):
"""Fit a spline-class spatial profile to a spectrum cross-section
"""
# 2010-11-29 14:14 IJC: Added poor attempt at catching bad pixels
import analysis as an
import numpy as np
spc = np.array(spc)
try:
npix = len(spc)
except:
npix = 0
if w is None:
w = np.ones(spc.shape)
else:
w[True - np.isfinite(w)] = 0.
if guess is None:
guessParam = [1., 0., 0.]
else:
guessParam = guess
if verbose>1:
#message("profileError>>"+str(profileError))
#message("guessParam>>"+str(guessParam))
#message("spc>>"+str(spc))
message("w>>"+str(w))
good_index = w<>0.0
xaxis = np.arange(-npix/2,npix/2,dtype=float) # take outside of fitting loop
bestFit = optimize.fmin_powell(profileError, guessParam, args=(spc,profile,w * good_index, xaxis),disp=verbose, full_output=True)
best_bic = bestFit[1] + len(bestFit[0]) * np.log(good_index.sum())
keepCleaning = True
if False:
while keepCleaning is True:
if verbose: print "best bic is: ", best_bic
#w_residuals = profileError(bestFit[0], spc, profile, w, retresidual=True)
#good_values = (abs(w_residuals)<>abs(w_residuals).max())
diffs = np.hstack(([0], np.diff(spc)))
d2 = diffs * (-np.ones(len(spc)))**np.arange(len(spc))
d3 = np.hstack((np.vstack((d2[1::], d2[0:len(spc)-1])).mean(0), [0]))
good_values = abs(d3)<>abs(d3).max()
good_index *= good_values
xaxis = np.arange(-npix/2,npix/2,dtype=float) # take outside of fitting loop
latestFit = optimize.fmin_powell(profileError, guessParam, args=(spc,profile,w * good_index, xaxis),disp=verbose, full_output=True)
latest_bic = latestFit[1] + len(latestFit[0]) * np.log(good_index.sum())
if latest_bic < best_bic:
best_bic = latest_bic
bestFit = latestFit
keepCleaning = True
else:
keepCleaning = False
if good_index.any() is False:
keepCleaning = False
#good_index = good_index * an.removeoutliers(w_residuals, 5, retind=True)[1]
if verbose:
message("initial guess chisq>>%3.2f" % profileError(guessParam, spc, profile,w,xaxis))
message("final fit chisq>>%3.2f" % profileError(bestFit[0], spc, profile,w, xaxis))
if retall:
ret = bestFit
else:
ret = bestFit[0]
return ret
[docs]
def fitProfileSlices(splineProfile, profileStack, stdProfile, goodPixels,verbose=False, bkg_radii=None, extract_radius=None):
"""Fit a given spatial profile to a spectrum
Helper function for :func:`extractSpectralProfiles`
"""
npix, fitwidth = profileStack.shape
stdProfile = stdProfile.copy()
goodPixels[np.isnan(stdProfile) + stdProfile==0] = False
stdProfile[True - goodPixels] = 1e9
#varPData = stdProfile**2
if extract_radius is None:
extract_radius = fitwidth
x2 = np.arange(npix)
x = np.arange(-fitwidth/2, fitwidth/2)
extractionAperture = np.abs(x) < extract_radius
nextract = extractionAperture.sum()
fitparam = np.zeros((npix, 3),float)
fitprofiles = np.zeros((npix,nextract),float)
tempx = np.arange(-fitwidth/2,fitwidth/2,dtype=float)
# Start out by getting a rough estimate of any additional bad
# pixels (to flag):
#backgroundAperture = (np.abs(x) > bkg_radii[0]) * (np.abs(x) < bkg_radii[1])
#background = an.wmean(profileStack[:, backgroundAperture], (goodPixels/varPData)[:, backgroundAperture], axis=1)
#badBackground = True - np.isfinite(background)
#background[badBackground] = 0.
#subPData = profileStack - background
#standardSpectrum = an.wmean(subPData[:, extractionAperture], (goodPixels/varPData)[:,extractionAperture], axis=1) * extractionAperture.sum()
if False:
varStandardSpectrum = an.wmean(varPData[:, extractionAperture], goodPixels[:, extractionAperture], axis=1) * extractionAperture.sum()
badSpectrum = True - np.isfinite(standardSpectrum)
standardSpectrum[badSpectrum] = 1.
varStandardSpectrum[badSpectrum] = varStandardSpectrum[True - badSpectrum].max() * 1e9
mod = background + standardSpectrum * splineProfile(x) * (splineProfile(x).sum() / splineProfile(x[extractionAperture]).sum())
anyNewBadPixels = True
nbp0 = goodPixels.size - goodPixels.sum()
gp0 = goodPixels.copy()
while anyNewBadPixels is True:
for ii in range(npix):
if verbose>1:
print "*****In fitProfileSlices, pixel %i/%i" % (ii+1, npix)
if ii>0: # take the median of the last several guesses
gindex = [max(0,ii-2), min(npix, ii+2)]
guess = np.median(fitparam[gindex[0]:gindex[1]], 0)
else:
guess = None
if verbose>1:
message("ii>>%i"%ii)
message("goodPixels>>"+str(goodPixels.astype(float)[ii,:]))
message("stdProfile>>"+str(stdProfile[ii,:]))
thisfit = fitprofile(profileStack[ii,extractionAperture], splineProfile, (goodPixels.astype(float)/stdProfile**2)[ii,extractionAperture],verbose=verbose, guess=guess, retall=True)
bestfit = thisfit[0]
bestchi = thisfit[1]
if verbose>1:
print "this fit: ",bestfit
fitparam[ii,:] = bestfit
fitprofiles[ii,:] = makeFitModel(bestfit, profileStack[ii,extractionAperture],splineProfile)
if verbose>1:
print "finished pixel %i/%i" % (ii+1,npix)
deviations = (fitprofiles - profileStack[:, extractionAperture])/stdProfile[:, extractionAperture]
for kk in range(nextract):
kk0 = extractionAperture.nonzero()[0][kk]
thisdevfit = an.polyfitr(x2, deviations[:, kk], 1, 3)
theseoutliers = np.abs((deviations[:, kk] - np.polyval(thisdevfit, x2))/an.stdr(deviations, 3)) > 5
goodPixels[theseoutliers, kk0] = False
nbp = goodPixels.size - goodPixels.sum()
if nbp <= nbp0:
anyNewBadPixels = False
else:
nbp0 = nbp
for ii in range(fitparam.shape[1]):
fitparam[:,ii] = bfixpix(fitparam[:,ii], goodPixels[:,extractionAperture].sum(1) <= (nextract/2.), retdat=True)
return fitparam, fitprofiles
[docs]
def extractSpectralProfiles(args, **kw):
"""
Extract spectrum
:INPUTS:
args : tuple or list
either a tuple of (splineProfile, profileStack, errorStack,
profileMask), or a list of such tuples (from makeprofile).
:OPTIONS:
bkg_radii : 2-sequence
inner and outer radii to use in computing background
extract_radius : int
radius to use for both flux normalization and extraction
:RETURNS:
3-tuple:
[0] -- spectrum flux (in electrons)
[1] -- background flux
[2] -- best-fit pixel shift
:EXAMPLE:
::
out = spec.traceorders('aug16s0399.fits',nord=7)
out2 = spec.makeprofile('aug16s0399.fits',out,retall=True)
out3 = spec.extractSpectralProfiles(out2)
:SEE_ALSO:
:func:`optimalExtract`
:NOTES:
Note that this is non-optimal with highly tilted or curved
spectra, for the reasons described by Marsh (1989) and Mukai
(1990).
"""
# 2010-08-24 21:24 IJC: Updated comments
if kw.has_key('verbose'):
verbose = kw['verbose']
else:
verbose = False
if kw.has_key('bkg_radii'):
bkg_radii = kw['bkg_radii']
else:
bkg_radii = [15, 20]
if verbose: message("Setting option 'bkg_radii' to: " + str(bkg_radii))
if kw.has_key('extract_radius'):
extract_radius = kw['extract_radius']
else:
extract_radius = 10
#print 'starting...', args[0].__class__, isinstance(args[0],tuple)#, len(*arg)
fitprofiles = []
if isinstance(args,list): # recurse with each individual set of arguments
#nord = len(arg[0])
spectrum = []
background = []
pixshift = []
for ii, thesearg in enumerate(args):
if verbose: print "----------- Iteration %i/%i" % (ii+1, len(args))
#print 'recursion loop', thesearg.__class__, verbose, kw
tempout = extractSpectralProfiles(thesearg, **kw)
spectrum.append(tempout[0])
background.append(tempout[1])
pixshift.append(tempout[2])
fitprofiles.append(tempout[3])
spectrum = np.array(spectrum)
background = np.array(background)
pixshift = np.array(pixshift)
else:
#print 'actual computation', args.__class__, verbose, kw
splineProfile = args[0]
profileStack = args[1]
errorStack = args[2]
profileMask = args[3]
#print splineProfile.__class__, profileStack.__class__, errorStack.__class__, profileMask.__class__
#print profileStack.shape, errorStack.shape, profileMask.shape
fps = dict()
if kw.has_key('verbose'): fps['verbose'] = kw['verbose']
if kw.has_key('bkg_radii'): fps['bkg_radii'] = kw['bkg_radii']
if kw.has_key('extract_radius'): fps['extract_radius'] = kw['extract_radius']
fitparam, fitprofile = fitProfileSlices(splineProfile, profileStack, errorStack, profileMask, **fps)
spectrum = fitparam[:,0]
background = fitparam[:,1]
pixshift = fitparam[:,2]
fitprofiles.append(fitprofile)
ret = (spectrum, background, pixshift, fitprofiles)
return ret
[docs]
def gaussint(x):
"""
:PURPOSE:
Compute the integral from -inf to x of the normalized Gaussian
:INPUTS:
x : scalar
upper limit of integration
:NOTES:
Designed to copy the IDL function of the same name.
"""
# 2011-10-07 15:41 IJMC: Created
from scipy.special import erf
scalefactor = 1./np.sqrt(2)
return 0.5 + 0.5 * erf(x * scalefactor)
[docs]
def slittrans(*varargin):
"""+
:NAME:
slittrans
:PURPOSE:
Compute flux passing through a slit assuming a gaussian PSF.
:CATEGORY:
Spectroscopy
:CALLING SEQUENCE:
result = slittrans(width,height,fwhm,xoffset,yoffset,CANCEL=cancel)
:INPUTS:
width - Width of slit.
height - Height of slit.
fwhm - Full-width at half-maximum of the gaussian image.
xoffset - Offset in x of the image from the center of the slit.
yoffset - Offset in y of the image from the center of the slit.
Note: the units are arbitrary but they must be the same across
all of the input quantities.
KEYWORD PARAMETERS:
CANCEL - Set on return if there is a problem
OUTPUTS:
Returned is the fraction of the total gaussian included in the slit.
EXAMPLE:
result = slittrans(0.3,15,0.6,0,0)
Computes the fraction of the flux transmitted through a slit
0.3x15 arcseconds with a PSF of 0.6 arcseconds FWHM. The PSF is
centered on the slit.
MODIFICATION HISTORY:
Based on M Buie program, 1991 Mar., Marc W. Buie, Lowell Observatory
Modified 2000 Apr., M. Cushing to include y offsets.
2011-10-07 15:45 IJMC: Converted to Python
2011-11-14 16:29 IJMC: Rewrote to use :func:`erf` rather than
:func:`gaussint`
"""
#function slittrans,width,height,fwhm,xoffset,yoffset,CANCEL=cancel
from scipy.special import erf
cancel = 0
n_params = len(varargin)
if n_params <> 5:
print 'Syntax - result = slittrans(width,height,fwhm,xoffset,yoffset,',
print ' CANCEL=cancel)'
cancel = 1
return cancel
width, height, fwhm, xoffset, yoffset = varargin[0:5]
#cancel = cpar('slittrans',width,1,'Width',[2,3,4,5],0)
#if cancel then return,-1
#cancel = cpar('slittrans',height,2,'Height',[2,3,4,5],0)
#if cancel then return,-1
#cancel = cpar('slittrans',fwhm,3,'FWHM',[2,3,4,5],0)
#if cancel then return,-1
#cancel = cpar('slittrans',xoffset,4,'Xoffset',[2,3,4,5],0)
#if cancel then return,-1
#cancel = cpar('slittrans',yoffset,5,'Yoffset',[2,3,4,5],0)
#if cancel then return,-1
# Go ahead
a = 0.5 * width
b = 0.5 * height
#s = fwhm/(np.sqrt(8.0*np.log(2.0)))
#slit = ( 1.0 - gaussint( -(a+xoffset)/s ) - gaussint( -(a-xoffset)/s ) ) * \
# ( 1.0 - gaussint( -(b+yoffset)/s ) - gaussint( -(b-yoffset)/s ) )
invs2 = (np.sqrt(4.0*np.log(2.0)))/fwhm # sigma * sqrt(2)
slit4 = 0.25 * ( (erf( -(a+xoffset)*invs2) + erf( -(a-xoffset)*invs2) ) ) * \
( (erf( -(b+yoffset)*invs2) + erf( -(b-yoffset)*invs2) ) )
#print 'smin/max,',slit.min(), slit5.max()
#print 'smin/max,',slit4.min(), slit4.max()
return slit4
[docs]
def atmosdisp(wave, wave_0, za, pressure, temp, water=2., fco2=0.0004, obsalt=0.):
""":NAME:
atmosdisp
PURPOSE:
Compute the atmosperic dispersion relative to lambda_0.
CATEGORY:
Spectroscopy
CALLING SEQUENCE:
result = atmosdisp(wave,wave_0,za,pressure,temp,[water],[obsalt],$
CANCEL=cancel)
INPUTS:
wave - wavelength in microns
wave_0 - reference wavelength in microns
za - zenith angle of object [in degrees]
pressure - atmospheric pressure in mm of Hg
temp - atmospheric temperature in degrees C
OPTIONAL INPUTS:
water - water vapor pressure in mm of Hg.
fco2 - relative concentration of CO2 (by pressure)
obsalt - The observatory altitude in km.
KEYWORD PARAMETERS:
CANCEL - Set on return if there is a problem
OUTPUTS:
Returns the atmospheric disperion in arcseconds.
PROCEDURE:
Computes the difference between the dispersion at two
wavelengths. The dispersion for each wavelength is derived from
Section 4.3 of Green's "Spherical Astronomy" (1985).
EXAMPLE:
MODIFICATION HISTORY:
2000-04-05 - written by M. Cushing, Institute for Astronomy, UH
2002-07-26 - cleaned up a bit.
2003-10-20 - modified formula - WDV
2011-10-07 15:51 IJMC: Converted to Python, with some unit conversions
-"""
#function atmosdisp,wave,wave_0,za,pressure,temp,water,obsalt,CANCEL=cancel
from nsdata import nAir
# Constants
mmHg2pa = 101325./760. # Pascals per Torr (i.e., per mm Hg)
rearth = 6378.136e6 #6371.03 # mean radius of earth in km [Allen's]
hconst = 2.926554e-2 # R/(mu*g) in km/deg K, R=gas const=8.3143e7
# mu=mean mol wght. of atm=28.970, g=980.665
tempk = temp + 273.15
pressure_pa = pressure * mmHg2pa
water_pp = water/pressure # Partial pressure
hratio = (hconst * tempk)/(rearth + obsalt)
# Compute index of refraction
nindx = nAir(wave,P=pressure_pa,T=tempk,pph2o=water_pp, fco2=fco2)
nindx0 = nAir(wave_0,P=pressure_pa,T=tempk,pph2o=water_pp, fco2=fco2)
# Compute dispersion
acoef = (1. - hratio)*(nindx - nindx0)
bcoef = 0.5*(nindx*nindx - nindx0*nindx0) - (1. + hratio)*(nindx - nindx0)
tanz = np.tan(np.deg2rad(za))
disp = 206265.*tanz*(acoef + bcoef*tanz*tanz)
#print nindx
#print nindx0
#print acoef
#print bcoef
#print tanz
#print disp
return disp
[docs]
def parangle(HA, DEC, lat):
"""
+
NAME:
parangle
PURPOSE:
To compute the parallactic angle at a given position on the sky.
CATEGORY:
Spectroscopy
CALLING SEQUENCE:
eta, za = parangle(HA, DEC, lat)
INPUTS:
HA - Hour angle of the object, in decimal hours (0,24)
DEC - Declination of the object, in degrees
lat - The latitude of the observer, in degrees
KEYWORD PARAMETERS:
CANCEL - Set on return if there is a problem
OUTPUTS:
eta - The parallactic angle
za - The zenith angle
PROCEDURE:
Given an objects HA and DEC and the observers latitude, the
zenith angle and azimuth are computed. The law of cosines
then gives the parallactic angle.
EXAMPLE:
NA
MODIFICATION HISTORY:
2000-04-05 - written by M. Cushing, Institute for Astronomy,UH
2002-08-15 - cleaned up a bit.
2003-10-21 - changed to pro; outputs zenith angle as well - WDV
2011-10-07 17:58 IJMC: Converted to Python
-"""
#pro parangle, HA, DEC, lat, eta, za, CANCEL=cancel
cancel = 0
d2r = np.deg2rad(1.)
r2d = np.rad2deg(1.)
# If HA equals zero then it is easy.
HA = HA % 24
# Check to see if HA is greater than 12.
if hasattr(HA, '__iter__'):
HA = np.array(HA, copy=False)
HAind = HA > 12
if HAind.any():
HA[HAind] = 24. - HA[HAind]
else:
if HA>12.:
HA = 24. - HA
HA = HA*15.
# Determine Zenith angle and Azimuth
cos_za = np.sin(lat*d2r) * np.sin(DEC*d2r) + \
np.cos(lat*d2r) * np.cos(DEC*d2r) * np.cos(HA*d2r)
za = np.arccos(cos_za) * r2d
cos_az = (np.sin(DEC*d2r) - np.sin(lat*d2r)*np.cos(za*d2r)) / \
(np.cos(lat*d2r) * np.sin(za*d2r))
az = np.arccos(cos_az)*r2d
if hasattr(az, '__iter__'):
azind = az==0
if azind.any() and DEC<lat:
az[azind] = 180.
else:
if az==0. and DEC<lat:
az = 180.
tan_eta = np.sin(HA*d2r)*np.cos(lat*d2r) / \
(np.cos(DEC*d2r)*np.sin(lat*d2r) - \
np.sin(DEC*d2r)*np.cos(lat*d2r)*np.cos(HA*d2r))
eta = np.arctan(tan_eta)*r2d
if hasattr(eta, '__iter__'):
etaind = eta < 0
ezind = (eta==0) * (az==0)
zaind = za > 90
if etaind.any():
eta[etaind] += 180.
elif ezind.any():
eta[ezind] = 180.
if zaind.any():
eta[zaind] = np.nan
else:
if eta < 0:
eta += 180.
elif eta==0 and az==0:
eta = 180.
if za>90:
eta = np.nan
HA = HA/15.0
return eta, za
[docs]
def lightloss(objfile, wguide, seeing, press=None, water=None, temp=None, fco2=None, wobj=None, dx=0, dy=0, retall=False):
""" +
NAME:
lightloss
PURPOSE:
To determine the slit losses from a spectrum.
