def _calculateSkyCoords(self):
"""
Calculates sky coordinates for the slits.
Resamples the direct image pixels back to the
slit image pixels. Takes the average in x-direction
which is assumed to be across the slit, so that
the sky coordinates are at the centre of the slit
in this direction.
"""
#calculate the RA and DEC of the centre of the centre slit
centre = np.array([[self.result['xcenter'], self.result['ycenter']], ], np.float_)
sky = self.result['WCS'].wcs_pix2sky(centre, 1)
self.result['RA'] = sky[0][0]
self.result['DEC'] = sky[0][1]
#each slit pixels
for key, value in self.slits.items():
ymin = value['ymin'] + self.result['y']
ymax = value['ymax'] + self.result['y']
y = np.arange(ymin, ymax)
#y is now sampled to the supersampled direct image
#what we ultimately need is the coordinates
#for slit pixels...
resample = value['pixels']
ybin = m.frebin(y, resample)
#for x, given that it is the slit width
#we need the mean value
xmin = value['xmin'] + self.result['x']
xmax = value['xmax'] + self.result['x']
mean = np.mean(np.array([xmin, xmax]))
#x = np.zeros(len(y)) + mean
xbin = np.zeros(len(ybin)) + mean
pixels = []
for a, b in zip(xbin, ybin):
pixels.append([a, b])
pixels = np.asanyarray(pixels)
sky = self.result['WCS'].wcs_pix2sky(pixels, 1)
#sky = self.result['WCS'].toworld(pixels)
self.slits[key]['coordinates'] = sky
#record x and y in the slit frame
self.slits[key]['coordinatesXY'] = pixels
self.slits[key]['coordinatesX'] = np.ones(value['pixels'])
self.slits[key]['coordinatesY'] = np.arange(value['pixels']) + 1
#RA and DEC of the centre of the slit
centre = np.array([[mean, (ymax + ymin) / 2.], ])
sky = self.result['WCS'].wcs_pix2sky(centre, 1)
self.slits[key]['RA'] = sky[0][0]
self.slits[key]['DEC'] = sky[0][1]
if self.debug:
print self.slits[key]['RA'], self.slits[key]['DEC']
def _calculateDifference(self):
"""
Calculates the difference between the derived observed and SDSS spectra.
Interpolates the SDSS spectra to the same wavelength scale.
.. Warning:: We do not conserve flux here when interpolating. This is because
we do not actually interpolate flux but flux density which is per
angstrom.
"""
ms = (self.fitting['SDSSwave'] >= np.min(self.fitting['obsWavelengths'])) &\
(self.fitting['SDSSwave'] <= np.max(self.fitting['obsWavelengths']))
newflux = m.frebin(self.fitting['SDSSflux'][ms],
len(self.fitting['obsSpectraConvolved']))#, total=True)
self.fitting['spectraRatio'] = self.fitting['obsSpectraConvolved'] / newflux
self.fitting['interpFlux'] = newflux
def _calculateDifference(self):
"""
Calculates the difference between the derived observed and SDSS spectra.
Interpolates the SDSS spectra to the same wavelength scale.
.. Warning:: We do not conserve flux here when interpolating. This is because
we do not actually interpolate flux but flux density which is per
angstrom.
"""
ms = (self.fitting['SDSSwave'] >= np.min(self.fitting['obsWavelengths'])) &\
(self.fitting['SDSSwave'] <= np.max(self.fitting['obsWavelengths']))
newflux = m.frebin(
self.fitting['SDSSflux'][ms],
len(self.fitting['obsSpectraConvolved'])) #, total=True)
self.fitting[
'spectraRatio'] = self.fitting['obsSpectraConvolved'] / newflux
self.fitting['interpFlux'] = newflux
def hrebin(imagefile, newx, newy, output='rebinned.fits', ext=0, total=False):
"""
Expand or contract a FITS image and update the header.
