def jyBm2jyPix(in_image, out_image):
"""
Blindly convert pixel values of in_image to Jy/pix, assuming they are Jy/bm.
we check nothing and overwrite out_image so you best watch yourself.
"""
rad_to_arcsec = 206264.81
twopi_over_eightLnTwo = 1.133
hdr = imhead(in_image)
flux_conversion = (rad_to_arcsec * np.abs(hdr['incr'][0])) * (
rad_to_arcsec * np.abs(hdr['incr'][1])) / (
twopi_over_eightLnTwo * hdr['restoringbeam']['major']['value'] *
hdr['restoringbeam']['minor']['value'])
flux_string = "(IM0 * %f)" % flux_conversion
rmtables(out_image)
immath(imagename=in_image,
expr=flux_string,
outfile=out_image,
mode='evalexpr')
new_unit = 'Jy/pixel'
imhead(imagename=out_image, mode='put', hdkey='BUNIT', hdvalue=new_unit)
return 0
def feather_one(sd_map, int_map, int_pb, tag=''):
"""
Feather together a single dish map sd_map with an interferometric
map int_map, with primary beams probably handled.
returns the names of the pbcorrected and not pbcorrected images.
"""
outfile = int_map + tag + '.feather'
outfile_pbcord = outfile + '.pbcor'
outfile_uncorr = outfile
#rmtables(outfile_uncorr)
#rmtables(outfile_pbcord)
rmtables(sd_map + ".TMP.intGrid.intPb")
imregrid(imagename=sd_map,
template=int_map,
output=sd_map + ".TMP.intGrid",
overwrite=True)
immath(imagename=[sd_map + ".TMP.intGrid", int_pb],
expr='IM0*IM1',
outfile=sd_map + ".TMP.intGrid.intPb")
feather(imagename=outfile_uncorr,
highres=int_map,
lowres=sd_map + ".TMP.intGrid.intPb")
immath(imagename=[outfile_uncorr, int_pb],
expr='IM0/IM1',
outfile=outfile_pbcord)
# clean up after self
rmtables(sd_map + ".TMP.intGrid.intPb")
rmtables(sd_map + ".TMP.intGrid")
return [outfile_pbcord, outfile_uncorr]
def noneg_model(modelname, ms, imagename, **kwargs):
"""
Given a model image, set all model components positive, then ft them into
the ms's model column
"""
if os.path.exists(modelname + ".positive"):
shutil.rmtree(modelname + ".positive")
immath(
imagename=modelname,
expr='iif(IM0<0, 0.0, IM0)',
outfile=modelname + ".positive",
)
if os.path.exists(modelname):
if os.path.exists(modelname + ".old"):
shutil.rmtree(modelname + ".old")
os.rename(modelname, modelname + ".old")
tclean(vis=ms,
imagename=imagename,
startmodel=modelname + ".positive",
niter=0,
deconvolver='mtmfs',
specmode='mfs',
nterms=1,
calcpsf=False,
calcres=False,
interactive=False,
savemodel='modelcolumn',
**kwargs)
def make_clnMod_fromImg(sd_map, int_map, tag='', clean_up=True):
"""
Take sd_map and regrid it into pixel coordinates of int_map, converting from
Jy/bm into Jy/pix.
Both input images are CASA images presumed to have surface brightess
units of Jy/bm
tag is an optional string that will be included in the outpu file name
Returns the name of the output file.
Note: sd_map and int_map can be of any provenance as long as the SB units
are Jy/bm. Results have only been validated for the case that the sd_map
pixels are larger than the int_map pixels however.
"""
regridded_sd_map = sd_map + '.TMP.regrid'
out_sd_map = sd_map + tag + '.regrid.jyPix'
imregrid(imagename=sd_map,
template=int_map,
output=regridded_sd_map,
overwrite=True)
sd_header = imhead(regridded_sd_map)
int_header = imhead(int_map)
rad_to_arcsec = 206264.81
twopi_over_eightLnTwo = 1.133
flux_conversion = (rad_to_arcsec * np.abs(sd_header['incr'][0])) * (
rad_to_arcsec * np.abs(sd_header['incr'][1])
) / (twopi_over_eightLnTwo * sd_header['restoringbeam']['major']['value'] *
sd_header['restoringbeam']['minor']['value'])
print "===================="
print "THESE SHOULD BE ARCSEC:" + sd_header['restoringbeam']['major'][
'unit'] + " " + sd_header['restoringbeam']['minor']['unit']
print "THESE SHOULD BE RADIANS: " + sd_header['axisunits'][
0] + " " + sd_header['axisunits'][1]
print "if not the unit conversions were wrong...."