CATEGORY:
Spectroscopy
CALLING SEQUENCE:
### TBD lightloss, obj, std, wguide, seeing, out, CANCEL=cancel
INPUTS:
obj - FITS file of the object spectrum
wguide - wavelength at which guiding was done
seeing - seeing FWHM at the guiding wavelength
OPTIONAL INPUTS:
press - mm Hg typical value (615 for IRTF, unless set)
water - mm Hg typical value (2 for IRTF, unless set)
temp - deg C typical value (0 for IRTF, unless set)
fco2 - relative concentration of CO2 (0.004, unless set)
wobj - wavelength scale for data
dx - horizontal offset of star from slit center
dy - vertical offset of star from slit center
retall- whether to return much diagnostic info, or just lightloss.
NOTES:
'seeing', 'dx', and 'dy' should all be in the same units, and
also the same units used to define the slit dimensions in the
obj FITS file header
KEYWORD PARAMETERS:
CANCEL - Set on return if there is a problem
OUTPUTS:
array : fractional slit loss at each wavelength value
OR:
tuple of arrays: (slitloss, disp_obj, diff, fwhm, dx_obj, dy_obj)
PROCEDURE:
Reads a Spextool FITS file.
EXAMPLE:
None
REQUIREMENTS:
:doc:`phot`
:doc:`pyfits`
MODIFICATION HISTORY:
2003-10-21 - Written by W D Vacca
2011-10-07 20:19 IJMC: Converted to Python, adapted for single objects
2011-10-14 14:01 IJMC: Added check for Prism mode (has
different slit dimension keywords) and different
pyfits header read mode.
2011-11-07 15:53 IJMC: Added 'retall' keyword
-
"""
#pro lightloss, objfile, stdfile, wguide, seeing, outfile,CANCEL=cancel
try:
from astropy.io import fits as pyfits
except:
import pyfits
from phot import hms, dms
r2d = np.rad2deg(1)
# --- Open input files
obj = pyfits.getdata(objfile)
objhdu = pyfits.open(objfile)
#objhdr = pyfits.getheader(objfile)
objhdu.verify('silentfix')
objhdr = objhdu[0].header
if wobj is None:
try:
wobj = obj[:,0,:]
#fobj = obj[:,1,:]
#eobj = obj[:,2,:]
except:
wobj = obj[0,:]
#fobj = obj[1,:]
#eobj = obj[2,:]
# --- Read header keywords
tele = objhdr['TELESCOP']
try:
slitwd = objhdr['SLTW_ARC']
slitht = objhdr['SLTH_ARC']
except: # for SpeX prism mode:
slitwd, slitht = map(float, objhdr['SLIT'].split('x'))
#xaxis = objhdr['XAXIS']
#xunits = objhdr['XUNITS']
#yaxis = objhdr['YAXIS']
#yunits = objhdr['YUNITS']
posang_obj = objhdr['POSANGLE']
HA_objstr = objhdr['HA']
DEC_objstr = objhdr['DEC']
# --- Process keywords
#coord_str = HA_objstr + ' ' + DEC_objstr
#get_coords, coords, InString=coord_str
HA_obj, DEC_obj = dms(HA_objstr), dms(DEC_objstr)
if posang_obj<0.0:
posang_obj += 180.
if posang_obj >= 180.0:
posang_obj -= 180.0
if tele=='NASA IRTF':
obsdeg = 19.0
obsmin = 49.0
obssec = 34.39
obsalt = 4.16807 # observatory altitude in km
teldiam = 3.0 # diameter of primary in meters
if press is None:
press = 615.0 # mm Hg typical value
if water is None:
water = 2.0 # mm Hg typical value
if temp is None:
temp = 0.0 # deg C typical value
else:
print 'Unknown Telescope - stopping!'
return
if fco2 is None:
fco2 = 0.0004
obslat = dms('%+02i:%02i:%02i' % (obsdeg, obsmin, obssec))
# --- Compute Parallactic Angle
pa_obj, za_obj = parangle(HA_obj, DEC_obj, obslat)
dtheta_obj = posang_obj - pa_obj
#print posang_obj, pa_obj, dtheta_obj, za_obj
#print posang_obj, pa_obj, dtheta_obj, HA_obj
# --- Compute Differential Atmospheric Dispersion
disp_obj = atmosdisp(wobj, wguide, za_obj, press, temp, \
water=water, obsalt=obsalt, fco2=fco2)
# --- Compute FWHM at each input wavelength
diff = 2.0*1.22e-6*3600.0*r2d*(wobj/teldiam) # arcsec
fwhm = (seeing*(wobj/wguide)**(-0.2))
fwhm[fwhm < diff] = diff[fwhm < diff]
# --- Compute Relative Fraction of Light contained within the slit
dx_obj = (disp_obj*np.sin(dtheta_obj/r2d)) + dx
dy_obj = (disp_obj*np.cos(dtheta_obj/r2d)) + dy
slitloss = slittrans(slitwd,slitht,fwhm,dx_obj,dy_obj)
#debug_check = lightloss2(wobj, slitwd, slitht, posang_obj/57.3, pa_obj/57.3, za_obj/57.3, wguide, seeing, retall=retall)
if retall:
return (slitloss, disp_obj, diff, fwhm, dx_obj, dy_obj)
else:
return slitloss
[docs]
def lightloss2(wobj, slitwd, slitht, slitPA, targetPA, zenith_angle, wguide, seeing, press=615., water=2., temp=0., fco2=0.004, obsalt=4.16807, teldiam=3., dx=0, dy=0, retall=False, ydisp=None, xdisp=None, fwhm=None):
""" +
NAME:
lightloss2
PURPOSE:
To determine the slit losses from an observation (no FITS file involved)
CATEGORY:
Spectroscopy
CALLING SEQUENCE:
### TBD lightloss, obj, std, wguide, seeing, out, CANCEL=cancel
INPUTS:
wobj - wavelength scale for data
slitwd - width of slit, in arcsec
slitht - height of slit, in arcsec
slitPA - slit Position Angle, in radians
targetPA - Parallactic Angle at target, in radians
zenith_angle - Zenith Angle, in radians
wguide - wavelength at which guiding was done
seeing - seeing FWHM at the guiding wavelength
OPTIONAL INPUTS:
press - mm Hg typical value (615, unless set)
water - mm Hg typical value (2 , unless set)
temp - deg C typical value (0 , unless set)
fco2 - relative concentration of CO2 (0.004, unless set)
obsalt- observatory altitude, in km
teldiam- observatory limiting aperture diameter, in m
dx - horizontal offset of star from slit center
dy - vertical offset of star from slit center
retall- whether to return much diagnostic info, or just lightloss.
ydisp - The position of the spectrum in the slit at all
values of wobj. This should be an array of the same
size as wobj, with zero corresponding to the vertical
middle of the slit and positive values tending toward
zenith. In this case xdisp will be computed as XXXX
rather than from the calculated atmospheric
dispersion; dx and dy will also be ignored.
fwhm - Full-width at half-maximum of the spectral trace, at
all values of wobj. This should be an array of the
same size as wobj, measured in arc seconds.
NOTES:
'slitwidth', 'slitheight', 'seeing', 'dx', 'dy', and 'fwhm'
(if used) should all be in the same units: arc seconds.
OUTPUTS:
array : fractional slit loss at each wavelength value
OR:
tuple of arrays: (slitloss, disp_obj, diff, fwhm, dx_obj, dy_obj)
PROCEDURE:
All input-driven. For the SpeXTool-version analogue, see
:func:`lightloss`
EXAMPLE:
import numpy as np
import spec
w = np.linspace(.5, 2.5, 100) # Wavelength, in microns
d2r = np.deg2rad(1.)
#targetPA, za = spec.parangle(1.827, 29.67*d2r, lat=20.*d2r)
targetPA, za = 105.3, 27.4
slitPA = 90. * d2r
spec.lightloss2(w, 3., 15., slitPA, targetPA*d2r, za*d2r, 2.2, 1.0)
REQUIREMENTS:
:doc:`phot`
MODIFICATION HISTORY:
2003-10-21 - Written by W D Vacca
2011-10-07 20:19 IJMC: Converted to Python, adapted for single objects
2011-10-14 14:01 IJMC: Added check for Prism mode (has
different slit dimension keywords) and different
pyfits header read mode.
2011-11-07 15:53 IJMC: Added 'retall' keyword
2011-11-07 21:17 IJMC: Cannibalized from SpeXTool version
2011-11-25 15:06 IJMC: Added ydisp and fwhm options.
-
"""
#pro lightloss, objfile, stdfile, wguide, seeing, outfile,CANCEL=cancel
try:
from astropy.io import fits as pyfits
except:
import pyfits
from phot import hms, dms
r2d = np.rad2deg(1)
d2r = np.deg2rad(1.)
if slitPA<0.0:
slitPA += np.pi
if slitPA >= np.pi:
slitPA -= np.pi
# --- Compute Parallactic Angle
dtheta_obj = slitPA - targetPA
#print slitPA, targetPA, dtheta_obj, zenith_angle
# --- Compute FWHM at each input wavelength
diff = 2.0*1.22e-6*3600.0*r2d*(wobj/teldiam) # arcsec
if fwhm is None:
fwhm = (seeing*(wobj/wguide)**(-0.2))
fwhm[fwhm < diff] = diff[fwhm < diff]
if ydisp is None or xdisp is None:
# --- Compute Differential Atmospheric Dispersion
disp_obj = atmosdisp(wobj, wguide, zenith_angle*r2d, press, temp, \
water=water, obsalt=obsalt, fco2=fco2)
if ydisp is None:
dy_obj = (disp_obj*np.cos(dtheta_obj)) + dy
else:
dy_obj = ydisp
if xdisp is None:
dx_obj = (disp_obj*np.sin(dtheta_obj)) + dx
else:
dx_obj = xdisp
else:
dx_obj = np.array(xdisp, copy=False)
dy_obj = np.array(ydisp, copy=False)
if retall:
disp_obj = (dy_obj - dy) / np.cos(dtheta_obj)
# if xdisp is None and ydisp is None:
# guide_index = np.abs(wobj - wguide).min() == np.abs(wobj-wguide)
# dy = ydisp[guide_index].mean()
# dy_obj = ydisp
# dx_obj = (dy_obj - dy) * np.tan(dtheta_obj)
# --- Compute Relative Fraction of Light contained within the slit
slitloss = slittrans(slitwd, slitht, fwhm, dx_obj, dy_obj)
if retall:
return slitloss, disp_obj, diff, fwhm, dx_obj, dy_obj
else:
return slitloss
[docs]
def humidpressure(RH, T):
""" +
NAME:
humidpressure
PURPOSE:
To convert relative humidity into a H2O vapor partial pressure
CATEGORY:
Spectroscopy
CALLING SEQUENCE:
humidpressure(RH, 273.15)
INPUTS:
RH - relative humidity, in percent
T - temperature, in Kelvin
OUTPUTS:
h2o_pp : water vapor partial pressure, in Pascals
PROCEDURE:
As outlined in Butler (1998): "Precipitable Water at KP", MMA
Memo 238 (which refers in turn to Liebe 1989, "MPM - An
Atmospheric Millimeter-Wave Propagation Model"). Liebe
claims that this relation has an error of <0.2% from -40 C to
+40 C.
EXAMPLE:
None
MODIFICATION HISTORY:
2011-10-08 17:08 IJMC: Created.
-
"""
# units of Pa
#theta = 300./T
#return 2.408e11 * RH * np.exp(-22.64*theta) * (theta*theta*theta*theta)
theta_mod = 6792./T # theta * 22.64
return 9.1638 * RH * np.exp(-theta_mod) * \
(theta_mod*theta_mod*theta_mod*theta_mod)
[docs]
def runlblrtm(lamrange, pwv=2., zang=0., alt=4.2, co2=390, res=200, dotrans=True, dorad=False,
pwv_offset=4.,
verbose=False, _save='/Users/ianc/temp/testatmo.mat',
_wd='/Users/ianc/proj/atmo/aerlbl_v12.0_package/caltech_wrapper_v02/',
scriptname='runtelluric.m',
command = '/Applications/Octave.app/Contents/Resources/bin/octave'):
"""
Run LBLRTM to compute atmospheric transmittance and/or radiance.
:INPUTS:
lamrange : 2-sequence
approximate minimum and maximum wavelengths
:OPTIONS:
pwv : float
mm of Precipitable Water Vapor above observation site. If
negative, then the value abs(pwv_offset-pwv) will be used instead.
pwv_offset : float
Only used if (pwv < 0); see above for description.
zang : float
observation angle from zenith, in degrees
alt : float
Observation elevation, in km.
co2 : float
CO2 concentration in ppm by volume. Concentration is assumed
to be uniform throughout the atmosphere.
res : float
approximate spectral resolution desired
dotrans : bool
compute atmospheric transmittance
dorad : bool
compute atmospheric radiance. NOT CURRENTLY WORKING
_save : str
path where temporary MAT save file will be stored
_wd : str
path where MATLAB wrapper scripts for LBLRTM are located
scriptname : str
filename for temporary MATLAB/OCTAVE script (saved after exit)
command : str
path to MATLAB/OCTAVE executable
:OUTPUTS:
A 2- or 3-tuple: First element is wavelength in microns, second
element is transmittance (if requested). Radiance will (if
requested) always be the last element, and in f_nu units: W/cm2/sr/(cm^-1)
:REQUIREMENTS:
SciPy
OCTAVE or MATLAB
`LBLRTM <http://rtweb.aer.com/lblrtm_code.html>`_
D. Feldman's set of MATLAB `wrapper scripts
<http://www.mathworks.com/matlabcentral/fileexchange/6461-lblrtm-wrapper-version-0-2/>`_
"""
# 2011-10-13 13:59 IJMC: Created.
# 2011-10-19 23:35 IJMC: Added scipy.__version__ check for MAT IO
# 2011-10-25 17:00 IJMC: Added pwv_offset option.
# 2011-11-07 08:47 IJMC: Now path(path..) uses _wd input option; call --wd option.
# 2012-07-20 21:25 IJMC: Added 'alt' option for altitude.
# 2012-09-16 15:13 IJMC: Fixed for 'dotrans' and 'dorad'
# 2014-02-17 18:48 IJMC: Specified approximate units of radiance output.
import os
from scipy.io import loadmat
from scipy import __version__
# Define variables:
def beforev8(ver):
v1, v2, v3 = ver.split('.')
if v1==0 and v2<8:
before = True
else:
before = False
return before
if pwv<0:
pwv = np.abs(pwv_offset - pwv)
#lamrange = [1.12, 2.55] # microns
#res = 200; # lambda/dlambda
# Try to delete old files, if possible.
if os.path.isfile(_save):
#try:
# os.remove(_save)
#except:
while os.path.isfile(_save):
_save = _save.replace('.mat', '0.mat')
if os.path.isfile(scriptname):
#try:
# os.remove(scriptname)
#except:
while os.path.isfile(scriptname):
scriptname = scriptname.replace('.m', '0.m')
aerlbl_dir = '/'.join(_wd.split('/')[0:-2]) + '/'
# Make the matlab script:
matlines = []
matlines.append("_path = '%s';\n" % _wd )
matlines.append("_save = '%s';\n" % _save )
matlines.append("_dir0 = pwd;\n")
matlines.append("path(path, '%s')\n" % _wd )
matlines.append("lamrange = [%1.5f, %1.5f];\n" % (lamrange[0], lamrange[1]))
if verbose:
matlines.append("pwd\n")
matlines.append("wt = which('telluric_simulator')\n")
matlines.append("strfind(path, 'caltech')\n")
matlines.append("tran_out = telluric_simulator(lamrange, '--dotran %i', '--dorad %i', '--R %s', '--alt %1.4f', '--pwv %1.4f', '--co2 %1.1f', '--zang %1.3f', '--verbose %i', '--wd %s');\n" % (int(dotrans), int(dorad), res, alt, pwv, co2, zang, verbose, aerlbl_dir) )
matlines.append("save('-v6', _save, 'tran_out');\n")
# Write script to disk, and execute it:
f = open(scriptname, 'w')
f.writelines(matlines)
f.close()
os.system('%s %s' % (command, scriptname))
# Open the MAT file and extract the desired output:
# try:
if os.path.isfile(_save):
mat = loadmat(_save)
else:
trycount = 1
print "Saved file '%s' could not be loaded..." % _save
while trycount < 5:
os.system('%s %s' % (command, scriptname))
try:
mat = loadmat(_save)
trycount = 10
except:
trycount += 1
if trycount < 10: # never successfully loaded the file.
pdb.set_trace()
if beforev8(__version__):
w = mat['tran_out'][0][0].wavelength.ravel()
else:
w = mat['tran_out']['wavelength'][0][0].ravel()
if dotrans:
if beforev8(__version__):
t = mat['tran_out'][0][0].transmittance.ravel()
else:
t = mat['tran_out']['transmittance'][0][0].ravel()
if dorad:
if beforev8(__version__):
r = mat['tran_out'][0][0].radiance.ravel()
else:
r = mat['tran_out']['radiance'][0][0].ravel()
os.remove(scriptname)
os.remove(_save)
if dotrans:
ret = (w, t)
else:
ret = (w,)
if dorad:
ret = ret + (r,)
return ret
[docs]
def zenith(ra, dec, ha, obs):
""" Compute zenith angle (in degrees) for an observation.
:INPUTS:
ra : str
Right Ascension of target, in format: HH:MM:SS.SS
dec : str
Declination of target, in format: +ddd:mm:ss
ha : str
Hour Angle of target, in format: +HH:MM:SS.SS
obs : str
Name of observatory site (keck, irtf, lick, lapalma, ctio,
andersonmesa, mtgraham, kpno) or a 3-tuple containing
(longitude_string, latitude_string, elevation_in_meters)
:OUTPUTS:
Zenith angle, in degrees, for the specified observation
:REQUIREMENTS:
:doc:`phot`
Numpy
"""
# 2011-10-14 09:43 IJMC: Created
import phot
#observer = ephem.Observer()
if obs=='lick':
obs_long, obs_lat = '-121:38.2','37:20.6'
obs_elev = 1290
elif obs=='keck':
obs_long, obs_lat = '-155:28.7','19:49.7'
obs_elev = 4160
elif obs=='irtf':
obs_long, obs_lat = '-155:28:21.3', '19:49:34.8'
obs_elev = 4205
elif obs=='lapalma':
obs_long, obs_lat = '17:53.6','28:45.5'
obs_elev = 4160
elif obs=='ctio':
obs_long, obs_lat = '-70:48:54','-30:9.92'
obs_elev = 2215
elif obs=='andersonmesa': #
obs_long, obs_lat = '-111:32:09', '30:05:49'
obs_elev = 2163
elif obs=='mtgraham':
obs_long, obs_lat = '-109:53:23', '32:42:05'
obs_elev = 3221
elif obs=='kpno':
obs_long, obs_lat = '-111:25:48', '31:57:30'
obs_elev = 2096
elif len(obs)==3:
obs_long, obs_lat, obs_elev = obs
else:
print "Unknown or imparseable observatory site."
return -1
lat = phot.dms(obs_lat) * np.pi/180.
long = phot.dms(obs_long) * np.pi/180.
ra = (phot.hms(ra)) * np.pi/180.
dec= (phot.dms(dec)) * np.pi/180.