Based on IDL routine hrebin.pro, but removed some functionality
:todo: remove the pywcs dependency
:param imagefile: name of the FITS file to be rebinned
:param newx: size of the new image in the X direction, integer scalar
:param newy: size of the new image in the Y direction, integer scalar
:param output: Name of the outputfile or None
:param ext: the FITS extension of the header and data
:return: None or updated img and hdr
"""
#todo: remove the pywcs dependency
#todo: add a routine for exact rebinning, now done with the same
fh = pf.open(imagefile)
oldhdr = fh[ext].header
oldimg = fh[ext].data
fh.close()
wcs = pywcs.WCS(oldhdr)
#old size
#xsize = oldhdr['NAXIS1']
#ysize = oldhdr['NAXIS2']
ysize, xsize = oldimg.shape
#ratios
xratio = newx / float(xsize)
yratio = newy / float(ysize)
#change in the aspect ratio
lambd = yratio / xratio
#Ratio of pixel areas
pix_ratio = xratio * yratio
#chech whether the new size is an exact match
exact = (xsize % newx == 0) | (newx % xsize == 0) and\
(ysize % newy == 0) | (newy % ysize == 0)
#rebin based on whether the rebinning is exact or not
if exact:
oldimg = manipulate.frebin(oldimg, newx, newy, total=total)
else:
oldimg = manipulate.frebin(oldimg, newx, newy, total=total)
#start updating the new header
oldhdr['NAXIS1'] = int(newx)
oldhdr['NAXIS2'] = int(newy)
#add comment orig size and new size
oldhdr.add_comment('rebinned: original image was %i by %i' %
(xsize, ysize))
#Correct the position of the reference pixel.
#Note that CRPIX values are given in FORTRAN (first pixel is (1,1)) convention
crpix = wcs.wcs.crpix
#update astrometry
if exact and xratio > 1:
crpix1 = (crpix[0] - 1.0) * xratio + 1.0
else:
crpix1 = (crpix[0] - 0.5) * xratio + 0.5
if exact and yratio > 1:
crpix2 = (crpix[1] - 1.0) * yratio + 1.0
else:
crpix2 = (crpix[1] - 0.5) * yratio + 0.5
oldhdr['CRPIX1'] = crpix1
oldhdr['CRPIX2'] = crpix2
#distortion correction, not implemented
#oldhdr['CDELT1'] /= xratio
#oldhdr['CDELT2'] /= yratio
oldhdr['CD1_1'] /= xratio
oldhdr['CD1_2'] /= yratio
oldhdr['CD2_1'] /= xratio
oldhdr['CD2_2'] /= yratio
#modify the B scale
#oldhdr['BSCALE'] /= pix_ratio
#write out a new FITS file
hdu = pf.PrimaryHDU(oldimg)
hdu.header = oldhdr
hdu.header.add_history(
'Updated: %s' % datetime.datetime.isoformat(datetime.datetime.now()))
if os.path.isfile(output):
os.remove(output)
hdu.writeto(output)
return oldimg, oldhdr
def hrebin(imagefile, newx, newy, output='rebinned.fits', ext=0, total=False):
"""
Expand or contract a FITS image and update the header.