print "===================="
flux_string = "(IM0 * %f)" % flux_conversion
rmtables(out_sd_map)
immath(imagename=regridded_sd_map,
expr=flux_string,
outfile=out_sd_map,
mode='evalexpr')
new_unit = 'Jy/pixel'
imhead(imagename=out_sd_map, mode='put', hdkey='BUNIT', hdvalue=new_unit)
# clean up after self-
rmtables(regridded_sd_map)
return out_sd_map
def fix_image_calib(sd_image,sd_img_fixed,cal_factor=1.0,beam_factor=1.0):
"""
Rescale image brightness (cal_factor) and BMAJ,BMIN (beam_factor)
in a CASA image sd_image (any CASA image). output file to
sd_img_fixed.
"""
immath(imagename=[sd_image],expr='IM0*'+str(cal_factor),outfile=sd_img_fixed)
hdr = imhead(imagename=sd_image,mode='summary')
bmaj_str = str(hdr['restoringbeam']['major']['value'] * beam_factor)+hdr['restoringbeam']['major']['unit']
bmin_str = str(hdr['restoringbeam']['minor']['value'] * beam_factor)+hdr['restoringbeam']['minor']['unit']
# you have to do BMIN first
imhead(sd_img_fixed,mode='put',hdkey='BMIN',hdvalue=bmin_str)
imhead(sd_img_fixed,mode='put',hdkey='BMAJ',hdvalue=bmaj_str)
return sd_img_fixed
def calc_fidelity(inimg,
refimg,
pbimg='',
psfimg='',
fudge_factor=1.0,
scale_factor=1.0,
pb_thresh=0.25,
clean_up=True,
outfile=''):
"""Calculate fidelity of inimg with reference to refimg.
Use Gaussian PSF with parameters described in inimg header, unless
an explicit psfimg is provided.
If a primary beam image (pbimg) is provided, use it to restrict
the area over which the fidelity is calculated, using pb_thresh as
the lower limit (relative to max(pbimg)).
clean_up controls whether intermediate files created in the
process are removed or not (all contain the string TMP). These can
be useful for sanity checking, but proper behavior is not
guaranteed if any are present already when the routine is called.
outfile specifies the file-name root for a fidelity image and a fractional error
image that will be created.
--- ****a pbimg is required to creat the outfile
inimg, refimg, [psfimg], and [pbimg] should be CASA images. outfile is a CASA image.
All input images should have the same axes and axis order.
***pbimg, if provided, should furthermore have the same pixel coordinates as inimg
(Cell size, npix, coordinate reference pixel, etc)***
***all input images should have the same number and ordering of axes!!!
fudge_factor multiples the beamwidth obtained from the input image, before
convolving refimg for comparison
scale_factor multiplies the inimg pixel values (i.e. it recalibrates them)
--> use these reluctantly and only if you know what you are doing
OUTPUTS: a dictionary containing
f1 = 1 - max(abs(inimg-refimg)) / max(refimg) - 'classic' definition
f2 = 1 - sum( refimg .* abs(inimg-refimg) ) / sum( refimg .* inimg)
--> this is a somewhat poorly behaved fidelity definition that was evaluated for ngVLA
(appearing in the draft ngVLA science requirements, May 2019)
--> it is equivalent to a weighted sum of fractional errors, with the fraction taken with
respect to the formed image inimg and the weight being inimg*refimg
f2b = 1 - sum( refimg .* abs(inimg-refimg) ) / sum( refimg .* refimg)
--> this is the original (ngVLA science requirememts, Nov. 2017) and better-behaved
ngVLA fidelity definition, with the fraction taken with respect to the model (refimg),
and the weight being refimg^2
f3 = 1 - sum( beta .* abs(inimg-refimg) ) / sum( beta.^2 )
--> this is the current ngVLA fidelity definition that has been adopted, where
beta_i = max(abs(inimg_i,),abs(refimg_i))
In all of the above "i" is a pixel index, .* and .^ are element- (pixel-) wise operations,
and sums are over pixels
Various ALMA-adopted fidelity measures are also reported, and the correlation coefficient
HISTORY:
August/September 2019 - B. Mason (nrao) - original version
"""
ia = iatool()