# Compute terms for coordinate conversion
if hasattr(ha, '__iter__'):
zang = []
for ha0 in ha:
har = (phot.hms(ha0)) * np.pi/180.
term1 = np.sin(lat)*np.sin(dec)+np.cos(lat)*np.cos(dec)*np.cos(har)
term2 = np.cos(lat)*np.sin(dec)-np.sin(lat)*np.cos(dec)*np.cos(har)
term3 = -np.cos(dec)*np.sin(har)
rad = np.abs(term3 +1j*term2)
az = np.arctan2(term3,term2)
alt = np.arctan2(term1, rad)
zang.append(90. - (alt*180./np.pi))
else:
har = (phot.hms(ha0)) * np.pi/180.
term1 = np.sin(lat)*np.sin(dec)+np.cos(lat)*np.cos(dec)*np.cos(har)
term2 = np.cos(lat)*np.sin(dec)-np.sin(lat)*np.cos(dec)*np.cos(har)
term3 = -np.cos(dec)*np.sin(har)
rad = np.abs(term3 +1j*term2)
az = np.arctan2(term3,term2)
alt = np.arctan2(term1, rad)
zang = 90. - (alt*180./np.pi)
## Compute airmass
#z = pi/2. - alt
#airmass = 1./(np.cos(z) + 0.50572*(96.07995-z*180./pi)**(-1.6364))
return zang
[docs]
def spexsxd_scatter_model(dat, halfwid=48, xlims=[470, 1024], ylims=[800, 1024], full_output=False, itime=None):
"""Model the scattered light seen in SpeX/SXD K-band frames.
:INPUTS:
dat : str or numpy array
filename of raw SXD frame to be corrected, or a Numpy array
containing its data.
:OPTIONS:
halfwid : int
half-width of the spectral orders. Experience shows this is
approximately 48 pixels. This value is not fit!
xlims : list of length 2
minimum and maximum x-pixel values to use in the fitting
ylims : list of length 2
minimum and maximum y-pixel values to use in the fitting
full_output : bool
whether to output only model, or the tuple (model, fits, chisq, nbad)
itime : float
integration time, in seconds, with which to scale the initial
guesses
:OUTPUT:
scatter_model : numpy array
Model of the scattered light component, for subtraction or saving.
OR:
scatter_model, fits, chis, nbad
:REQUIREMENTS:
:doc:`pyfits`, :doc:`numpy`, :doc:`fit_atmo`, :doc:`analysis`, :doc:`phasecurves`
:TO_DO_LIST:
I could stand to be more clever in modeling the scattered light
components -- perhaps fitting for the width, or at least
allowing the width to be non-integer.
"""
# 2011-11-10 11:10 IJMC: Created
import analysis as an
import phasecurves as pc
try:
from astropy.io import fits as pyfits
except:
import pyfits
############################################################
# Define some helper functions:
############################################################
def tophat(param, x):
"""Grey-pixel tophat function with set width
param: [cen_pix, amplitude, background]
x : must be array of ints, arange(0, size-1)
returns the model."""
# 2011-11-09 21:37 IJMC: Created
intpix, fracpix = int(param[0]), param[0] % 1
th = param[1] * ((-halfwid <= (x - intpix)) * ((x - intpix) < halfwid))
# th = * th.astype(float)
if (intpix >= halfwid) and ((intpix - halfwid) < x.size):
th[intpix - halfwid] = param[1]*(1. - fracpix)
if (intpix < (x.size - halfwid)) and ((intpix + halfwid) >= 0):
th[intpix + halfwid] = param[1]*fracpix
return th + param[2]
def tophat2g(param, x, p0prior=None):
"""Grey-pixel double-tophat plus gaussian
param: [cen_pix1, amplitude1, cen_pix2, amplitude2, g_area, g_sigma, g_center, background]
x : must be ints, arange(0, size-1)
returns the model.""" # 2011-11-09 21:37 IJMC: Created
#th12 =
#th2 =
#gauss =
# if p0prior is not None:
# penalty =
return tophat([param[0], param[1], 0], x) + \
tophat([param[2], param[3], 0], x) + \
gaussian(param[4:7], x) + param[7]
############################################################
# Parse inputs
############################################################
halfwid = int(halfwid)
if isinstance(dat, np.ndarray):
if itime is None:
itime = 1.
else:
if itime is None:
try:
itime = pyfits.getval(dat, 'ITIME')
except:
itime = 1.
dat = pyfits.getdata(dat)
nx, ny = dat.shape
scatter_model = np.zeros((nx, ny), dtype=float)
chis, fits, nbad = [], [], []
iivals = np.arange(xlims[1]-1, xlims[0], -1, dtype=int)
position_offset = 850 - ylims[0]
est_coefs = np.array([ -5.02509772e-05, 2.97212397e-01, -7.65702234e+01])
estimated_position = np.polyval(est_coefs, iivals) + position_offset
estimated_error = 0.5
# to hold scattered light position fixed, rather than fitting for
# that position, uncomment the following line:
#holdfixed = [0]
holdfixed = None
############################################################
# Start fitting
############################################################
for jj, ii in enumerate(iivals):
col = dat[ylims[0]:ylims[1], ii]
ecol = np.ones(col.size, dtype=float)
x = np.arange(col.size, dtype=float)
if len(fits)==0:
all_guess = [175 + position_offset, 7*itime, \
70 + position_offset, 7*itime, \
250*itime, 5, 89 + position_offset, 50]
else:
all_guess = fits[-1]
all_guess[0] = estimated_position[jj]
model_all = tophat2g(all_guess, x)
res = (model_all - col)
badpix = np.abs(res) > (4*an.stdr(res, nsigma=4))
ecol[badpix] += 1e9
fit = an.fmin(pc.errfunc, all_guess, args=(tophat2g, x, col, 1./ecol**2), full_output=True, maxiter=1e4, maxfun=1e4, disp=False, kw=dict(testfinite=False), holdfixed=holdfixed)
best_params = fit[0].copy()
res = tophat2g(best_params, x) - col
badpix = np.abs(res) > (4*an.stdr(res, nsigma=4))
badpix[((np.abs(np.abs(x - best_params[0]) - 48.)) < 2) + \
((np.abs(np.abs(x - best_params[2]) - 48.)) < 2)] = False
badpix += (np.abs(res) > (20*an.stdr(res, nsigma=4)))
ecol = np.ones(col.size, dtype=float)
ecol[badpix] += 1e9
best_chisq = pc.errfunc(best_params, tophat2g, x, col, 1./ecol**2)
# Make sure you didn't converge on the wrong model:
for this_offset in ([-2, 0, 2]):
this_guess = fit[0].copy()
this_guess[2] += this_offset
this_guess[0] = estimated_position[jj]
#pc.errfunc(this_guess, tophat2g, x, col, 1./ecol**2)
this_fit = an.fmin(pc.errfunc, this_guess, args=(tophat2g, x, col, 1./ecol**2), full_output=True, maxiter=1e4, maxfun=1e4, disp=False, kw=dict(testfinite=False), holdfixed=holdfixed)
#print this_offset1, this_offset2, this_fit[1]
if this_fit[1] < best_chisq:
best_chisq = this_fit[1]
best_params = this_fit[0].copy()
fits.append(best_params)
chis.append(best_chisq)
nbad.append(badpix.sum())
mod2 = tophat2g(best_params, x)
scatter_model[ylims[0]:ylims[1], ii] = tophat(list(best_params[0:2])+[0], x)
if full_output:
return scatter_model, fits, chis, nbad
else:
return scatter_model
[docs]
def spexsxd_scatter_fix(fn1, fn2, **kw):
""" Fix scattered light in SpeX/SXD K-band and write a new file.
:INPUTS:
fn1 : str
file to be fixed
fn2 : str
new filename of fixed file.
:OPTIONS:
clobber : bool
whether to overwrite existing FITS files
Other options will be passed to :func:`spexsxd_scatter_model`
:OUTPUTS:
status : int
0 if a problem, 1 if everything is Okay
"""
# 2011-11-10 13:34 IJMC: Created
try:
from astropy.io import fits as pyfits
except:
import pyfits
if kw.has_key('clobber'):
clobber = kw.pop('clobber')
else:
clobber = False
try:
dat0 = pyfits.getdata(fn1)
hdr0 = pyfits.getheader(fn1)
except:
print "Could not read one of data or header from %s" % fn1
return 0
try:
hdr0.update('SCAT_FIX', 1, 'SXD Scattered light fixed by spexsxd_scatter_fix()')
except:
print "Could not update header properly."
return 0
try:
if hdr0.has_key('ITIME') and not kw.has_key('itime'):
kw['itime'] = hdr0['ITIME']
mod = spexsxd_scatter_model(dat0, **kw)
except:
print "Could not not model scattered light with spexsxd_scatter_model()"
return 0
try:
if not isinstance(mod, np.ndarray):
mod = mod[0]
pyfits.writeto(fn2, dat0 - mod, header=hdr0, clobber=clobber, output_verify='ignore')
except:
print "Could not write updated file to %s; clobber is %s" % (fn2, clobber)
return 0
return 1
[docs]
def tophat2(param, x):
"""Grey-pixel tophat function with set width
param: [cen_pix, amplitude, background]
newparam: [amplitude, full width, cen_pix, background]
x : must be array of ints, arange(0, size-1)
returns the model."""
# 2011-11-09 21:37 IJMC: Created
import analysis as an
oversamp = 100.
x2 = np.linspace(x[0], x[-1], (x.size - 1)*oversamp + 1)
halfwid = np.round(param[1] / 2.).astype(int)
halfwid = param[1] / 2.
intpix, fracpix = int(param[2]), param[2] % 1
intwid, fracwid = int(halfwid), halfwid % 1
th2 = 0.0 + param[0] * ((-halfwid <= (x2 - param[2])) * ((x2 - param[2]) < halfwid))
th = an.binarray(np.concatenate((th2, np.zeros(oversamp-1))), oversamp)
return th + param[3]
[docs]
def tophat_alt(param, x):
"""Standard tophat function (alternative version).
:INPUTS:
p : sequence
p[0] -- Amplitude
p[1] -- full width dispersion
p[2] -- central offset (mean location)
p[3] -- vertical offset (OPTIONAL)
x : scalar or sequence
values at which to evaluate function
:OUTPUTS:
y : scalar or sequence
1.0 where |x| < 0.5, 0.5 where |x| = 0.5, 0.0 otherwise.
"""
# 2012-04-11 12:50 IJMC: Created
if len(param)==3:
vertical_offset = 0.
elif len(param)>3:
vertical_offset = param[3]
else:
print "Input `param` to function `tophat` requires length > 3."
return -1
amplitude, width, center = param[0:3]
absx = np.abs((x - center) )
if hasattr(x, '__iter__'):
ret = np.zeros(x.shape, float)
#ind1 = absx < (0.5*width)
#ret[ind1] = 1.0
i0 = np.searchsorted(x, center-width/2)
i1 = np.searchsorted(x, center+width/2)
ret[i0:i1] = 1.
#ind2 = absx ==(0.5*width)
#ret[ind2] = 0.5
else:
if absx < 0.5:
ret = 1.0
elif absx==0.5:
ret = 0.5
else:
ret = 0.
return amplitude * ret + vertical_offset
[docs]
def model_resel(param, x):
"""Model a spectral resolution element.
:INPUTS:
param : sequence
param[0, 1, 2] - amplitude, sigma, and central location of
Gaussian line profile (cf. :func:`analysis.gaussian`).
param[3, 4, 5] - amplitude, width, and central location of
top-hat-like background (cf. :func:`tophat`).
param[6::] - additional (constant or polynomial) background
components, for evaluation with :func:`numpy.polyval`
x : sequence
Values at which to evaluate model function (i.e., pixels).
Typically 1D.
:OUTPUTS:
line : NumPy array
model of the resolution element, of same shape as `x`.
:DESCRIPTION:
Model a spectral resolution element along the spatial
direction. This consists of a (presumably Gaussian) line
profile superimposed on the spectral trace's top-hat-like
background, with an additional constant (or polynomial)
out-of-echelle-order background component.
"""
# 2012-04-11 12:57 IJMC: Created
# 2012-04-12 13:29 IJMC: Try to be more clever and save time in
# the back2 polyval call.
lineprofile = gaussian(param[0:3], x)
back1 = tophat(param[3:6], x)
if hasattr(param[6::], '__iter__') and len(param[6::])>1:
back2 = np.polyval(param[6::], x)
else:
back2 = param[6::]
return lineprofile + back1 + back2
[docs]
def add_partial_pixel(x0, y0, z0, z):
"""
:INPUTS:
x0 : int or sequence of ints
first index of z at which data will be added
y0 : int or sequence of ints
second index of z at which data will be added
z0 : scalar or sequence
values which will be added to z
z : 2D NumPy array
initial data, to which partial-pixels will be added
"""
# 2012-04-11 14:24 IJMC: Created
nx, ny = z.shape[0:2]
x = np.arange(nx)
y = np.arange(ny)
dx, dy = 1, 1
ret = np.array(z, copy=True)
if hasattr(x0, '__iter__'):
if len(x0)==len(y0) and len(x0)==len(z0):
for x00, y00, z00 in zip(x0, y0, z0):
ret = add_partial_pixel(x00, y00, z00, ret)
else:
print "Inputs x0, y0, and z0 must have the same length!"
ret = -1
if False:
if hasattr(x0, '__iter__'):
x0 = np.array(x0, copy=False)
y0 = np.array(y0, copy=False)
z0 = np.array(z0, copy=False)
else:
x0 = np.array([x0])
y0 = np.array([y0])
z0 = np.array([z0])
ix0 = np.tile(np.vstack((np.floor(x0), np.floor(x0)+1)).astype(int), (2,1))
iy0 = np.tile(np.vstack((np.floor(y0), np.floor(y0)+1)).astype(int), (2,1))
xfrac = x0 % 1
yfrac = y0 % 1
weights0 = np.vstack([(1. - xfrac) * (1. - yfrac), \
xfrac * (1. - yfrac), \
(1. - xfrac) * yfrac, \
xfrac * yfrac])
for ii in range(ix0.shape[1]):
print ii, weights0[:,ii]*z0[ii]
ret[ix0[:,ii], iy0[:,ii]] = ret[ix0[:,ii], iy0[:,ii]] + weights0[:,ii]*z0[ii]
else:
ix0 = map(int, [np.floor(x0), np.floor(x0)+1]*2)
iy0 = map(int, [np.floor(y0)]*2 + [np.floor(y0)+1]*2)
xfrac = x0 % 1
yfrac = y0 % 1
weights0 = [(1. - xfrac) * (1. - yfrac), \
xfrac * (1. - yfrac), \
(1. - xfrac) * yfrac, \
xfrac * yfrac]
ix = []
iy = []
weights = []
for ii in range(4):
#print ix0[ii], iy0[ii], weights0[ii]
if ix0[ii]>=0 and ix0[ii]<nx and iy0[ii]>=0 and iy0[ii]<ny:
ix.append(ix0[ii]), iy.append(iy0[ii])
weights.append(weights0[ii])
npix = len(ix)
if npix>0:
sumweights = sum(weights)
for ii in range(npix):
#print ix[ii], iy[ii], weights[ii]
ret[ix[ii], iy[ii]] += weights[ii] * z0
return ret
[docs]
def modelSpectralTrace(param, shape=None, nscat=None, npw=None, npy=None, noy=None, now=None, ndist=None, x=None, y=None, transpose=False):
"""Model a raw spectral trace!
:INPUTS:
NOTE that most inputs should be in the _rectified_ frame.
Trace background pedestal level : 1D array
Width of background pedestarl level : scalar (for now)
Center of trace : 1D array
Offset of object spectrum, relative to center : scalar
width of 1D PSF : scalar
Area of 1D psf : 1D array
Distortion (x and y, somehow???)
Scattered light background : scalar (for now)
"""
# 2012-04-11 17:50 IJMC: Created
########################################
# Parse various inputs:
########################################
param = np.array(param, copy=False)
nx, ny = shape
# Construct pixel vectors:
if x is None:
x = np.arange(nx)
if y is None:
y = np.arange(ny)
def makevec(input):
if not hasattr(input, '__iter__'):
ret = [input]*ny
#ret = np.polyval([input], y)
elif len(input)==1:
ret = [input[0]]*ny
#ret = np.polyval(input, y)
else:
ret = np.polyval(input, y)
return ret
# Define all the parameters:
pedestal_level = param[0:ny]
obj_flux = param[ny:ny*2]
pedestal_width = param[ny*2:ny*2+npw]
obj_fwhm = param[ny*2+npw:ny*2+npw+now]
spectrum_yparam = param[ny*2+npw+now : ny*2+npw+now+npy]
obj_yparam = param[ny*2+npw+now+npy : ny*2+npw+now+npy+noy]
disp_distort = param[ny*2+npw+now+npy+noy : ny*2+npw+now+npy+noy+ndist]
scat_param = list(param[-ndist:])
pedestal_width = makevec(pedestal_width)
obj_fwhm = makevec(obj_fwhm)
obj_yloc = makevec(obj_yparam) #, mode='cheby')
spectrum_yloc = makevec(spectrum_yparam) #, mode='cheby')
########################################
# Generate spectral model
########################################
rectified_model = np.zeros((nx, ny), float)
for ii in range(ny):
obj_param = [obj_flux[ii], obj_fwhm[ii], obj_yloc[ii]]
bkg_param = [pedestal_level[ii], pedestal_width[ii], spectrum_yloc[ii]]
oneDparam = obj_param + bkg_param + scat_param
rectified_model[:, ii] = model_resel(oneDparam, x)
#ypts = x
#xpts = ii + np.polyval(disp_distort, x-obj_yloc[ii])
#distorted_model = spec.add_partial_pixel(ypts, xpts, oneD, model)
interp_model = 0*rectified_model
if False:
# One way to do it:
newxcoords = (y - np.polyval(disp_distort, (x.reshape(nx,1)-obj_yloc.reshape(1,ny))))
meshmod = interpolate.RectBivariateSpline(x, y, rectified_model, kx=1, ky=1, s=0.)
for ii in range(nx):
interp_model[ii,:] = meshmod(x[ii], newxcoords[ii,:])
else:
# An alternate (not entirely equivalanet) approach, about equally fast:
shifts = np.polyval(disp_distort, x)
intshifts = np.floor(shifts)
minshift = intshifts.min()
shifts -= minshift
intshifts -= minshift
for ii in range(nx):
kern = np.zeros(intshifts.max() - intshifts.min()+2, float)
kern[intshifts[ii]] += 1. - (shifts[ii] - intshifts[ii])
kern[intshifts[ii]+1] += (shifts[ii] - intshifts[ii])
interp_model[ii] = np.convolve(rectified_model[ii], kern, 'same')
return interp_model
[docs]
def makeSpexSlitlessSky(skyfns, scatcen=[980, 150], scatdim=[60, 300]):
"""
Generate a normalized Sky frame from SpeX slitless spectroscopy data.
:INPUTS:
skyfns : list of strs
Filenames of slitless-spectroscopy sky frames
scatcen : 2-sequence
center of region to use in median-normalizing the frame.
scatdim : 2-sequence
full width of region to use in median-normalizing the frame.
:OUTPUTS:
(sky, skyHeader)
"""
# 2012-04-19 09:38 IJMC: Created
import phot
nsky = len(skyfns)
skyscat = phot.subreg2(skyfns, center=scatcen, dim=scatdim)
normfact = np.median(skyscat.reshape(nsky, np.prod(scatdim)), 1)
hdr = pyfits.getheader(skyfns[0])
skyframe = np.zeros((hdr['naxis1'], hdr['naxis2']), float)
for skyfn, factor in zip(skyfns, normfact):
skyframe += (pyfits.getdata(skyfn) / factor / nsky)
hdr.update('SKYRGCEN', str(scatcen))
hdr.update('SKYRGDIM', str(scatdim))
hdr.update('SKYNOTE', 'sky frame, median=1 in SKYRG')
return skyframe, hdr
[docs]
def resamplespec(w1, w0, spec0, oversamp=100):
"""
Resample a spectrum while conserving flux density.