Based on IDL routine hrebin.pro, but removed some functionality
:todo: remove the pywcs dependency
:param imagefile: name of the FITS file to be rebinned
:param newx: size of the new image in the X direction, integer scalar
:param newy: size of the new image in the Y direction, integer scalar
:param output: Name of the outputfile or None
:param ext: the FITS extension of the header and data
:return: None or updated img and hdr
"""
#todo: remove the pywcs dependency
#todo: add a routine for exact rebinning, now done with the same
fh = pf.open(imagefile)
oldhdr = fh[ext].header
oldimg = fh[ext].data
fh.close()
wcs = pywcs.WCS(oldhdr)
#old size
#xsize = oldhdr['NAXIS1']
#ysize = oldhdr['NAXIS2']
ysize, xsize = oldimg.shape
#ratios
xratio = newx / float(xsize)
yratio = newy / float(ysize)
#change in the aspect ratio
lambd = yratio / xratio
#Ratio of pixel areas
pix_ratio = xratio * yratio
#chech whether the new size is an exact match
exact = (xsize % newx == 0) | (newx % xsize == 0) and\
(ysize % newy == 0) | (newy % ysize == 0)
#rebin based on whether the rebinning is exact or not
if exact:
oldimg = manipulate.frebin(oldimg, newx, newy, total=total)
else:
oldimg = manipulate.frebin(oldimg, newx, newy, total=total)
#start updating the new header
oldhdr['NAXIS1'] = int(newx)
oldhdr['NAXIS2'] = int(newy)
#add comment orig size and new size
oldhdr.add_comment('rebinned: original image was %i by %i' % (xsize, ysize))
#Correct the position of the reference pixel.
#Note that CRPIX values are given in FORTRAN (first pixel is (1,1)) convention
crpix = wcs.wcs.crpix
#update astrometry
if exact and xratio > 1:
crpix1 = (crpix[0] - 1.0) * xratio + 1.0
else:
crpix1 = (crpix[0] - 0.5) * xratio + 0.5
if exact and yratio > 1:
crpix2 = (crpix[1] - 1.0) * yratio + 1.0
else:
crpix2 = (crpix[1] - 0.5) * yratio + 0.5
oldhdr['CRPIX1'] = crpix1
oldhdr['CRPIX2'] = crpix2
#distortion correction, not implemented
#oldhdr['CDELT1'] /= xratio
#oldhdr['CDELT2'] /= yratio
oldhdr['CD1_1'] /= xratio
oldhdr['CD1_2'] /= yratio
oldhdr['CD2_1'] /= xratio
oldhdr['CD2_2'] /= yratio
#modify the B scale
#oldhdr['BSCALE'] /= pix_ratio
#write out a new FITS file
hdu = pf.PrimaryHDU(oldimg)
hdu.header = oldhdr
hdu.header.add_history('Updated: %s' % datetime.datetime.isoformat(datetime.datetime.now()))
if os.path.isfile(output):
os.remove(output)
hdu.writeto(output)
return oldimg, oldhdr
def _fitSlitsToDirectImage(self):
"""
Fits a slit mask profiles to a direct image to recover the position and orientation of the slitmask.
By default the counts are not normalized to a peak count, but this can
be controlled using the optional keyword normalize. This should be used for
debugging purposes only. I included this part because in the beginning we
were missing flux calibration.
:Note: This is a very slow algorithm because of the insanely many nested for loops...
:todo: rewrite the algorithm.