ia.open(inimg)
# average over the stokes axis to get it down to 3 axes which is what our other one has
imvals = np.squeeze(ia.getchunk()) * scale_factor
img_cs = ia.coordsys()
# how to trim the freq axis--
#img_shape = (ia.shape())[0:3]
img_shape = ia.shape()
ia.close()
# get beam info
hdr = imhead(imagename=inimg, mode='summary')
bmaj_str = str(hdr['restoringbeam']['major']['value'] *
fudge_factor) + hdr['restoringbeam']['major']['unit']
bmin_str = str(hdr['restoringbeam']['minor']['value'] *
fudge_factor) + hdr['restoringbeam']['minor']['unit']
bpa_str = str(hdr['restoringbeam']['positionangle']
['value']) + hdr['restoringbeam']['positionangle']['unit']
# i should probably also be setting the beam * fudge_factor in the *header* of the input image
if len(pbimg) > 0:
ia.open(pbimg)
pbvals = np.squeeze(ia.getchunk())
pbvals /= np.max(pbvals)
pbvals = np.where(pbvals < pb_thresh, 0.0, pbvals)
#good_pb_ind=np.where( pbvals >= pb_thresh)
#bad_pb_ind=np.where( pbvals < pb_thresh)
#pbvals[good_pb_ind] = 1.0
#if bad_pb_ind[0]:
# pbvals[bad_pb_ind] = 0.0
else:
pbvals = imvals * 0.0 + 1.0
#good_pb_ind = np.where(pbvals)
#bad_pb_ind = [np.array([])]
##
##############
# open, smooth, and regrid reference image
#
smo_ref_img = refimg + '.TMP.smo'
# if given a psf image, use that for the convolution. need to regrid onto input
# model coordinate system first. this is mostly relevant for the single dish
# if the beam isn't very gaussian (as is the case for alma sim tp)
if len(psfimg) > 0:
# consider testing and fixing the case the reference image isn't jy/pix
ia.open(refimg)
ref_cs = ia.coordsys()
ref_shape = ia.shape()
ia.close()
ia.open(psfimg)
psf_reg_im = ia.regrid(csys=ref_cs.torecord(),
shape=ref_shape,
outfile=psfimg + '.TMP.regrid',
overwrite=True,
axes=[0, 1])
psf_reg_im.done()
ia.close()
ia.open(refimg)
# default of scale= -1.0 autoscales the PSF to have unit area, which preserves "flux" in units of the input map
# scale=1.0 sets the PSF to have unit *peak*, which results in flux per beam in the output
ref_convd_im = ia.convolve(outfile=smo_ref_img,
kernel=psfimg + '.TMP.regrid',
overwrite=True,
scale=1.0)
ref_convd_im.setbrightnessunit('Jy/beam')
ref_convd_im.done()
ia.close()
if clean_up:
rmtables(psfimg + '.TMP.regrid')
else:
# consider testing and fixing the case the reference image isn't jy/pix
ia.open(refimg)
im2 = ia.convolve2d(outfile=smo_ref_img,
axes=[0, 1],
major=bmaj_str,
minor=bmin_str,
pa=bpa_str,
overwrite=True)
im2.done()
ia.close()
smo_ref_img_regridded = smo_ref_img + '.TMP.regrid'
ia.open(smo_ref_img)
im2 = ia.regrid(csys=img_cs.torecord(),
shape=img_shape,
outfile=smo_ref_img_regridded,
overwrite=True,
axes=[0, 1])
refvals = np.squeeze(im2.getchunk())
im2.done()
ia.close()
ia.open(smo_ref_img_regridded)
refvals = np.squeeze(ia.getchunk())
ia.close()
# set all pixels to zero where the PB is low - to avoid NaN's
imvals = np.where(pbvals, imvals, 0.0)
refvals = np.where(pbvals, refvals, 0.0)
#if len(bad_pb_ind) > 0:
#imvals[bad_pb_ind] = 0.0
#refvals[bad_pb_ind] = 0.0
deltas = (imvals - refvals).flatten()
# put both image and model values in one array to calculate Beta for F_3-
allvals = np.array([np.abs(imvals.flatten()), np.abs(refvals.flatten())])
# the max of (image_pix_i,model_pix_i), in one flat array of length nixels
maxvals = allvals.max(axis=0)
# carilli definition. rosero eq1
f_eq1 = 1.0 - np.max(np.abs(deltas)) / np.max(refvals)
f_eq2 = 1.0 - (refvals.flatten() * np.abs(deltas)).sum() / (refvals *
imvals).sum()
f_eq2b = 1.0 - (refvals.flatten() *
np.abs(deltas)).sum() / (refvals * refvals).sum()
#f_eq3 = 1.0 - (maxvals[gi] * np.abs(deltas[gi])).sum() / (maxvals[gi] * maxvals[gi]).sum()
f_eq3 = 1.0 - (pbvals.flatten() * maxvals * np.abs(deltas)).sum() / (
pbvals.flatten() * maxvals * maxvals).sum()