:INPUTS:
w1 : sequence
new wavelength grid (i.e., center wavelength of each pixel)
w0 : sequence
old wavelength grid (i.e., center wavelength of each pixel)
spec0 : sequence
old spectrum (e.g., flux density or photon counts)
oversamp : int
factor by which to oversample input spectrum prior to
rebinning. The worst fractional precision you achieve is
roughly 1./oversamp.
:NOTE:
Format is the same as :func:`numpy.interp`
:REQUIREMENTS:
:doc:`tools`
"""
from tools import errxy
# 2012-04-25 18:40 IJMC: Created
nlam = len(w0)
x0 = np.arange(nlam, dtype=float)
x0int = np.arange((nlam-1.)*oversamp + 1., dtype=float)/oversamp
w0int = np.interp(x0int, x0, w0)
spec0int = np.interp(w0int, w0, spec0)/oversamp
# Set up the bin edges for down-binning
maxdiffw1 = np.diff(w1).max()
w1bins = np.concatenate(([w1[0] - maxdiffw1],
.5*(w1[1::] + w1[0:-1]), \
[w1[-1] + maxdiffw1]))
# Bin down the interpolated spectrum:
junk, spec1, junk2, junk3 = errxy(w0int, spec0int, w1bins, xmode=None, ymode='sum', xerr=None, yerr=None)
return spec1
#wcoords = spec.wavelengthMatch(mastersky0, wsky, thissky, ethissky, guess=wcoef1, order=None)
[docs]
def wavelengthMatch(spectrum, wtemplate, template, etemplate, guess=None, nthread=1):
"""
Determine dispersion solution for a spectrum, from a template.
:INPUTS:
spectrum : 1D sequence
Spectrum for which a wavelength solution is desired.
wtemplate : 1D sequence
Known wavelength grid of a template spectrum.
template : 1D sequence
Flux (e.g.) levels of the template spectrum with known
wavelength solution.
etemplate : 1D sequence
Uncertainties on the template values. This can be important
in a weighted fit!
:OPTIONS:
guess : sequence
Initial guess for the wavelength solution. This is very
helpful, if you have it! The guess should be a sequence
containing the set of Chebychev polynomial coefficients,
followed by a scale factor and DC offset (to help in scaling
the template).
If guess is None, attempt to fit a simple linear dispersion relation.
order : int > 0
NOT YET IMPLEMENTED! BUT EVENTUALLY: if guess is None, this
sets the polynomial order of the wavelength solution.
FOR THE MOMENT: if guess is None, return a simple linear
solution. This is likely to fail entirely for strongly
nonlinear dispersion solutions or poorly mismatched template
and spectrum.
nthread : int > 0
Number of processors to use for MCMC searching.
:RETURNS:
(wavelength, wavelength_polynomial_coefficients, full_parameter_set)
:NOTES:
This implementation uses a rather crude MCMC sampling approach
to sample parameters space and 'home in' on better solutions.
There is probably a way to do this that is both faster and more
optimal...
Note that if 'spectrum' and 'template' are of different lengths,
the longer one will be trimmed at the end to make the lengths match.
:REQUIREMENTS:
`emcee <http://TBD>`_
"""
#2012-04-25 20:53 IJMC: Created
# 2012-09-23 20:17 IJMC: Now spectrum & template can be different length.
# 2013-03-09 17:23 IJMC: Added nthread option
import emcee
import phasecurves as pc
nlam_s = len(spectrum)
nlam_t = len(template)
if nlam_s <= nlam_t:
template = np.array(template, copy=True)[0:nlam_s]
etemplate = np.array(etemplate, copy=True)[0:nlam_s]
nlam = nlam_s
spectrum_trimmed = False
else: # nlam_s > nlam_t:
spectrum0 = np.array(spectrum, copy=True)
spectrum = spectrum0[0:nlam_t]
wtemplate = np.array(wtemplate, copy=True)[0:nlam_t]
nlam = nlam_t
spectrum_trimmed = True
# Create a normalized vector of coordinates for computing
# normalized polynomials:
dx0 = 1. / (nlam - 1.)
x0n = dx0 * np.arange(nlam) - 1.
#x0n = 2*np.arange(nlam, dtype=float) / (nlam - 1.) - 1
if guess is None:
# Start with a simple linear wavelength solution.
guess = [np.diff(wtemplate).mean() * len(template)/2, np.mean(wtemplate), np.median(template)/np.median(spectrum), 1.]
# Define arguments for use by fitting routines:
fitting_args = (makemodel, x0n, spectrum, wtemplate, template, 1./etemplate**2)
# Try to find an initial best fit:
bestparams = an.fmin(pc.errfunc, guess, args=fitting_args, disp=False)
# Initial fit is likely a local minimum, so explore parameter
# space using an MCMC approach.
ndim = len(guess)
nwalkers = ndim * 50
sampler = emcee.EnsembleSampler(nwalkers, ndim, pc.lnprobfunc, args=fitting_args, threads=nthread)
# Initialize the sampler with various starting positions:
e_params1 = np.vstack((np.array(guess)/10., np.zeros(ndim) + .01)).max(0)
e_params2 = np.vstack((bestparams/10., np.zeros(ndim) + .01)).max(0)
p0 = np.vstack(([guess, bestparams], \
[np.random.normal(guess, e_params1) for ii in xrange(nwalkers/2-1)], \
[np.random.normal(bestparams, e_params2) for ii in xrange(nwalkers/2-1)]))
# Run the sampler for a while:
pos, prob, state = sampler.run_mcmc(p0, 300) # Burn-in
bestparams = sampler.flatchain[np.nonzero(sampler.lnprobability.ravel()==sampler.lnprobability.ravel().max())[0][0]]
# Optimize the latest set of best parameters.
bestparams = an.fmin(pc.errfunc, bestparams, args=fitting_args, disp=False)
dispersionSolution = bestparams[0:-2]
if spectrum_trimmed:
x0n_original = dx0 * np.arange(nlam_s) - 1.
wavelengths = np.polyval(dispersionSolution, x0n_original)
else:
wavelengths = np.polyval(dispersionSolution, x0n)
return dispersionSolution, wavelengths, bestparams
[docs]
def makemodel(params, xvec, specvec, wtemplate):
"""Helper function for :func:`wavelengthMatch`: generate a scaled,
interpolative model of the template."""
wcoef = params[0:-2]
scale, offset = params[-2::]
neww = np.polyval(wcoef, xvec)
return offset + scale * np.interp(wtemplate, neww, specvec, left=0, right=0)
[docs]
def normalizeSpecFlat(flatdat, nspec=1, minsep=50, median_width=51, readnoise=40, badpixelmask=None, traces=None):
"""Trace and normalize a spectroscopic flat field frame.
:INPUTS:
flatdat : 2D NumPy array
Master, unnormalized flat frame: assumed to be measured in
photoelectrons (for computing uncertainties).
nspec : int
Number of spectral orders to find and normalize
minsep : int
Minimum separation, in pixels, between spectral orders that
will be found.
median_width : int
Width of median-filter kernel used to compute the low-
readnoise : scalar
Detector read noise, in electrons. For computing uncertainties.
badpixelmask : 2D NumPy array
bad pixel mask: 1 at bad pixel locations, 0 elsewhere.
traces : 2D NumPy Array
(nord, pord) shaped numpy array representing the polynomial
coefficients for each order (suitable for use with
np.polyval), as produced by :func:`traceorders`
"""
# 2012-04-28 06:22 IJMC: Created
# 2012-07-24 21:04 IJMC: Now, as a final step, all bad indices are set to unity.
# 2014-12-17 20:07 IJMC: Added 'traces' option
import analysis as an
from scipy import signal
if badpixelmask is None:
badpixelmask = np.zeros(flatdat.shape, bool)
# Enforce positivity and de-weight negative flux values:
e_flatdat = np.sqrt(flatdat + readnoise**2)
badindices = ((flatdat<=0) + badpixelmask).nonzero()
e_flatdat[badindices] = flatdat[badindices] * 1e9
flatdat[badindices] = 1.
# Find spectral orders, using derivatives (will probably fail if
# spec. overlaps the edge!):
ordvec = an.meanr(flatdat, axis=1, nsigma=3)
filtvec = signal.medfilt(ordvec, 9)
dvec1 = np.diff(filtvec)
dvec2 = -np.diff(filtvec)
dvec1[dvec1<0] = 0.
dvec2[dvec2<0] = 0.
x1 = np.arange(dvec1.size)
available1 = np.ones(dvec1.size, dtype=bool)
available2 = np.ones(dvec1.size, dtype=bool)
pos1 = []
pos2 = []
for ii in range(nspec):
thisx1 = x1[dvec1==dvec1[available1].max()][0]
available1[np.abs(x1 - thisx1) < minsep] = False
pos1.append(thisx1)
thisx2 = x1[dvec2==dvec2[available2].max()][0]
available2[np.abs(x1 - thisx2) < minsep] = False
pos2.append(thisx2)
limits = np.array(zip(np.sort(pos1), np.sort(pos2)))
# Generate and normalize the spectral traces:
masterflat = np.ones(flatdat.shape, dtype=float)
if traces is not None:
nx = flatdat.shape[1]
xvec = np.arange(nx)
ymat = np.tile(np.arange(flatdat.shape[0]), (nx, 1)).T
for ii in range(nspec):
if traces is None:
profvec = np.median(flatdat[limits[ii,0]:limits[ii,1], :], axis=0)
e_profvec = np.sqrt(an.wmean(flatdat[limits[ii,0]:limits[ii,1], :], 1./e_flatdat[limits[ii,0]:limits[ii,1], :]**2, axis=0) / np.diff(limits[ii]))[0]
e_profvec[e_profvec <= 0] = profvec.max()*1e9
smooth_prof = signal.medfilt(profvec, median_width)
masterflat[limits[ii,0]:limits[ii,1], :] = flatdat[limits[ii,0]:limits[ii,1], :] / smooth_prof
else:
traceloc = np.polyval(traces[ii], xvec)
limind = (limits[:,1] > traceloc.mean()).nonzero()[0][0]
order_ind_2d = ((ymat - traceloc) > (limits[limind,0] - traceloc.mean())) * ((ymat - traceloc) < (limits[limind,1] - traceloc.mean()))
profvec = np.array([np.median(flatdat[order_ind_2d[:,jj], jj]) for jj in xrange(nx)])
smooth_prof = signal.medfilt(profvec, median_width)
for jj in xrange(nx):
masterflat[order_ind_2d[:,jj],jj] = flatdat[order_ind_2d[:,jj],jj] / smooth_prof[jj]
# Ideally, we would do some sort of weighted fitting here. Instead,
# for now, just take a running median:
masterflat[badindices] = 1.
return masterflat
[docs]
def optspecextr_idl(frame, gain, readnoise, x1, x2, idlexec, clobber=True, tempframefn='tempframe.fits', specfn='tempspec.fits', scriptfn='temp_specextract.pro', IDLoptions="adjfunc='adjgauss', adjoptions={center:1,centerfit:1,centerdeg:3}, bgdeg=3", inmask=None):
"""Run optimal spectral extraction in IDL; pass results to Python.
:INPUTS:
frame : str
filename, or 2D Numpy Array, or list of filenames containing
frames from which spectra will be extracted. This should be
in units of ADU (not electrons) for the noise properties to
come out properly.
Also, the spectral trace must run vertically across the frame.
gain : scalar
Detector gain, in electrons / ADU
readnoise : scalar
Detector read noise, in electrons
x1, x2 : ints, or lists of ints
Start and stop indices of the spectral trace across the frame.
If multiple frames are input and a single x1/x2 is input, the
same value will be used for each frame. Note however that
multiple x1/x2 can also be input (one for each frame).
idlexec : str
Path to the IDL executable. OPTSPECEXTR.PRO and its
associated files must be in your IDL path. If set to None,
then it will be set to: os.popen('which idl').read().strip()
:OPTIONS:
clobber : bool
Whether to overwrite files when writing input data to TEMPFRAMFN.
tempframefn : str
If input 'frame' is an array, it will be written to this
filename in order to pass it to IDL.
specfn : str
IDL will write the spectral data to this filename in order to
pass it back to Python.
scriptfn : str
Filename in which the short IDL script will be written.
IDLoptions : str
Options to pass to OPTSPECEXTR.PRO. For example:
"adjfunc='adjgauss', adjoptions={center:1,centerfit:1,centerdeg:3}, bgdeg=3"
Note that this Python code will break if you _don't_ trace
the spectrum (adjoptions, etc.); this is an area for future
work if I ever use a spectrograph with straight traces.
inmask : None or str
Name of the good pixel mask for OPTSPECEXTR.PRO. Equal to 1
for good pixels, and 0 for bad pixels.
:OUTPUTS:
For each input frame, a list of four items:
[0] -- Extracted spectrum, ADU per pixel
[1] -- Uncertainty (1 sigma) of extracted spectrum
[2] -- Location of trace (in pixels) across the frame
[3] -- Width of trace across the frame
:NOTES:
Note that this more closely follows Horne et al. than does
:func:`optimalExtract`, and is faster than both that function
and (especially!) :func:`extractSpectralProfiles`. The only
downside (if it is one) is that this function requires IDL.
:TO-DO:
Add options for user input of a variance frame, or of sky variance.
Allow more flexibility (tracing, input/output options, etc.)
:REQUIREMENTS:
IDL
`OPTSPECEXTR <http://physics.ucf.edu/~jh/ast/software.html>`_
"""
# 2012-08-18 16:36 IJMC: created
# 2012-08-19 09:39 IJMC: Added 'inmask' option.
import os
try:
from astropy.io import fits as pyfits
except:
import pyfits
# Put the input frames in the proper format:
if isinstance(frame, np.ndarray):
frameisfilename = False
if frame.ndim==2:
frames = [frame]
elif frame.ndim==1:
print "Input array should be 2D or 3D -- no telling what will happen next!"
else:
frames = frame
else:
frameisfilename = True
if isinstance(frame, str):
frames = [frame]
else:
frames = frame
if not hasattr(x1, '__iter__'):
x1 = [x1] * len(frames)
if not hasattr(x2, '__iter__'):
x2 = [x2] * len(frames)
if idlexec is None:
idlexec = os.popen('which idl').read().strip()
# Loop through all files:
specs = []
ii = 0
for frame in frames:
if frameisfilename:
tempframefn = frame
else:
pyfits.writeto(tempframefn, frame, clobber=clobber)
# Prepare the temporary IDL script:
idlcmds = []
idlcmds.append("frame = readfits('%s')\n" % tempframefn)
idlcmds.append("gain = %1.3f\n" % gain)
idlcmds.append("readnoise = %i\n" % readnoise)
idlcmds.append("varim = abs(frame) / gain + readnoise^2\n")
idlcmds.append("x1 = %i & x2 = %i\n" % (x1[ii],x2[ii]))
if inmask is not None:
idlcmds.append("inmask = readfits('%s')\n" % inmask)
IDLoptions += ', inmask=inmask'
idlcmds.append("spec = optspecextr(frame, varim, readnoise, gain, x1, x2, adjparms=adjparm, opvar=opvar, %s)\n" % IDLoptions)
idlcmds.append("spec_err_loc_width = [[spec], [sqrt(opvar)], [adjparm.traceest], [adjparm.widthest]]\n")
idlcmds.append("writefits,'%s', spec_err_loc_width\n" % (specfn))
idlcmds.append("exit\n")
# Write it to disk, and execute it.
f = open(scriptfn, 'w')
f.writelines(idlcmds)
f.close()
os.system('%s %s' % (idlexec, scriptfn))
# Read the spectrum into Python, and iterate.
spec = pyfits.getdata(specfn)
specs.append(spec)
ii += 1
# Clean up after ourselves:
if not frameisfilename and os.path.isfile(tempframefn):
os.remove(tempframefn)
if os.path.isfile(specfn):
os.remove(specfn)
if os.path.isfile(scriptfn):
os.remove(scriptfn)
# If only one file was run, we don't need to return a list.
if ii==1:
specs = specs[0]
return specs
[docs]
def optimalExtract(*args, **kw):
"""
Extract spectrum, following Horne 1986.
:INPUTS:
data : 2D Numpy array
Appropriately calibrated frame from which to extract
spectrum. Should be in units of ADU, not electrons!
variance : 2D Numpy array
Variances of pixel values in 'data'.
gain : scalar
Detector gain, in electrons per ADU
readnoise : scalar
Detector readnoise, in electrons.
:OPTIONS:
goodpixelmask : 2D numpy array
Equals 0 for bad pixels, 1 for good pixels
bkg_radii : 2- or 4-sequence
If length 2: inner and outer radii to use in computing
background. Note that for this to be effective, the spectral
trace should be positions in the center of 'data.'
If length 4: start and end indices of both apertures for
background fitting, of the form [b1_start, b1_end, b2_start,
b2_end] where b1 and b2 are the two background apertures, and
the elements are arranged in strictly ascending order.
extract_radius : int or 2-sequence
radius to use for both flux normalization and extraction. If
a sequence, the first and last indices of the array to use
for spectral normalization and extraction.
dispaxis : bool
0 for horizontal spectrum, 1 for vertical spectrum
bord : int >= 0
Degree of polynomial background fit.
bsigma : int >= 0
Sigma-clipping threshold for computing background.
pord : int >= 0
Degree of polynomial fit to construct profile.
psigma : int >= 0
Sigma-clipping threshold for computing profile.
csigma : int >= 0
Sigma-clipping threshold for cleaning & cosmic-ray rejection.
finite : bool
If true, mask all non-finite values as bad pixels.
nreject : int > 0
Number of pixels to reject in each iteration.
:RETURNS:
3-tuple:
[0] -- spectrum flux (in electrons)
[1] -- uncertainty on spectrum flux
[1] -- background flux
:EXAMPLE:
::
:SEE_ALSO:
:func:`superExtract`.
:NOTES:
Horne's classic optimal extraction algorithm is optimal only so
long as the spectral traces are very nearly aligned with
detector rows or columns. It is *not* well-suited for
extracting substantially tilted or curved traces, for the
reasons described by Marsh 1989, Mukai 1990. For extracting
such spectra, see :func:`superExtract`.