:rtype: dictionary
"""
#generates a model array from the slit values, takes into account potential
#throughput of a slit
model = np.array([])
for vals in self.slits.values():
#must rebin/interpolate to the right scale, conserve surface flux
new = m.frebin(vals['profile'], int(vals['height']), total=True)
model = np.append(model, new)
self.fitting[vals['file']] = new
if self.normalize:
model /= np.max(model)
#mask out negative values and especially those equal to zero
msk = model > 0.0
self.fitting['model'] = model[msk]
self.fitting['mask'] = msk
#basic model information
mean = np.mean(self.fitting['model'])
sig = np.sqrt(np.sum((self.fitting['model'] - mean) ** 2))
diff = self.fitting['model'] - mean
self.fitting['sig'] = sig
self.fitting['diff'] = diff
#generate rotations
if self.rotation:
rotations = np.arange(-self.fitting['rotation'], self.fitting['rotation'], self.fitting['rotstep'])
rotations[(rotations < 1e-8) & (rotations > -1e-8)] = 0.0
#make a copy of the direct image
origimage = self.direct['rotatedImage'].copy()
else:
rotations = [0, ]
#x and yranges to cover in the grid
xran = range(-self.fitting['xrange'], self.fitting['xrange'], self.fitting['xstep'])
yran = range(-self.fitting['yrange'], self.fitting['yrange'], self.fitting['ystep'])
#define some helper variables
out = []
corrmin = -1e10
cm = 1e30
#loop over a range of rotations, x and y positions
#this should be done in some other way than having three nested loops... not good
for r in rotations:
if self.rotation:
if r != 0.0:
#d = interpolation.rotate(origimage, r, reshape=False)
d, hdrR = modify.hrot2(origimage.copy(), self.direct['rotatedHeader'], r, xc=None, yc=None)
else:
d = origimage.copy()
hdrR = self.direct['rotatedHeader']
else:
d = self.direct['rotatedImage'].copy()
hdrR = self.direct['rotatedHeader']
#only for debugging purposes
if self.normalize:
d /= np.max(d)
for x in xran:
for y in yran:
#all slits
dirdata = np.array([])
for s in self.slits.values():
#direct image data
dirdat = d[s['ymin'] + y:s['ymax'] + y,\
s['xmin'] + x:s['xmax'] + x]
#sum the counts inside the slit
dirdat = np.sum(dirdat, axis=1)
dirdata = np.append(dirdata, dirdat.ravel())
#remove the masked pixels
dirdata = dirdata[self.fitting['mask']]
#chi**2, p-value
chisq = stats.chisquare(dirdata, self.fitting['model'])
#chisq = self._chiSquare(dirdata, self.fitting['model'], 1./self.fitting['model'])
#correlation coeff
mean = np.mean(dirdata)
sig = np.sqrt(np.sum((dirdata - mean) ** 2))
diff = dirdata - mean
corr = np.sum(self.fitting['diff'] * diff) / self.fitting['sig'] / sig
#save info
tmp = [r, x, y, chisq[0], chisq[0] / s['pixels'], chisq[1], corr]
#tmp = [r, x, y, chisq, chisq / s['pixels'], 0.0, corr]
out.append(tmp)
if chisq[0] < cm:
#if chisq < cm:
cm = chisq[0]
#cm = chisq
minpos = tmp
dir = dirdata
img = d
hdr = hdrR
if corr > corrmin:
corrmin = corr
minpos2 = tmp
dir2 = dirdata
if self.debug:
print r, x, y, chisq[0] / s['pixels'], chisq[1], corr
#print r, x, y, chisq / s['pixels'], 0.0, corr
#save results
self.result['outputs'] = out
self.result['minimaPosition'] = minpos
self.result['FinalImage'] = img
self.result['FinalHeader'] = hdr
self.result['WCS'] = pywcs.WCS(hdr)
#self.result['WCS'] = wcs.Projection(hdr)
#choose the method
if 'chi' in self.fitting['method']:
self.result['rotation'] = minpos[0]
self.result['posangle'] = minpos[0] + self.slits[self.centralFile]['POSANGLE']
self.result['x'] = minpos[1]
self.result['y'] = minpos[2]
self.result['bestfit'] = dir
self.result['xcenter'] = self.direct['xposition'] + minpos[1]
self.result['ycenter'] = self.direct['yposition'] + minpos[2]
self.result['pvalue'] = minpos[5]
elif 'corr' in self.fitting['method']:
self.result['rotation'] = minpos2[0]
self.result['posangle'] = minpos[0] + self.slits[self.centralFile]['POSANGLE']
self.result['x'] = minpos2[1]
self.result['y'] = minpos2[2]
self.result['bestfit'] = dir2
self.result['xcenter'] = self.direct['xposition'] + minpos2[1]
self.result['ycenter'] = self.direct['yposition'] + minpos2[2]
self.result['pvalue'] = minpos2[5]
else:
raise NotImplementedError, 'This minimization method has not yet been implemented...'
if self.debug:
print minpos
print minpos2