"""
# 2012-08-20 08:24 IJMC: Created from previous, low-quality version.
# 2012-09-03 11:37 IJMC: Renamed to replace previous, low-quality
# version. Now bkg_radii and extract_radius
# can refer to either a trace-centered
# coordinate system, or the specific
# indices of all aperture edges. Added nreject.
from scipy import signal
# Parse inputs:
frame, variance, gain, readnoise = args[0:4]
# Parse options:
if kw.has_key('goodpixelmask'):
goodpixelmask = np.array(kw['goodpixelmask'], copy=True).astype(bool)
else:
goodpixelmask = np.ones(frame.shape, dtype=bool)
if kw.has_key('dispaxis'):
if kw['dispaxis']==1:
frame = frame.transpose()
variance = variance.transpose()
goodpixelmask = goodpixelmask.transpose()
if kw.has_key('verbose'):
verbose = kw['verbose']
else:
verbose = False
if kw.has_key('bkg_radii'):
bkg_radii = kw['bkg_radii']
else:
bkg_radii = [15, 20]
if verbose: message("Setting option 'bkg_radii' to: " + str(bkg_radii))
if kw.has_key('extract_radius'):
extract_radius = kw['extract_radius']
else:
extract_radius = 10
if verbose: message("Setting option 'extract_radius' to: " + str(extract_radius))
if kw.has_key('bord'):
bord = kw['bord']
else:
bord = 1
if verbose: message("Setting option 'bord' to: " + str(bord))
if kw.has_key('bsigma'):
bsigma = kw['bsigma']
else:
bsigma = 3
if verbose: message("Setting option 'bsigma' to: " + str(bsigma))
if kw.has_key('pord'):
pord = kw['pord']
else:
pord = 2
if verbose: message("Setting option 'pord' to: " + str(pord))
if kw.has_key('psigma'):
psigma = kw['psigma']
else:
psigma = 4
if verbose: message("Setting option 'psigma' to: " + str(psigma))
if kw.has_key('csigma'):
csigma = kw['csigma']
else:
csigma = 5
if verbose: message("Setting option 'csigma' to: " + str(csigma))
if kw.has_key('finite'):
finite = kw['finite']
else:
finite = True
if verbose: message("Setting option 'finite' to: " + str(finite))
if kw.has_key('nreject'):
nreject = kw['nreject']
else:
nreject = 100
if verbose: message("Setting option 'nreject' to: " + str(nreject))
if finite:
goodpixelmask *= (np.isfinite(frame) * np.isfinite(variance))
variance[True-goodpixelmask] = frame[goodpixelmask].max() * 1e9
nlam, fitwidth = frame.shape
xxx = np.arange(-fitwidth/2, fitwidth/2)
xxx0 = np.arange(fitwidth)
if len(bkg_radii)==4: # Set all borders of background aperture:
backgroundAperture = ((xxx0 > bkg_radii[0]) * (xxx0 <= bkg_radii[1])) + \
((xxx0 > bkg_radii[2]) * (xxx0 <= bkg_radii[3]))
else: # Assume trace is centered, and use only radii.
backgroundAperture = (np.abs(xxx) > bkg_radii[0]) * (np.abs(xxx) <= bkg_radii[1])
if hasattr(extract_radius, '__iter__'):
extractionAperture = (xxx0 > extract_radius[0]) * (xxx0 <= extract_radius[1])
else:
extractionAperture = np.abs(xxx) < extract_radius
nextract = extractionAperture.sum()
xb = xxx[backgroundAperture]
#Step3: Sky Subtraction
if bord==0: # faster to take weighted mean:
background = an.wmean(frame[:, backgroundAperture], (goodpixelmask/variance)[:, backgroundAperture], axis=1)
else:
background = 0. * frame
for ii in range(nlam):
fit = an.polyfitr(xb, frame[ii, backgroundAperture], bord, bsigma, w=(goodpixelmask/variance)[ii, backgroundAperture], verbose=verbose-1)
background[ii, :] = np.polyval(fit, xxx)
# (my 3a: mask any bad values)
badBackground = True - np.isfinite(background)
background[badBackground] = 0.
if verbose and badBackground.any():
print "Found bad background values at: ", badBackground.nonzero()
skysubFrame = frame - background
#Step4: Extract 'standard' spectrum and its variance
standardSpectrum = nextract * an.wmean(skysubFrame[:, extractionAperture], goodpixelmask[:, extractionAperture], axis=1)
varStandardSpectrum = nextract * an.wmean(variance[:, extractionAperture], goodpixelmask[:, extractionAperture], axis=1)
# (my 4a: mask any bad values)
badSpectrum = True - (np.isfinite(standardSpectrum))
standardSpectrum[badSpectrum] = 1.
varStandardSpectrum[badSpectrum] = varStandardSpectrum[True - badSpectrum].max() * 1e9
#Step5: Construct spatial profile; enforce positivity & normalization
normData = skysubFrame / standardSpectrum
varNormData = variance / standardSpectrum**2
# Iteratively clip outliers
newBadPixels = True
iter = -1
if verbose: print "Looking for bad pixel outliers in profile construction."
xl = np.linspace(-1., 1., nlam)
while newBadPixels:
iter += 1
if pord==0: # faster to take weighted mean:
profile = np.tile(an.wmean(normData, (goodpixelmask/varNormData), axis=0), (nlam,1))
else:
profile = 0. * frame
for ii in range(fitwidth):
fit = an.polyfitr(xl, normData[:, ii], pord, np.inf, w=(goodpixelmask/varNormData)[:, ii], verbose=verbose-1)
profile[:, ii] = np.polyval(fit, xl)
if profile.min() < 0:
profile[profile < 0] = 0.
profile /= profile.sum(1).reshape(nlam, 1)
#Step6: Revise variance estimates
modelData = standardSpectrum * profile + background
variance = (np.abs(modelData)/gain + (readnoise/gain)**2) / \
(goodpixelmask + 1e-9) # Avoid infinite variance
outlierSigmas = (frame - modelData)**2/variance
if outlierSigmas.max() > psigma**2:
maxRejectedValue = max(psigma**2, np.sort(outlierSigmas[:, extractionAperture].ravel())[-nreject])
worstOutliers = (outlierSigmas>=maxRejectedValue).nonzero()
goodpixelmask[worstOutliers] = False
newBadPixels = True
numberRejected = len(worstOutliers[0])
else:
newBadPixels = False
numberRejected = 0
if verbose: print "Rejected %i pixels on this iteration " % numberRejected
#Step5: Construct spatial profile; enforce positivity & normalization
varNormData = variance / standardSpectrum**2
if verbose: print "%i bad pixels found" % iter
# Iteratively clip Cosmic Rays
newBadPixels = True
iter = -1
if verbose: print "Looking for bad pixel outliers in optimal extraction."
while newBadPixels:
iter += 1
#Step 8: Extract optimal spectrum and its variance
gp = goodpixelmask * profile
denom = (gp * profile / variance)[:, extractionAperture].sum(1)
spectrum = ((gp * skysubFrame / variance)[:, extractionAperture].sum(1) / denom).reshape(nlam, 1)
varSpectrum = (gp[:, extractionAperture].sum(1) / denom).reshape(nlam, 1)
#Step6: Revise variance estimates
modelData = spectrum * profile + background
variance = (np.abs(modelData)/gain + (readnoise/gain)**2) / \
(goodpixelmask + 1e-9) # Avoid infinite variance
#Iterate until worse outliers are all identified:
outlierSigmas = (frame - modelData)**2/variance
if outlierSigmas.max() > csigma**2:
maxRejectedValue = max(csigma**2, np.sort(outlierSigmas[:, extractionAperture].ravel())[-nreject])
worstOutliers = (outlierSigmas>=maxRejectedValues).nonzero()
goodpixelmask[worstOutliers] = False
newBadPixels = True
numberRejected = len(worstOutliers[0])
else:
newBadPixels = False
numberRejected = 0
if verbose: print "Rejected %i pixels on this iteration " % numberRejected
if verbose: print "%i bad pixels found" % iter
ret = (spectrum, varSpectrum, profile, background, goodpixelmask)
return ret
[docs]
def superExtract(*args, **kw):
"""
Optimally extract curved spectra, following Marsh 1989.
:INPUTS:
data : 2D Numpy array
Appropriately calibrated frame from which to extract
spectrum. Should be in units of ADU, not electrons!
variance : 2D Numpy array
Variances of pixel values in 'data'.
gain : scalar
Detector gain, in electrons per ADU
readnoise : scalar
Detector readnoise, in electrons.
:OPTIONS:
trace : 1D numpy array
location of spectral trace. If None, :func:`traceorders` is
invoked.
goodpixelmask : 2D numpy array
Equals 0 for bad pixels, 1 for good pixels
npoly : int
Number of profile polynomials to evaluate (Marsh's
"K"). Ideally you should not need to set this -- instead,
play with 'polyspacing' and 'extract_radius.' For symmetry,
this should be odd.
polyspacing : scalar
Spacing between profile polynomials, in pixels. (Marsh's
"S"). A few cursory tests suggests that the extraction
precision (in the high S/N case) scales as S^-2 -- but the
code slows down as S^2.
pord : int
Order of profile polynomials; 1 = linear, etc.
bkg_radii : 2-sequence
inner and outer radii to use in computing background
extract_radius : int
radius to use for both flux normalization and extraction
dispaxis : bool
0 for horizontal spectrum, 1 for vertical spectrum
bord : int >= 0
Degree of polynomial background fit.
bsigma : int >= 0
Sigma-clipping threshold for computing background.
tord : int >= 0
Degree of spectral-trace polynomial (for trace across frame
-- not used if 'trace' is input)
csigma : int >= 0
Sigma-clipping threshold for cleaning & cosmic-ray rejection.
finite : bool
If true, mask all non-finite values as bad pixels.
qmode : str ('fast' or 'slow')
How to compute Marsh's Q-matrix. Valid inputs are
'fast-linear', 'slow-linear', 'fast-nearest,' 'slow-nearest,'
and 'brute'. These select between various methods of
integrating the nearest-neighbor or linear interpolation
schemes as described by Marsh; the 'linear' methods are
preferred for accuracy. Use 'slow' if you are running out of
memory when using the 'fast' array-based methods. 'Brute' is
both slow and inaccurate, and should not be used.
nreject : int
Number of outlier-pixels to reject at each iteration.
retall : bool
If true, also return the 2D profile, background, variance
map, and bad pixel mask.
:RETURNS:
object with fields for:
spectrum
varSpectrum
trace
:EXAMPLE:
::
import spec
import numpy as np
import pylab as py
def gaussian(p, x):
if len(p)==3:
p = concatenate((p, [0]))
return (p[3] + p[0]/(p[1]*sqrt(2*pi)) * exp(-(x-p[2])**2 / (2*p[1]**2)))
# Model some strongly tilted spectral data:
nx, nlam = 80, 500
x0 = np.arange(nx)
gain, readnoise = 3.0, 30.
background = np.ones(nlam)*10
flux = np.ones(nlam)*1e4
center = nx/2. + np.linspace(0,10,nlam)
FWHM = 3.
model = np.array([gaussian([flux[ii]/gain, FWHM/2.35, center[ii], background[ii]], x0) for ii in range(nlam)])
varmodel = np.abs(model) / gain + (readnoise/gain)**2
observation = np.random.normal(model, np.sqrt(varmodel))
fitwidth = 60
xr = 15
output_spec = spec.superExtract(observation, varmodel, gain, readnoise, polyspacing=0.5, pord=1, bkg_radii=[10,30], extract_radius=5, dispaxis=1)
py.figure()
py.plot(output_spec.spectrum.squeeze() / flux)
py.ylabel('(Measured flux) / (True flux)')
py.xlabel('Photoelectrons')
:TO_DO:
Iterate background fitting and reject outliers; maybe first time
would be unweighted for robustness.
Introduce even more array-based, rather than loop-based,
calculations. For large spectra computing the C-matrix takes
the most time; this should be optimized somehow.
:SEE_ALSO:
"""
# 2012-08-25 20:14 IJMC: Created.
# 2012-09-21 14:32 IJMC: Added error-trapping if no good pixels
# are in a row. Do a better job of extracting
# the initial 'standard' spectrum.
from scipy import signal
from pylab import *
from nsdata import imshow, bfixpix
# Parse inputs:
frame, variance, gain, readnoise = args[0:4]
frame = gain * np.array(frame, copy=False)
variance = gain**2 * np.array(variance, copy=False)
variance[variance<=0.] = readnoise**2
# Parse options:
if kw.has_key('verbose'):
verbose = kw['verbose']
else:
verbose = False
if verbose: from time import time
if kw.has_key('goodpixelmask'):
goodpixelmask = kw['goodpixelmask']
if isinstance(goodpixelmask, str):
goodpixelmask = pyfits.getdata(goodpixelmask).astype(bool)
else:
goodpixelmask = np.array(goodpixelmask, copy=True).astype(bool)
else:
goodpixelmask = np.ones(frame.shape, dtype=bool)
if kw.has_key('dispaxis'):
dispaxis = kw['dispaxis']
else:
dispaxis = 0
if dispaxis==0:
frame = frame.transpose()
variance = variance.transpose()
goodpixelmask = goodpixelmask.transpose()
if kw.has_key('pord'):
pord = kw['pord']
else:
pord = 2
if kw.has_key('polyspacing'):
polyspacing = kw['polyspacing']
else:
polyspacing = 1
if kw.has_key('bkg_radii'):
bkg_radii = kw['bkg_radii']
else:
bkg_radii = [15, 20]
if verbose: message("Setting option 'bkg_radii' to: " + str(bkg_radii))
if kw.has_key('extract_radius'):
extract_radius = kw['extract_radius']
else:
extract_radius = 10
if verbose: message("Setting option 'extract_radius' to: " + str(extract_radius))
if kw.has_key('npoly'):
npoly = kw['npoly']
else:
npoly = 2 * int((2.0 * extract_radius) / polyspacing / 2.) + 1
if kw.has_key('bord'):
bord = kw['bord']
else:
bord = 1
if verbose: message("Setting option 'bord' to: " + str(bord))
if kw.has_key('tord'):
tord = kw['tord']
else:
tord = 3
if verbose: message("Setting option 'tord' to: " + str(tord))
if kw.has_key('bsigma'):
bsigma = kw['bsigma']
else:
bsigma = 3
if verbose: message("Setting option 'bsigma' to: " + str(bsigma))
if kw.has_key('csigma'):
csigma = kw['csigma']
else:
csigma = 5
if verbose: message("Setting option 'csigma' to: " + str(csigma))
if kw.has_key('qmode'):
qmode = kw['qmode']
else:
qmode = 'fast'
if verbose: message("Setting option 'qmode' to: " + str(qmode))
if kw.has_key('nreject'):
nreject = kw['nreject']
else:
nreject = 100
if verbose: message("Setting option 'nreject' to: " + str(nreject))
if kw.has_key('finite'):
finite = kw['finite']
else:
finite = True
if verbose: message("Setting option 'finite' to: " + str(finite))
if kw.has_key('retall'):
retall = kw['retall']
else:
retall = False
if finite:
goodpixelmask *= (np.isfinite(frame) * np.isfinite(variance))
variance[True-goodpixelmask] = frame[goodpixelmask].max() * 1e9
nlam, fitwidth = frame.shape
# Define trace (Marsh's "X_j" in Eq. 9)
if kw.has_key('trace'):
trace = kw['trace']
else:
trace = None
if trace is None:
trace = 5
if not hasattr(trace, '__iter__'):
if verbose: print "Tracing not fully tested; dispaxis may need adjustment."
#pdb.set_trace()
tracecoef = traceorders(frame, pord=trace, nord=1, verbose=verbose-1, plotalot=verbose-1, g=gain, rn=readnoise, badpixelmask=True-goodpixelmask, dispaxis=dispaxis, fitwidth=min(fitwidth, 80))
trace = np.polyval(tracecoef.ravel(), np.arange(nlam))
#xxx = np.arange(-fitwidth/2, fitwidth/2)
#backgroundAperture = (np.abs(xxx) > bkg_radii[0]) * (np.abs(xxx) < bkg_radii[1])
#extractionAperture = np.abs(xxx) < extract_radius
#nextract = extractionAperture.sum()
#xb = xxx[backgroundAperture]
xxx = np.arange(fitwidth) - trace.reshape(nlam,1)
backgroundApertures = (np.abs(xxx) > bkg_radii[0]) * (np.abs(xxx) <= bkg_radii[1])
extractionApertures = np.abs(xxx) <= extract_radius
nextracts = extractionApertures.sum(1)
#Step3: Sky Subtraction
background = 0. * frame
for ii in range(nlam):
if goodpixelmask[ii, backgroundApertures[ii]].any():
fit = an.polyfitr(xxx[ii,backgroundApertures[ii]], frame[ii, backgroundApertures[ii]], bord, bsigma, w=(goodpixelmask/variance)[ii, backgroundApertures[ii]], verbose=verbose-1)
background[ii, :] = np.polyval(fit, xxx[ii])
else:
background[ii] = 0.
background_at_trace = np.array([np.interp(0, xxx[j], background[j]) for j in range(nlam)])
# (my 3a: mask any bad values)
badBackground = True - np.isfinite(background)
background[badBackground] = 0.
if verbose and badBackground.any():
print "Found bad background values at: ", badBackground.nonzero()
skysubFrame = frame - background
# Interpolate and fix bad pixels for extraction of standard
# spectrum -- otherwise there can be 'holes' in the spectrum from
# ill-placed bad pixels.
fixSkysubFrame = bfixpix(skysubFrame, True-goodpixelmask, n=8, retdat=True)
#Step4: Extract 'standard' spectrum and its variance
standardSpectrum = np.zeros((nlam, 1), dtype=float)
varStandardSpectrum = np.zeros((nlam, 1), dtype=float)
for ii in range(nlam):
thisrow_good = extractionApertures[ii] #* goodpixelmask[ii] *
standardSpectrum[ii] = fixSkysubFrame[ii, thisrow_good].sum()
varStandardSpectrum[ii] = variance[ii, thisrow_good].sum()
spectrum = standardSpectrum.copy()
varSpectrum = varStandardSpectrum
# Define new indices (in Marsh's appendix):
N = pord + 1
mm = np.tile(np.arange(N).reshape(N,1), (npoly)).ravel()
nn = mm.copy()
ll = np.tile(np.arange(npoly), N)
kk = ll.copy()
pp = N * ll + mm
qq = N * kk + nn
jj = np.arange(nlam) # row (i.e., wavelength direction)
ii = np.arange(fitwidth) # column (i.e., spatial direction)
jjnorm = np.linspace(-1, 1, nlam) # normalized X-coordinate
jjnorm_pow = jjnorm.reshape(1,1,nlam) ** (np.arange(2*N-1).reshape(2*N-1,1,1))
# Marsh eq. 9, defining centers of each polynomial:
constant = 0. # What is it for???
poly_centers = trace.reshape(nlam, 1) + polyspacing * np.arange(-npoly/2+1, npoly/2+1) + constant
# Marsh eq. 11, defining Q_kij (via nearest-neighbor interpolation)
# Q_kij = max(0, min(S, (S+1)/2 - abs(x_kj - i)))
if verbose: tic = time()
if qmode=='fast-nearest': # Array-based nearest-neighbor mode.
if verbose: tic = time()
Q = np.array([np.zeros((npoly, fitwidth, nlam)), np.array([polyspacing * np.ones((npoly, fitwidth, nlam)), 0.5 * (polyspacing+1) - np.abs((poly_centers - ii.reshape(fitwidth, 1, 1)).transpose(2, 0, 1))]).min(0)]).max(0)
elif qmode=='slow-linear': # Code is a mess, but it works.
invs = 1./polyspacing
poly_centers_over_s = poly_centers / polyspacing
xps_mat = poly_centers + polyspacing
xms_mat = poly_centers - polyspacing
Q = np.zeros((npoly, fitwidth, nlam), dtype=float)
for i in range(fitwidth):
ip05 = i + 0.5
im05 = i - 0.5
for j in range(nlam):
for k in range(npoly):
xkj = poly_centers[j,k]
xkjs = poly_centers_over_s[j,k]
xps = xps_mat[j,k] #xkj + polyspacing
xms = xms_mat[j,k] # xkj - polyspacing
if (ip05 <= xms) or (im05 >= xps):
qval = 0.
elif (im05) > xkj:
lim1 = im05
lim2 = min(ip05, xps)
qval = (lim2 - lim1) * \
(1. + xkjs - 0.5*invs*(lim1+lim2))
elif (ip05) < xkj:
lim1 = max(im05, xms)
lim2 = ip05
qval = (lim2 - lim1) * \
(1. - xkjs + 0.5*invs*(lim1+lim2))
else:
lim1 = max(im05, xms)
lim2 = min(ip05, xps)
qval = lim2 - lim1 + \
invs * (xkj*(-xkj + lim1 + lim2) - \
0.5*(lim1*lim1 + lim2*lim2))
Q[k,i,j] = max(0, qval)
elif qmode=='fast-linear': # Code is a mess, but it's faster than 'slow' mode
invs = 1./polyspacing
xps_mat = poly_centers + polyspacing
Q = np.zeros((npoly, fitwidth, nlam), dtype=float)
for j in range(nlam):
xkj_vec = np.tile(poly_centers[j,:].reshape(npoly, 1), (1, fitwidth))
xps_vec = np.tile(xps_mat[j,:].reshape(npoly, 1), (1, fitwidth))
xms_vec = xps_vec - 2*polyspacing
ip05_vec = np.tile(np.arange(fitwidth) + 0.5, (npoly, 1))
im05_vec = ip05_vec - 1
ind00 = ((ip05_vec <= xms_vec) + (im05_vec >= xps_vec))
ind11 = ((im05_vec > xkj_vec) * (True - ind00))
ind22 = ((ip05_vec < xkj_vec) * (True - ind00))
ind33 = (True - (ind00 + ind11 + ind22)).nonzero()
ind11 = ind11.nonzero()
ind22 = ind22.nonzero()
n_ind11 = len(ind11[0])
n_ind22 = len(ind22[0])
n_ind33 = len(ind33[0])
if n_ind11 > 0:
ind11_3d = ind11 + (np.ones(n_ind11, dtype=int)*j,)
lim2_ind11 = np.array((ip05_vec[ind11], xps_vec[ind11])).min(0)
Q[ind11_3d] = ((lim2_ind11 - im05_vec[ind11]) * invs * \
(polyspacing + xkj_vec[ind11] - 0.5*(im05_vec[ind11] + lim2_ind11)))
if n_ind22 > 0:
ind22_3d = ind22 + (np.ones(n_ind22, dtype=int)*j,)
lim1_ind22 = np.array((im05_vec[ind22], xms_vec[ind22])).max(0)
Q[ind22_3d] = ((ip05_vec[ind22] - lim1_ind22) * invs * \
(polyspacing - xkj_vec[ind22] + 0.5*(ip05_vec[ind22] + lim1_ind22)))
if n_ind33 > 0:
ind33_3d = ind33 + (np.ones(n_ind33, dtype=int)*j,)
lim1_ind33 = np.array((im05_vec[ind33], xms_vec[ind33])).max(0)
lim2_ind33 = np.array((ip05_vec[ind33], xps_vec[ind33])).min(0)
Q[ind33_3d] = ((lim2_ind33 - lim1_ind33) + invs * \
(xkj_vec[ind33] * (-xkj_vec[ind33] + lim1_ind33 + lim2_ind33) - 0.5*(lim1_ind33*lim1_ind33 + lim2_ind33*lim2_ind33)))
elif qmode=='brute': # Neither accurate, nor memory-frugal.
oversamp = 4.
jj2 = np.arange(nlam*oversamp, dtype=float) / oversamp
trace2 = np.interp(jj2, jj, trace)
poly_centers2 = trace2.reshape(nlam*oversamp, 1) + polyspacing * np.arange(-npoly/2+1, npoly/2+1, dtype=float) + constant
x2 = np.arange(fitwidth*oversamp, dtype=float)/oversamp
Q = np.zeros((npoly, fitwidth, nlam), dtype=float)
for k in range(npoly):
Q[k,:,:] = an.binarray((np.abs(x2.reshape(fitwidth*oversamp,1) - poly_centers2[:,k]) <= polyspacing), oversamp)
Q /= oversamp*oversamp*2
else: # 'slow' Loop-based nearest-neighbor mode: requires less memory
if verbose: tic = time()
Q = np.zeros((npoly, fitwidth, nlam), dtype=float)
for k in range(npoly):
for i in range(fitwidth):
for j in range(nlam):
Q[k,i,j] = max(0, min(polyspacing, 0.5*(polyspacing+1) - np.abs(poly_centers[j,k] - i)))
if verbose: print '%1.2f s to compute Q matrix (%s mode)' % (time() - tic, qmode)
# Some quick math to find out which dat columns are important, and
# which contain no useful spectral information:
Qmask = Q.sum(0).transpose() > 0
Qind = Qmask.transpose().nonzero()
Q_cols = [Qind[0].min(), Qind[0].max()]
nQ = len(Qind[0])
Qsm = Q[:,Q_cols[0]:Q_cols[1]+1,:]
# Prepar to iteratively clip outliers
newBadPixels = True
iter = -1
if verbose: print "Looking for bad pixel outliers."
while newBadPixels:
iter += 1
if verbose: print "Beginning iteration %i" % iter
# Compute pixel fractions (Marsh Eq. 5):
# (Note that values outside the desired polynomial region
# have Q=0, and so do not contribute to the fit)
#E = (skysubFrame / spectrum).transpose()
invEvariance = (spectrum**2 / variance).transpose()
weightedE = (skysubFrame * spectrum / variance).transpose() # E / var_E
invEvariance_subset = invEvariance[Q_cols[0]:Q_cols[1]+1,:]
# Define X vector (Marsh Eq. A3):
if verbose: tic = time()
X = np.zeros(N * npoly, dtype=float)
X0 = np.zeros(N * npoly, dtype=float)
for q in qq:
X[q] = (weightedE[Q_cols[0]:Q_cols[1]+1,:] * Qsm[kk[q],:,:] * jjnorm_pow[nn[q]]).sum()
if verbose: print '%1.2f s to compute X matrix' % (time() - tic)
# Define C matrix (Marsh Eq. A3)
if verbose: tic = time()
C = np.zeros((N * npoly, N*npoly), dtype=float)
buffer = 1.1 # C-matrix computation buffer (to be sure we don't miss any pixels)
for p in pp:
qp = Qsm[ll[p],:,:]
for q in qq:
# Check that we need to compute C:
if np.abs(kk[q] - ll[p]) <= (1./polyspacing + buffer):
if q>=p:
# Only compute over non-zero columns:
C[q, p] = (Qsm[kk[q],:,:] * qp * jjnorm_pow[nn[q]+mm[p]] * invEvariance_subset).sum()
if q>p:
C[p, q] = C[q, p]
if verbose: print '%1.2f s to compute C matrix' % (time() - tic)
##################################################
##################################################
# Just for reference; the following is easier to read, perhaps, than the optimized code:
if False: # The SLOW way to compute the X vector:
X2 = np.zeros(N * npoly, dtype=float)
for n in nn:
for k in kk:
q = N * k + n
xtot = 0.
for i in ii:
for j in jj:
xtot += E[i,j] * Q[k,i,j] * (jjnorm[j]**n) / Evariance[i,j]
X2[q] = xtot
# Compute *every* element of C (though most equal zero!)
C = np.zeros((N * npoly, N*npoly), dtype=float)
for p in pp:
for q in qq:
if q>=p:
C[q, p] = (Q[kk[q],:,:] * Q[ll[p],:,:] * (jjnorm.reshape(1,1,nlam)**(nn[q]+mm[p])) / Evariance).sum()
if q>p:
C[p, q] = C[q, p]
##################################################
##################################################
# Solve for the profile-polynomial coefficients (Marsh Eq. A)4:
if np.abs(np.linalg.det(C)) < 1e-10:
Bsoln = np.dot(np.linalg.pinv(C), X)
else:
Bsoln = np.linalg.solve(C, X)
Asoln = Bsoln.reshape(N, npoly).transpose()
# Define G_kj, the profile-defining polynomial profiles (Marsh Eq. 8)
Gsoln = np.zeros((npoly, nlam), dtype=float)
for n in range(npoly):
Gsoln[n] = np.polyval(Asoln[n,::-1], jjnorm) # reorder polynomial coef.
# Compute the profile (Marsh eq. 6) and normalize it:
if verbose: tic = time()
profile = np.zeros((fitwidth, nlam), dtype=float)
for i in range(fitwidth):
profile[i,:] = (Q[:,i,:] * Gsoln).sum(0)
#P = profile.copy() # for debugging
if profile.min() < 0:
profile[profile < 0] = 0.
profile /= profile.sum(0).reshape(1, nlam)
profile[True - np.isfinite(profile)] = 0.
if verbose: print '%1.2f s to compute profile' % (time() - tic)
#Step6: Revise variance estimates
modelSpectrum = spectrum * profile.transpose()
modelData = modelSpectrum + background
variance0 = np.abs(modelData) + readnoise**2
variance = variance0 / (goodpixelmask + 1e-9) # De-weight bad pixels, avoiding infinite variance
outlierVariances = (frame - modelData)**2/variance
if outlierVariances.max() > csigma**2:
newBadPixels = True
# Base our nreject-counting only on pixels within the spectral trace:
maxRejectedValue = max(csigma**2, np.sort(outlierVariances[Qmask])[-nreject])
worstOutliers = (outlierVariances>=maxRejectedValue).nonzero()
goodpixelmask[worstOutliers] = False
numberRejected = len(worstOutliers[0])
#pdb.set_trace()
else:
newBadPixels = False
numberRejected = 0
if verbose: print "Rejected %i pixels on this iteration " % numberRejected
# Optimal Spectral Extraction: (Horne, Step 8)
fixSkysubFrame = bfixpix(skysubFrame, True-goodpixelmask, n=8, retdat=True)
spectrum = np.zeros((nlam, 1), dtype=float)
#spectrum1 = np.zeros((nlam, 1), dtype=float)
varSpectrum = np.zeros((nlam, 1), dtype=float)
goodprof = profile.transpose() #* goodpixelmask
for ii in range(nlam):
thisrow_good = extractionApertures[ii] #* goodpixelmask[ii]
denom = (goodprof[ii, thisrow_good] * profile.transpose()[ii, thisrow_good] / variance0[ii, thisrow_good]).sum()
if denom==0:
spectrum[ii] = 0.
varSpectrum[ii] = 9e9
else:
spectrum[ii] = (goodprof[ii, thisrow_good] * skysubFrame[ii, thisrow_good] / variance0[ii, thisrow_good]).sum() / denom
#spectrum1[ii] = (goodprof[ii, thisrow_good] * modelSpectrum[ii, thisrow_good] / variance0[ii, thisrow_good]).sum() / denom
varSpectrum[ii] = goodprof[ii, thisrow_good].sum() / denom
#if spectrum.size==1218 and ii>610:
# pdb.set_trace()
#if spectrum.size==1218: pdb.set_trace()
ret = baseObject()
ret.spectrum = spectrum
ret.raw = standardSpectrum
ret.varSpectrum = varSpectrum
ret.trace = trace
ret.units = 'electrons'
ret.background = background_at_trace
ret.function_name = 'spec.superExtract'
if retall:
ret.profile_map = profile
ret.extractionApertures = extractionApertures
ret.background_map = background
ret.variance_map = variance0
ret.goodpixelmask = goodpixelmask
ret.function_args = args
ret.function_kw = kw
return ret
[docs]
def spextractor(frame, gain, readnoise, framevar=None, badpixelmask=None, mode='superExtract', trace=None, options=None, trace_options=None, verbose=False):
"""Extract a spectrum from a frame using one of several methods.
:INPUTS:
frame : 2D Numpy array or filename
Contains a single spectral trace.
gain : None or scalar
Gain of data contained in 'frame;' i.e., number of collected
photoelectrons equals frame * gain.
readnoise : None or scalar
Read noise of detector, in electrons.
framevar : None, 2D Numpy array, or filename
Variance of values in 'frame.'
If and only if framevar is None, use gain and readnoise to
compute variance.
badpixelmask : None, 2D Numpy array, or filename
Mask of bad pixels in 'frame.' Bad pixels are set to 1, good
pixels are set to 0.
mode : str
Which spectral extraction mode to use. Options are:
superExtract -- see :func:`superExtract`
optimalExtract -- see :func:`optimalExtract`
spline -- see :func:`extractSpectralProfiles`
Must also input trace_options.
trace : None, or 1D Numpy Array
Spectral trace location: fractional pixel index along the
entire spectral trace. If None, :func:`traceorders` will be
called using the options in 'trace_options.'
options : None or dict
Keyword options to be passed to the appropriate spectral
extraction algorithm. Note that you should be able to pass the
same sets of parameters to :func:`superExtract` and
:func:`optimalExtract` (the necessary parameter sets overlap,
but are not identical).
trace_options : None or dict
Keyword options to be passed to :func:`traceorders` (if no
trace is input, or if mode='spline')
:RETURNS:
spectrum, error or variance of spectrum, sky background, ...
:NOTES:
When 'optimalextract' is used: if len(bkg_radii)==2 then the
background apertures will be reset based on the median location
of the trace. If extract_radius is a singleton, it will be
similarly redefined.
:EXAMPLE:
::
frame = pyfits.getdata('spectral_filename.fits')
gain, readnoise = 3.3, 30
output = spec.spextractor(frame, gain, readnoise, mode='superExtract', \
options=dict(bord=2, bkg_radii=[20, 30], extract_radius=15, \
polyspacing=1./3, pord=5, verbose=True, trace=trace, \
qmode='slow-linear'))
output2 = spec.spextractor(frame, gain, readnoise, mode='optimalExtract', \
options=dict(bkg_radii=[20,30], extract_radius=15, bord=2, \
bsigma=3, pord=3, psigma=8, csigma=5, verbose=1))
"""
# 2012-09-03 11:12 IJMC: Created
from tools import array_or_filename
# Parse inputs:
frame = array_or_filename(frame)
framevar = array_or_filename(framevar, noneoutput=np.abs(frame) / gain + (readnoise / gain)**2)
if options is None:
options = dict(dispaxis=0)
if verbose and not options.has_key('verbose'):
options['verbose'] = verbose
if trace is None:
if trace_options is None:
trace_options = dict(pord=2)
if not trace_options.has_key('dispaxis'):
if options.has_key('dispaxis'): trace_options['dispaxis'] = options['dispaxis']
trace_coef = traceorders(frame, **trace_options)
trace = np.polyval(trace_coef.ravel(), np.arange(frame.shape[1-options['dispaxis']]))
options['trace'] = trace
##################################################
# Extract spectrum in one of several ways:
##################################################
if mode.lower()=='superextract':
ret = superExtract(frame, framevar, gain, readnoise, **options)
elif mode.lower()=='spline':
# First, set things up:
try:
trace_coef += 0.0
except:
trace_coef = np.polyfit(np.arange(trace.size), trace, trace_options['pord']).reshape(1, trace_options['pord']+1)
trace_options['retall'] = True
if not trace_options.has_key('bkg_radii'):
if options.has_key('bkg_radii') and len(options['bkg_radii'])==2:
trace_options['bkg_radii'] = options['bkg_radii']
if not trace_options.has_key('extract_radius'):
if options.has_key('extract_radius') and not hasattr(options['extract_radius'], '__iter__'):
trace_options['extract_radius'] = options['extract_radius']
prof = makeprofile(frame, trace_coef, **trace_options)
ret = extractSpectralProfiles(prof, **trace_options)
elif mode.lower()=='optimalextract':
# First, re-define bkg_radii and extract_radius (if necessary):
options = dict().update(options) # prevents alterations of options
if options.has_key('bkg_radii') and len(options['bkg_radii'])==2:
t0 = np.median(trace)
bkg_radii = [t0-options['bkg_radii'][1], t0-options['bkg_radii'][0],
t0+options['bkg_radii'][0], t0+options['bkg_radii'][1]]
options['bkg_radii'] = bkg_radii
if verbose: print "Re-defining background apertures: ", bkg_radii
if options.has_key('extract_radius') and \
(not hasattr(options['extract_radius'], '__iter__') or \
(len(options['extract_radius'])==1)):
extract_radius = [t0 - options['extract_radius'], \
t0 + options['extract_radius']]
options['extract_radius'] = extract_radius
if verbose: print "Re-defining extraction aperture: ", extract_radius
ret = optimalExtract(frame, framevar, gain, readnoise, **options)
else:
print "No valid spectral extraction mode specified!"
ret = -1
return ret
[docs]
def scaleSpectralSky_pca(frame, skyframes, variance=None, mask=None, badpixelmask=None, npca=3, gain=3.3, readnoise=30):
"""
Use PCA and blank sky frames to subtract
frame : str or NumPy array
data frame to subtract sky from. Assumed to be in ADU, not
electrons (see gain and readnoise)
npca : int
number of PCA components to remove
f0 = pyfits.getdata(odome.procsci[0])
mask = pyfits.getdata(odome._proc + 'skyframes_samplemask.fits').astype(bool)
badpixelmask = pyfits.getdata( odome.badpixelmask).astype(bool)
The simplest way to fit sky to a 'frame' containing bright
spectral is to include the spectral-trace regions in 'mask' but
set the 'variance' of those regions extremely high (to de-weight
them in the least-squares fit).
To use for multi-object data, consider running multiple times
(once per order)
Returns the best-fit sky frame as determined from the first 'npca'
PCA components.
"""
# 2012-09-17 16:41 IJMC: Created
from scipy import sparse
from pcsa import pca
# Parse inputs
if not isinstance(frame, np.ndarray):
frame = pyfits.getdata(frame)
if variance is None:
variance = np.abs(frame)/gain + (readnoise/gain)**2
else:
variance = np.array(variance, copy=False)
weights = 1./variance
nx, ny = frame.shape
n_tot = nx*ny
if badpixelmask is None:
badpixelmask = np.zeros((nx, ny), dtype=bool)
elif isinstance(badpixelmask, str):
badpixelmask = pyfits.getdata(badpixelmask).astype(bool)
else:
badpixelmask = np.array(badpixelmask).astype(bool)
weights[badpixelmask] = 0.
if mask is None:
mask = np.ones(frame.shape, dtype=bool)
n_elem = n_tot
else:
n_elem = mask.astype(bool).sum()
maskind = mask.nonzero()
if isinstance(skyframes, np.ndarray) and skyframes.ndim==3:
pass
else:
skyframes = np.array([pyfits.getdata(fn) for fn in skyframes])
skyframes_ind = np.array([sf[maskind] for sf in skyframes])
frame_ind = frame[maskind]
pcaframes = np.zeros((npca, n_elem), dtype=np.float32)
iter = 0
pcas = []
for jj in range(npca):
pcas.append(pca(skyframes_ind, None, ord=jj+1))
pcaframes[0] = pcas[0][0][0]
for jj in range(1, npca):
pcaframes[jj] = (pcas[jj][0] - pcas[jj-1][0])[0]
# del pcas
svecs = sparse.csr_matrix(pcaframes).transpose()
fitcoef, efitcoef = an.lsqsp(svecs, frame_ind, weights[maskind])
skyvalues = (fitcoef.reshape(npca, 1) * pcaframes).sum(0)
#eskyvalues = (np.diag(efitcoef).reshape(npca, 1) * pcaframes).sum(0)
skyframe = np.zeros((nx, ny), dtype=float)
skyframe[maskind] = skyvalues
return skyframe
[docs]
def scaleSpectralSky_dsa(subframe, variance=None, badpixelmask=None, nk=31, pord=1, nmed=3, gain=3.3, readnoise=30, dispaxis=0, spatial_index=None):
"""
Use difference-imaging techniques to subtract moderately tilted
sky background. Doesn't work so well!
subframe : NumPy array
data subframe containing sky data to be subtracted (and,
perhaps, an object's spectral trace). Assumed to be in ADU, not
electrons (see gain and readnoise)
variance : None or NumPy array
variance of each pixel in 'subframe'
nmed : int
size of 2D median filter
pord : int
degree of spectral tilt. Keep this number low!
nk : int
Number of kernel pixels in :func:`dia.dsa`
nmed : int
Size of window for 2D median filter (to reject bad pixels, etc.)
dispaxis : int
set dispersion axis: 0 = horizontal and 1 = vertical
gain, readnoise : ints
If 'variance' is None, these are used to estimate the uncertainties.
spatial_index : None, or 1D NumPy array of type bool
Which spatial rows (if dispaxis=0) to use when fitting the tilt
of sky lines across the spectrum. If you want to use all, set
to None. If you want to ignore some (e.g., because there's a
bright object's spectrum there) then set those rows' elements
of spatial_index to 'False'.
:NOTES:
Note that (in my experience!) this approach works better when
you set all weights to unity, rather than using the suggested
(photon + read noise) variances.
Returns the best-fit sky frame.
"""
# 2012-09-17 16:41 IJMC: Created
from scipy import signal
import dia
# Parse inputs
if not isinstance(subframe, np.ndarray):
subframe = pyfits.getdata(subframe)
if variance is None:
variance = np.abs(subframe)/gain + (readnoise/gain)**2
else:
variance = np.array(variance, copy=False)
weights = 1./variance
if badpixelmask is None:
badpixelmask = np.zeros(subframe.shape, dtype=bool)
elif isinstance(badpixelmask, str):
badpixelmask = pyfits.getdata(badpixelmask).astype(bool)
else:
badpixelmask = np.array(badpixelmask).astype(bool)
weights[badpixelmask] = 0.
if dispaxis==1:
subframe = subframe.transpose()
variance = variance.transpose()
weights = weights.transpose()
badpixelmask = badpixelmask.transpose()
sub = subframe
if nmed > 1:
ss = signal.medfilt2d(sub, nmed)
else:
ss = sub.copy()
ref = np.median(ss, axis=0)
n, nlam = ss.shape
if spatial_index is None:
spatial_index = np.arange(n)
else:
spatial_index = np.array(spatial_index, copy=False).nonzero()
gaussianpar = np.zeros((n, 4))
for ii in range(n):
test = dia.dsa(ref, ss[ii], nk, w=weights[ii], noback=True)
gaussianpar[ii] = fitGaussian(test[1])[0]
position_fit = an.polyfitr(np.arange(n)[spatial_index], gaussianpar[:,2][spatial_index], pord, 3)
positions = np.polyval(position_fit, np.arange(n))
width_fit = np.median(gaussianpar[:,1])
skyframe = 0*ss
testDC_spec = np.ones(nlam)
testx = np.arange(nk, dtype=float)
lsqfits = np.zeros((n,2))
for ii in range(n):
testgaussian = gaussian([1, width_fit, positions[ii], 0.], testx)
testgaussian /= testgaussian.sum()
testspectrum = dia.rconvolve1d(ref, testgaussian)
skyframe[ii] = testspectrum
if dispaxis==1:
skyframe = skyframe.transpose()
return skyframe
[docs]
def scaleSpectralSky_cor(subframe, badpixelmask=None, maxshift=20, fitwidth=2, pord=1, nmed=3, dispaxis=0, spatial_index=None, refpix=None, tord=2):
"""
Use cross-correlation to subtract tilted sky backgrounds.
subframe : NumPy array
data subframe containing sky data to be subtracted (and,
perhaps, an object's spectral trace).
badpixelmask : None or NumPy array
A boolean array, equal to zero for good pixels and unity for bad
pixels. If this is set, the first step will be a call to
:func:`nsdata.bfixpix` to interpolate over these values.
nmed : int
size of 2D median filter for pre-smoothing.
pord : int
degree of spectral tilt. Keep this number low!
maxshift : int
Maximum acceptable shift. NOT YET IMPLEMENTED!
fitwidth : int
Maximum radius (in pixels) for fitting to the peak of the
cross-correlation.
nmed : int
Size of window for 2D median filter (to reject bad pixels, etc.)
dispaxis : int
set dispersion axis: 0 = horizontal and 1 = vertical
spatial_index : None, or 1D NumPy array of type *bool*
Which spatial rows (if dispaxis=0) to use when fitting the tilt
of sky lines across the spectrum. If you want to use all, set
to None. If you want to ignore some (e.g., because there's a
bright object's spectrum there) then set those rows' elements
of spatial_index to 'False'.
refpix : scalar
Pixel along spatial axis to which spectral fits should be
aligned; if a spectral trace is present, one should set
"refpix" to the location of the trace.
tord : int
Order of polynomial fits along spatial direction in aligned
2D-spectral frame, to account for misalignments or
irregularities of tilt direction.
:RETURNS:
a model of the sky background, of the same shape as 'subframe.'
"""
# 2012-09-22 17:04 IJMC: Created
# 2012-12-27 09:53 IJMC: Edited to better account for sharp edges
# in backgrounds.
from scipy import signal
from nsdata import bfixpix
# Parse inputs
if not isinstance(subframe, np.ndarray):
subframe = pyfits.getdata(subframe)
if badpixelmask is None:
pass
else:
badpixelmask = np.array(badpixelmask).astype(bool)
subframe = bfixpix(subframe, badpixelmask, retdat=True)
if dispaxis==1:
subframe = subframe.transpose()
# Define necessary variables and vectors:
npix, nlam = subframe.shape
if spatial_index is None:
spatial_index = np.ones(npix, dtype=bool)
else:
spatial_index = np.array(spatial_index, copy=False)
if refpix is None:
refpix = npix/2.
lampix = np.arange(nlam)
tpix = np.arange(npix)
alllags = np.arange(nlam-maxshift*2) - np.floor(nlam/2 - maxshift)
# Median-filter the input data:
if nmed > 1:
ssub = signal.medfilt2d(subframe, nmed)
else:
ssub = subframe.copy()
ref = np.median(ssub, axis=0)
#allcor = np.zeros((npix, nlam-maxshift*2))
shift = np.zeros(npix, dtype=float)
for ii in tpix:
# Cross-correlate to measure alignment at each row:
cor = np.correlate(ref[maxshift:-maxshift], signal.medfilt(ssub[ii], nmed)[maxshift:-maxshift], mode='same')
# Measure offset of each row:
maxind = alllags[(cor==cor.max())].mean()
fitind = np.abs(alllags - maxind) <= fitwidth
quadfit = np.polyfit(alllags[fitind], cor[fitind], 2)
shift[ii] = -0.5 * quadfit[1] / quadfit[0]
shift_polyfit = an.polyfitr(tpix[spatial_index], shift[spatial_index], pord, 3) #, w=weights)
refpos = np.polyval(shift_polyfit, refpix)
#pdb.set_trace()
fitshift = np.polyval(shift_polyfit, tpix) - refpos
# Interpolate each row to a common frame to create an improved reference:
newssub = np.zeros((npix, nlam))
for ii in tpix:
newssub[ii] = np.interp(lampix, lampix+fitshift[ii], ssub[ii])
#pdb.set_trace()
newref = np.median(newssub[spatial_index,:], axis=0)
tfits = np.zeros((nlam, tord+1), dtype=float)
newssub2 = np.zeros((npix, nlam))
for jj in range(nlam):
tfits[jj] = an.polyfitr(tpix, newssub[:,jj], tord, 3)
newssub2[:, jj] = np.polyval(tfits[jj], tpix)
# Create the final model of the sky background:
skymodel = np.zeros((npix, nlam), dtype=float)
shiftmodel = np.zeros((npix, nlam), dtype=float)
for ii in tpix:
#skymodel[ii] = np.interp(lampix, lampix-fitshift[ii], newref)
skymodel[ii] = np.interp(lampix, lampix-fitshift[ii], newssub2[ii])
shiftmodel[ii] = np.interp(lampix, lampix+fitshift[ii], ssub[ii])
#pdb.set_trace()
if dispaxis==1:
skymodel = skymodel.transpose()
return skymodel, shiftmodel, newssub, newssub2
[docs]
def defringe_sinusoid(subframe, badpixelmask=None, nmed=5, dispaxis=0, spatial_index=None, period_limits=[20, 100], retall=False, gain=None, readnoise=None, bictest=False, sinonly=False):
"""
Use simple fitting to subtract fringes and sky background.
subframe : NumPy array
data subframe containing sky data to be subtracted (and,
perhaps, an object's spectral trace). Assumed to be in ADU, not
electrons (see gain and readnoise)
nmed : int
Size of window for 2D median filter (to reject bad pixels, etc.)
dispaxis : int
set dispersion axis: 0 = horizontal and 1 = vertical
spatial_index : None, or 1D NumPy array of type bool
Which spatial rows (if dispaxis=0) to use when fitting the tilt
of sky lines across the spectrum. If you want to use all, set
to None. If you want to ignore some (e.g., because there's a
bright object's spectrum there) then set those rows' elements
of spatial_index to 'False'.
period_limits : 2-sequence
Minimum and maximum periods (in pixels) of fringe signals to
accept as 'valid.' Resolution elements with best-fit periods
outside this range will only be fit by a linear trend.
gain : scalar
Gain of detector, in electrons per ADU (where 'subframe' is in
units of ADUs).
readnoise : scalar
Readnoise of detector, in electrons (where 'subframe' is in
units of ADUs).
bictest : bool
If True, use 'gain' and 'readnoise' to compute the Bayesian
Information Criterion (BIC) for each fit; a sinusoid is only
removed if BIC(sinusoid fit) is lower than BIC(constant fit).
sinonly : bool
If True, the output "model 2D spectrum" will only contain the
sinusoidal component. Otherwise, it will contain DC and
linear-trend terms.
:NOTES:
Note that (in my experience!) this approach works better when
you set all weights to unity, rather than using the suggested
(photon + read noise) variances.
:REQUIREMENTS:
:doc:`analysis` (for :func:`analysis.fmin`)
:doc:`phasecurves` (for :func:`phasecurves.errfunc`)
:doc:`lomb` (for Lomb-Scargle periodograms)
SciPy 'signal' module (for median-filtering)
Returns the best-fit sky frame.
"""
# 2012-09-19 16:22 IJMC: Created
# 2012-12-16 15:11 IJMC: Made some edits to fix bugs, based on outlier indexing.
from scipy import signal
import lomb
from analysis import fmin # scipy.optimize.fmin would also be O.K.
from phasecurves import errfunc
twopi = np.pi*2
fudgefactor = 1.5
def ripple(params, x): # Sinusoid + constant offset
if params[1]==0:
ret = 0.0
else:
ret = params[0] * np.cos(twopi*x/params[1] - params[2]) + params[3]
return ret
def linripple(params, x): # Sinusoid + linear trend
if params[1]==0:
ret = 0.0
else:
ret = params[0] * np.cos(twopi*x/params[1] - params[2]) + params[3] + params[4]*x
return ret
# Parse inputs
if not isinstance(subframe, np.ndarray):
subframe = pyfits.getdata(subframe)
if gain is None: gain = 1
if readnoise is None: readnoise = 1
if badpixelmask is None:
badpixelmask = np.zeros(subframe.shape, dtype=bool)
elif isinstance(badpixelmask, str):
badpixelmask = pyfits.getdata(badpixelmask).astype(bool)
else:
badpixelmask = np.array(badpixelmask).astype(bool)
if dispaxis==1:
subframe = subframe.transpose()
badpixelmask = badpixelmask.transpose()
sub = subframe
if nmed > 1:
sdat = signal.medfilt2d(sub, nmed)
else:
sdat = sub.copy()
if bictest:
var_sdat = np.abs(sdat)/gain + (readnoise/gain)**2
npix, nlam = sdat.shape
if spatial_index is None:
spatial_index = np.ones(npix, dtype=bool)
else:
spatial_index = np.array(spatial_index, copy=False).astype(bool)
##############################
periods = np.logspace(0.5, np.log10(npix), npix*2)
x = np.arange(npix)
allfits = np.zeros((nlam, 5))
for jj in range(nlam):
vec = sdat[:, jj].copy()
this_index = spatial_index * (True - badpixelmask[:, jj])
#if jj==402: pdb.set_trace()
if this_index.sum() > 1:
# LinFit the data; exclude values inconsistent with a sinusoid:
linfit = an.polyfitr(x[this_index], vec[this_index], 1, 3)
linmodel = (x) * linfit[0] + linfit[1]
vec2 = vec - (linmodel - linfit[1])
maxexcursion = an.dumbconf(vec2[this_index], .683)[0] * (fudgefactor / .88)
index = this_index * (np.abs(vec2 - np.median(vec2[this_index])) <= (6*maxexcursion))
# Use Lomb-Scargle to find strongest period:
#lsp = lomb.lomb(vec2[index], x[index], twopi/periods)
freqs, lsp = lomb.fasper(x[index], vec2[index], 12., 0.5)
guess_period = (1./freqs[lsp==lsp.max()]).mean()
# If the best-fit period is within our limits, fit it:
if (guess_period <= period_limits[1]) and (guess_period >= period_limits[0]):
#periods2 = np.arange(guess_period-1, guess_period+1, 0.02)
#lsp2 = lomb.lomb(vec2[index], x[index], twopi/periods2)
#guess_period = periods2[lsp2==lsp2.max()].mean()
guess_dc = np.median(vec2[index])
guess_amp = an.dumbconf(vec2[index], .683, mid=guess_dc)[0] / 0.88
guess = [guess_amp, guess_period, np.pi, guess_dc]
if bictest:
w = 1./var_sdat[:,jj][index]
else:
w = np.ones(index.sum())
fit = fmin(errfunc, guess, args=(ripple, x[index], vec2[index], w), full_output=True, disp=False)
guess2 = np.concatenate((fit[0], [linfit[0]]))
fit2 = fmin(errfunc, guess2, args=(linripple, x[index], vec[index], w), full_output=True, disp=False)
if bictest:
ripple_model = linripple(fit2[0], x)
bic_ripple = (w*(vec - ripple_model)[index]**2).sum() + 6*np.log(index.sum())
bic_linfit = (w*(vec - linmodel)[index]**2).sum() + 2*np.log(index.sum())
#if jj==546:
# pdb.set_trace()
if bic_ripple >= bic_linfit:
fit2 = [[0, 1e9, 0, linfit[1], linfit[0]], 0]
else: # Clearly not a valid ripple -- just use the linear fit.
fit2 = [[0, 1e11, 0, linfit[1], linfit[0]], 0]
else: # *NO* good pixels!
#linfit = an.polyfitr(x[spatial_index], vec[spatial_index], 1, 3)
fit2 = [[0, 1e11, 0, np.median(vec[spatial_index]), 0], 0]
allfits[jj] = fit2[0]
# Median filter the resulting coefficients
if nmed > 1: # No median-filtering
newfits = np.array([signal.medfilt(fits, nmed) for fits in allfits.transpose()]).transpose()
newfits[:, 2] = allfits[:,2] # Don't smooth phase
else:
newfits = allfits
# Generate the model sky pattern
skymodel = np.zeros((npix, nlam))
for jj in range(nlam):
if sinonly:
coef = newfits[jj].copy()
coef[3:] = 0.
else:
coef = newfits[jj]
skymodel[:, jj] = linripple(coef, x)
if dispaxis==1:
skymodel = skymodel.transpose()
if retall:
ret = skymodel, allfits
else:
ret = skymodel
return ret
#
[docs]
def makexflat(subreg, xord, nsigma=3, minsnr=10, minfrac=0.5, niter=1):
"""Helper function for XXXX.
:INPUTS:
subreg : 2D NumPy array
spectral subregion, containing spectral background, sky,
and/or target flux measurements.
xord : scalar or sequence
Order of polynomial by which each ROW will be normalized. If
niter>0, xord can be a sequence of length (niter+1). A good
approach for, e.g., spectral dome flats is to set niter=1 and
xord=[15,2].
nsigma : scalar
Sigma-clipping level for calculation of column-by-column S/N
minsnr : scalar
Minimum S/N value to use when selecting 'good' columns for
normalization.
minfrac : scalar, 0 < minfrac < 1
Fraction of columns to use, selected by highest S/N, when
selecting 'good' columns for normalization.
niter : int
Number of iterations. If set to zero, do not iterate (i.e.,
run precisely once through.)
:NOTES:
Helper function for functions XXXX
"""
# 2013-01-20 14:20 IJMC: Created
ny, nx = subreg.shape
xall = np.arange(2048.)
subreg_new = subreg.copy()
iter = 0
if not hasattr(xord, '__iter__'): xord = [xord]*(niter+1)
while iter <= niter:
snr = an.snr(subreg_new, axis=0, nsigma=nsigma)
ind = (snr > np.sort(snr)[-int(minfrac*snr.size)]) * (snr > minsnr)
xxx = ind.nonzero()[0]
norm_subreg = subreg[:,ind] / np.median(subreg[:,ind], 0)
coefs = np.array([an.polyfitr(xxx, row, xord[iter], 3) for row in norm_subreg])
xflat = np.array([np.polyval(coef0, xall) for coef0 in coefs])
iter += 1
subreg_new = subreg / xflat
return xflat
[docs]
def make_spectral_flats(sky, domeflat, subreg_corners, badpixelmask=None, xord_pix=[15,2], xord_sky=[2,1], yord=2, minsnr=5, minfrac_pix=0.7, minfrac_sky=0.5, locs=None, nsigma=3):
"""
Construct appropriate corrective frames for multi-object
spectrograph data. Specifically: corrections for irregular slit
widths, and pixel-by-pixel detector sensitivity variations.
sky : 2D NumPy array
Master spectral sky frame, e.g. from median-stacking many sky
frames or masking-and-stacking dithered science spectra frames.
This frame is used to construct a map to correct science frames
(taken with the identical slit mask!) for irregular sky
backgrounds resulting from non-uniform slit widths (e.g.,
Keck/MOSFIRE).
Note that the dispersion direction should be 'horizontal'
(i.e., parallel to rows) in this frames.
domeflat : 2D NumPy array
Master dome spectral flat, e.g. from median-stacking many dome
spectra. This need not be normalized in the dispersion or
spatial directions. This frame is used to construct a flat map
of the pixel-to-pixel variations in detector sensitivity.
Note that the dispersion direction should be 'horizontal'
(i.e., parallel to rows) in this frames.
subreg_corners : sequence of 2- or 4-sequences
Indices (or merely starting- and ending-rows) for each
subregion of interest in 'sky' and 'domeflat' frames. Syntax
should be that of :func:`tools.extractSubregion`, or such that
subreg=sky[subreg_corners[0]:subreg_corners[1]]
badpixelmask : 2D NumPy array, or str
Nonzero for any bad pixels; these will be interpolated over
using :func:`nsdata.bfixpix`.
xord_pix : sequence
Polynomial orders for normalization in dispersion direction of
pixel-based flat (dome flat) on successive iterations; see
:func:`makexflat`.
xord_sky : sequence
Polynomial orders for normalization in dispersion direction of
slit-based correction (sky frame) on successive iterations;
see :func:`makexflat`.
yord : scalar
Polynomial order for normalization of pixel-based (dome) flat
in spatial direction.
locs : None, or sequence
Row-index in each subregion of the location of the
spectral-trace-of-interest. Only used for rectifying of sky
frame; leaving this at None will have at most a mild
effect. If not None, should be the same length as
subreg_corners. If subreg_corners[0] = [800, 950] then
locs[0] might be set to, e.g., 75 if the trace lies in the
middle of the subregion.
:RETURNS:
wideslit_skyflat, narrowslit_domeflat
:EXAMPLE:
::
try:
from astropy.io import fits as pyfits
except:
import pyfits
import spec
import ir
obs = ir.initobs('20121010')
sky = pyfits.getdata(obs._raw + 'masktersky.fits')
domeflat = pyfits.getdata(obs._raw + 'mudflat.fits')
allinds = [[7, 294], [310, 518], [532, 694], [710, 960], [976, 1360], [1378, 1673], [1689, 2022]]
locs = [221, 77, 53, 88, 201, 96, 194]
skycorrect, pixcorrect = spec.make_spectral_flats(sky, domeflat, allinds, obs.badpixelmask, locs=locs)
"""
# 2013-01-20 14:40 IJMC: Created
from tools import extractSubregion
from nsdata import bfixpix
# Check inputs:
if not isinstance(sky, np.ndarray):
sky = pyfits.getdata(sky)
if not isinstance(domeflat, np.ndarray):
domeflat = pyfits.getdata(domeflat)
ny0, nx0 = sky.shape
if badpixelmask is None:
badpixelmask = np.zeros((ny0, nx0), dtype=bool)
if not isinstance(badpixelmask, np.ndarray):
badpixelmask = pyfits.getdata(badpixelmask)
# Correct any bad pixels:
if badpixelmask.any():
bfixpix(sky, badpixelmask)
bfixpix(domeflat, badpixelmask)
wideslit_skyflat = np.ones((ny0, nx0), dtype=float)
narrowslit_domeflat = np.ones((ny0, nx0), dtype=float)
# Loop through all subregions specified:
for jj in range(len(subreg_corners)):
# Define extraction & alignment indices:
specinds = np.array(subreg_corners[jj]).ravel().copy()
if len(specinds)==2:
specinds = np.concatenate(([0, nx0], specinds))
if locs is None:
loc = np.mean(specinds[2:4])
else:
loc = locs[jj]
skysub = extractSubregion(sky, specinds, retall=False)
domeflatsub = extractSubregion(domeflat, specinds, retall=False)
badsub = extractSubregion(badpixelmask, specinds, retall=False)
ny, nx = skysub.shape
yall = np.arange(ny)
# Normalize Dome Spectral Flat in X-direction (rows):
xflat_dome = makexflat(domeflatsub, xord_pix, minsnr=minsnr, minfrac=minfrac_pix, niter=len(xord_pix)-1, nsigma=nsigma)
ymap = domeflatsub*0.0
xsubflat = domeflatsub / np.median(domeflatsub, 0) / xflat_dome
# Normalize Dome Spectral Flat in Y-direction (columns):
ycoefs = [an.polyfitr(yall, xsubflat[:,ii], yord, nsigma) for ii in range(nx)]
yflat = np.array([np.polyval(ycoef0, yall) for ycoef0 in ycoefs]).transpose()
pixflat = domeflatsub / (xflat_dome * yflat * np.median(domeflatsub, 0))
bogus = (pixflat< 0.02) + (pixflat > 50)
pixflat[bogus] = 1.
# Normalize Sky spectral Flat in X-direction (rows):
askysub = scaleSpectralSky_cor(skysub/pixflat, badsub, pord=2, refpix=loc, nmed=1)
xflat_sky = makexflat(askysub[1], xord_sky, minsnr=minsnr, minfrac=minfrac_sky, niter=len(xord_sky)-1, nsigma=nsigma)
wideslit_skyflat[specinds[2]:specinds[3], specinds[0]:specinds[1]] = xflat_sky
narrowslit_domeflat[specinds[2]:specinds[3], specinds[0]:specinds[1]] = pixflat
print "Done with subregion %i" % (jj+1)
return wideslit_skyflat, narrowslit_domeflat
[docs]
def calibrate_stared_mosfire_spectra(scifn, outfn, skycorrect, pixcorrect, subreg_corners, **kw):
"""Correct non-dithered WIDE-slit MOSFIRE spectral frames for:
pixel-to-pixel nonuniformities (i.e., traditional flat-fielding)
detector nonlinearities
tilted spectral lines
non-uniform slit widths (which cause non-smooth backgrounds)
Note that the dispersion direction should be 'horizontal'
(i.e., parallel to rows) in this frame.
:INPUTS:
scifn : str
Filename of raw, uncalibrated science frame (in ADUs, not electrons)
outfn : str
Name into which final, calibrated file should be written.
skycorrect : str or 2D NumPy array
Slitloss correction map (i.e., for slits of nonuniform width),
such as generated by :func:`make_spectral_flats`.
pixcorrect : str or 2D NumPy array
Pixel-by-pixel sensitivity correction map (i.e., flat field),
such as generated by :func:`make_spectral_flats`.
subreg_corners : sequence of 2- or 4-sequences
Indices (or merely starting- and ending-rows) for each
subregion of interest in 'sky' and 'domeflat' frames. Syntax
should be that of :func:`tools.extractSubregion`, or such that
subreg=sky[subreg_corners[0]:subreg_corners[1]]
linearize : bool
Whether to linearity-correct the data.
If linearizing: linearity correction is computed & applied
AFTER applying the pixel-by-pixel (flatfield) correction, but
BEFORE the slitloss (sky) correction.
bkg_radii : 2-sequence
Inner and outer radius for background computation and removal;
measured in pixels from the center of the profile. The values
[11,52] seems to work well for MOSFIRE K-band spectra.
locs : None, or sequence
Row-index in each subregion of the location of the
spectral-trace-of-interest. Only used for rectifying of sky
frame; leaving this at None will have at most a mild
effect. If not None, should be the same length as
subreg_corners. If subreg_corners[0] = [800, 950] then
locs[0] might be set to, e.g., 75 if the trace lies in the
middle of the subregion.
gain: scalar
Detector gain, in electrons per ADU
readnoise: scalar
Detector readnoise, in electrons
polycoef : str, None, or sequence
Polynomial coefficients for detector linearization: see
:func:`ir.linearity_correct` and :func:`ir.linearity_mosfire`.
unnormalized_flat : str or 2D NumPy array
If not 'None', this is used to define the subregion header
keywords ('subreg0', 'subreg1', etc.) for each slit's
two-dimensional spectrum. These keywords are required by much
of my extraction machinery!
:EXAMPLE:
::
try:
from astropy.io import fits as pyfits
except:
import pyfits
import spec
import ir
obs = ir.initobs('20121010')
allinds = [[7, 294], [310, 518], [532, 694], [710, 960], [976, 1360], [1378, 1673], [1689, 2022]]
locs = [221, 77, 53, 88, 201, 96, 194]
unnorm_flat_fn = obs._raw + 'mudflat.fits'
if False:
sky = pyfits.getdata(obs._raw + 'masktersky.fits')
unnorm_domeflat = pyfits.getdata(unnorm_flat_fn)
skycorrect, pixcorrect = spec.make_spectral_flats(sky, unnorm_domeflat, allinds, badpixelmask=obs.badpixelmask, locs=locs)
else:
skycorrect = obs._raw + 'skycorrect.fits'
pixcorrect = obs._raw + 'pixcorrect.fits'
linearcoef='/Users/ianc/proj/transit/data/mosfire_linearity/linearity/mosfire_linearity_cnl_coefficients_bic-optimized.fits'
rawfn = obs.rawsci
outfn = obs.procsci
spec.calibrate_stared_mosfire_spectra(rawfn, outfn, skycorrect, pixcorrect, allinds, linearize=True, badpixelmask=obs.badpixelmask, locs=locs, polycoef=linearcoef, verbose=True, clobber=True, unnormalized_flat=unnorm_flat_fn)
:NOTES:
This routine is *slow*, mostly because of the call to
:func:`defringe_sinusoid`. Consider running multiple processes
in parallel, to speed things up!
:SEE_ALSO:
:func:`ir.mosfire_speccal`, :func:`defringe_sinusoid`
"""
# 2013-01-20 16:17 IJMC: Created.
# 2013-01-25 15:59 IJMC: Fixed various bugs relating to
# determinations of subregion boundaries.
try:
from astropy.io import fits as pyfits
except:
import pyfits
from nsdata import bfixpix
import ir
import os
from tools import extractSubregion, findRectangles
import sys
#pdb.set_trace()
if not isinstance(scifn, str) and hasattr(scifn, '__iter__'):
for scifn0, outfn0 in zip(scifn, outfn):
calibrate_stared_mosfire_spectra(scifn0, outfn0, skycorrect, pixcorrect, subreg_corners, **kw)
else:
# Set defaults:
names = ['linearize', 'bkg_radii', 'gain', 'readnoise', 'polycoef', 'verbose', 'badpixelmask', 'locs', 'clobber', 'unnormalized_flat']
defaults = [False, [11, 52], 2.15, 21.1, None, False, None, None, False, None]
for n,d in zip(names, defaults):
exec('%s = kw["%s"] if kw.has_key("%s") else d' % (n, n, n))
# Check inputs:
scihdr = pyfits.getheader(scifn)
sci = pyfits.getdata(scifn)
exptime = scihdr['TRUITIME']
ncoadd = scihdr['COADDONE']
nread = scihdr['READDONE']
if not isinstance(pixcorrect, np.ndarray):
scihdr.update('SLITFLAT', pixcorrect)
pixcorrect = pyfits.getdata(pixcorrect)
else:
scihdr.update('PIXFLAT', 'User-specified data array')
if not isinstance(skycorrect, np.ndarray):
scihdr.update('SLITFLAT', skycorrect)
skycorrect = pyfits.getdata(skycorrect)
else:
scihdr.update('SLITFLAT', 'User-specified data array')
if unnormalized_flat is not None:
if not isinstance(unnormalized_flat, np.ndarray):
scihdr.update('MUDFLAT', unnormalized_flat)
unnormalized_flat = pyfits.getdata(unnormalized_flat)
else:
scihdr.update('MUDFLAT', 'User-specified data array')
ny0, nx0 = sci.shape
if badpixelmask is None:
badpixelmask = np.zeros((ny0, nx0), dtype=bool)
if not isinstance(badpixelmask, np.ndarray):
badpixelmask = pyfits.getdata(badpixelmask)
# Correct any bad pixels:
if badpixelmask.any():
bfixpix(sci, badpixelmask)
if linearize:
# If you edit this section, be sure to update the same
# section in ir.mosfire_speccal!
fmap = ir.makefmap_mosfire()
sci_lin, liniter = ir.linearity_correct(sci/pixcorrect, nread-1, ncoadd, 1.45, exptime, fmap, ir.linearity_mosfire, verbose=verbose, linfuncargs=dict(polycoef=polycoef), retall=True)
ir.hdradd(scihdr, 'COMMENT', 'Linearity-corrected by calibrate_stared_mosfire_spectra.')
ir.hdradd(scihdr, 'LIN_ITER', liniter)
if isinstance(polycoef, str):
ir.hdradd(scihdr, 'LINPOLY', os.path.split(polycoef)[1])
else:
if isinstance(polycoef, np.ndarray) and polycoef.ndim==3:
ir.hdradd(scihdr, 'LINPOLY', 'user-input array used for linearity correction.')
else:
ir.hdradd(scihdr, 'LINPOLY', str(polycoef))
else:
sci_lin = sci/pixcorrect
ir.hdradd(scihdr, 'COMMENT', 'NOT linearity-corrected.')
newframe = np.zeros((ny0, nx0), dtype=float)
# Loop through all subregions specified:
for jj in range(len(subreg_corners)):
# Define extraction & alignment indices:
specinds = np.array(subreg_corners[jj]).ravel().copy()
if len(specinds)==2:
specinds = np.concatenate(([0, nx0], specinds))
if locs is None:
loc = np.mean(specinds[2:4])
else:
loc = locs[jj]
# Prepare various necessities:
scisub = extractSubregion(sci, specinds, retall=False)
badsub = extractSubregion(badpixelmask, specinds, retall=False)
slitflat = extractSubregion(skycorrect, specinds, retall=False)
ny, nx = scisub.shape
yall = np.arange(ny)
# Define subregion boundaries from the flat-field frame.
# Ideally, this should go *outside* the loop (no need to
# re-calculate this for every file...)
if unnormalized_flat is not None:
samplemask, boundaries = ir.spectral_subregion_mask(unnormalized_flat)
samplemask *= (unnormalized_flat > 500)
samplemask_corners = findRectangles(samplemask, minsepy=42, minsepx=150)
nspec = samplemask_corners.shape[0]
for kk in range(nspec):
ir.hdradd(scihdr, 'SUBREG%i' % kk, str(samplemask_corners[kk]))
ir.hdradd(scihdr, 'NSUBREG', nspec)
# Align the subregion, so that sky lines run along detector columns:
aligned_scisub = scaleSpectralSky_cor(scisub/slitflat, badsub, pord=2, refpix=loc, nmed=1)
# Model the sky background, using linear trends plus a sinusoid:
if locs is None:
spatind = np.ones(ny, dtype=bool)
else:
spatind = (np.abs(np.arange(ny) - loc) > bkg_radii[0]) * (np.abs(np.arange(ny) - loc) < bkg_radii[1])
sscisub = defringe_sinusoid(aligned_scisub[1], badpixelmask=badsub, period_limits=[9,30], gain=gain, readnoise=readnoise, bictest=False, spatial_index=spatind, nmed=1)
# Compute the corrected subregion:
aligned_flattened_skycorrected_subregion = \
(aligned_scisub[1] - sscisub) * slitflat + aligned_scisub[3]
#try2 = (aligned_scisub[1] - sscisub) * slitflat + aligned_scisub
newframe[specinds[2]:specinds[3],specinds[0]:specinds[1]] = aligned_flattened_skycorrected_subregion
print "Done with subregion %i" % (jj+1)
scihdr.update('MOSCAL', 'Calibrated by calibrate_stared_mosfire_spectra')
pyfits.writeto(outfn, newframe.astype(float), scihdr, clobber=clobber)
return
[docs]
def reconstitute_gmos_roi(infile, outfile, **kw):
"""Convert GMOS frames taken with custom ROIs into standard FITS frames.
:INPUTS:
in : string or sequence of strings
Input filename or filenames.
out : string or sequence of strings
Output filename or filenames.
:OPTIONS:
clobber : bool
Passed to PyFITS; whether to overwrite existing files.
"""
# 2013-03-28 21:37 IJMC: Created at the summit of Mauna Kea
import ir
def trimlims(instr):
trimchars = '[]:,'
for trimchar in trimchars:
instr = instr.replace(trimchar, ' ')
return map(float, instr.strip().split())
# Handle recursive case:
if not isinstance(infile, str):
for thisin, thisout in zip(infile, outfile):
reconstitute_gmos_roi(thisin, thisout, **kw)
return
# Handle non-recursive (single-file) case:
file = pyfits.open(infile)
hdr = file[0].header
nroi = hdr['detnroi']
biny, binx = map(int, file[1].header['ccdsum'].split())
detsize = trimlims(hdr['detsize'])
binning = 2
frame = np.zeros(((detsize[3]-detsize[2]+1)/binx, (detsize[1]-detsize[0]+1)/biny), dtype=int)
print frame.shape
# Read in the coordinates for each ROI:
roi_xys = np.zeros((nroi, 4), dtype=int)
for ii in range(1, nroi+1):
roi_xys[ii-1, 0] = (hdr['detro%ix' % ii] - 1)
roi_xys[ii-1, 1] = (hdr['detro%ix' % ii] + hdr['detro%ixs' % ii] - 1)
roi_xys[ii-1, 2] = (hdr['detro%iy' % ii] - 1) /binx
roi_xys[ii-1, 3] = (hdr['detro%iy' % ii] - 1)/binx + hdr['detro%iys' % ii] - 1
# Read in the C
nhdus = len(file) - 1
detsecs = np.zeros((nhdus, 4), dtype=int)
datasecs = np.zeros((nhdus, 4), dtype=int)
biassecs = np.zeros((nhdus, 4), dtype=int)
for ii in range(nhdus):
detsecs[ii] = trimlims(file[ii+1].header['detsec'])
datasecs[ii] = trimlims(file[ii+1].header['datasec'])
biassecs[ii] = trimlims(file[ii+1].header['biassec'])
di2 = (detsecs[ii,0]-1)/biny, detsecs[ii,1]/biny, (detsecs[ii,2]-1)/binx, detsecs[ii,3]/binx
thisdat = file[ii+1].data
#pdb.set_trace()
if biassecs[ii,0]==1025:
frame[di2[2]:di2[3], di2[0]:di2[1]] = thisdat[:, :-32]
elif biassecs[ii,0]==1:
frame[di2[2]:di2[3], di2[0]:di2[1]] = thisdat[:, 32:]
else:
print 'bombed. bummer! I should have written a better function'
file.close()
hdr = pyfits.getheader(infile)
for ii in range(nroi):
ir.hdradd(hdr, 'SUBREG%i' % ii, '[%i %i %i %i]' % tuple(roi_xys[ii]))
pyfits.writeto(outfile, frame, hdr, **kw)
return #detsecs, datasecs, biassecs
[docs]
def rotationalProfile(delta_epsilon, delta_lam):
"""Compute the rotational profile of a star, assuming solid-body
rotation and linear limb darkening.
This uses Eq. 18.14 of Gray's Photospheres, 2005, 3rd Edition.
:INPUTS:
delta_epsilon : 2-sequence
[0] : delta_Lambda_L = lambda * V * sin(i)/c; the rotational
displacement at the stellar limb.
[1] : epsilon, the linear limb darkening coefficient, used in
the relation I(theta) = I0 + epsilon * (cos(theta) - 1).
[2] : OPTIONAL! The central location of the profile (otherwise
assumed to be located at delta_lam=0).
delta_lam : scalar or sequence
Wavelength minus offset: Lambda minus lambda_0. Grid upon
which computations will be done.
:EXAMPLE:
::
import pylab as py
import spec
dlam = py.np.linspace(-2, 2, 200) # Create wavelength grid
profile = spec.rotationalProfile([1, 0.6], dlam)
py.figure()
py.plot(dlam, profile)
"""
# 2013-05-26 10:37 IJMC: Created.
delta_lambda_L, epsilon = delta_epsilon[0:2]
if len(delta_epsilon)>2: # optional lambda_offset
lamdel2 = 1. - ((delta_lam - delta_epsilon[2])/delta_lambda_L)**2
else:
lamdel2 = 1. - (delta_lam/delta_lambda_L)**2
if not hasattr(delta_lam, '__iter__'):
delta_lam = np.array([delta_lam])
ret = (4*(1.-epsilon) * np.sqrt(lamdel2) + np.pi*epsilon*lamdel2) / \
(2*np.pi * delta_lambda_L * (1. - epsilon/3.))
ret[lamdel2<0] = 0.
return ret
[docs]
def modelline(param, prof2, dv):
"""Generate a rotational profile, convolve it with a second input
profile, normalize it (simply), and return.
:INPUTS:
param : 1D sequence
param[0:3] -- see :func:`rotationalProfile`
param[4] -- multiplicative scaling factor
dv : velocity grid
prof2 : second input profile
"""
# 2013-08-07 09:55 IJMC: Created
nk = dv.size
rot_model = rotationalProfile(param[0:3], dv)
conv_rot_model = np.convolve(rot_model, prof2, 'same')
norm_conv_rot_model = conv_rot_model - (conv_rot_model[0] + (conv_rot_model[-1] - conv_rot_model[0])/(nk - 1.) * np.arange(nk))
return norm_conv_rot_model * param[3]
Reference in New Issue View Git Blame Copy Permalink