# if an output image was requested, and a pbimg was given; make one.
if ((len(outfile) > 0) & (len(pbimg) > 0)):
weightfile = 'mypbweight.TMP.im'
rmtables(weightfile)
immath(imagename=[pbimg],
mode='evalexpr',
expr='ceil(IM0/max(IM0) - ' + str(pb_thresh) + ')',
outfile=weightfile)
betafile = 'mybeta.TMP.im'
rmtables(betafile)
immath(imagename=[inimg, smo_ref_img_regridded],
mode='evalexpr',
expr='iif(abs(IM0) > abs(IM1),abs(IM0),abs(IM1))',
outfile=betafile)
# 19sep19 - change to the actual F_3 contrib ie put abs() back in
rmtables(outfile)
print " Writing fidelity error image: " + outfile
immath(imagename=[inimg, smo_ref_img_regridded, weightfile, betafile],
expr='IM3*IM2*abs(IM0-IM1)/sum(IM3*IM3*IM2)',
outfile=outfile)
# 19sep19 - add fractional error (rel to beta) to output
rmtables(outfile + '.frac')
print " Writing fractional error image: " + outfile + '.frac'
immath(imagename=[inimg, smo_ref_img_regridded, weightfile, betafile],
expr='IM2*(IM0-IM1)/IM3',
outfile=outfile + '.frac')
if clean_up:
rmtables(weightfile)
rmtables(betafile)
# pearson correlation coefficient evaluated above beta = 1% peak reference image
gi = np.where(np.abs(maxvals) > 0.01 * np.abs(refvals.max()))
ii = imvals.flatten()
mm = refvals.flatten()
mm -= mm.min()
# (x-mean(x)) * (y-mean(y)) / sigma_x / sigma_y
cc = (ii[gi] - ii[gi].mean()) * (mm[gi] - mm[gi].mean()) / (
np.std(ii[gi]) * np.std(mm[gi]))
#cc = (ii[gi] - ii[gi].mean()) * (mm[gi] - mm[gi].mean()) / (np.std(mm[gi]))**2
corco = cc.sum() / cc.shape[0]
fa = np.abs(mm) / np.abs(mm - ii)
fa_0p1 = np.median(fa[(np.abs(ii) > 1e-3 * mm.max()) |
(np.abs(mm) > 1e-3 * mm.max())])
fa_1 = np.median(fa[(np.abs(ii) > 1e-2 * mm.max()) |
(np.abs(mm) > 1e-2 * mm.max())])
fa_3 = np.median(fa[(np.abs(ii) > 3e-2 * mm.max()) |
(np.abs(mm) > 3e-2 * mm.max())])
fa_10 = np.median(fa[(np.abs(ii) > 1e-1 * mm.max()) |
(np.abs(mm) > 1e-1 * mm.max())])
#gi2 = (np.abs(ii) > 1e-3 * mm.max()) | (np.abs(mm) > 1e-3 * mm.max())
print "*************************************"
print 'image: ', inimg, 'reference image:', refimg
print "Eq1 / Eq2 / Eq2b / Eq3 / corrCoeff "
print f_eq1, f_eq2, f_eq2b, f_eq3, corco
print ' ALMA: ', fa_0p1, fa_1, fa_3, fa_10
print "*************************************"
fidelity_results = {
'f1': f_eq1,
'f2': f_eq2,
'f2b': f_eq2b,
'f3': f_eq3,
'falma': [fa_0p1, fa_1, fa_3, fa_10]
}
if clean_up:
rmtables(smo_ref_img)
rmtables(smo_ref_img_regridded)
return fidelity_results
def make_clnMod_fromImg(sd_map, int_map, pb_map='', tag='', clean_up=True):
"""
Take sd_map and regrid it into pixel coordinates of int_map, converting from
Jy/bm into Jy/pix.
Both input images are CASA images presumed to have surface brightess
units of Jy/bm
tag is an optional string that will be included in the outpu file name
If optional pb_map [RECOMMENDED] is provided, then apply the PB before
making model. pb_map should be in same pixel & coordinate basis as int_map
Returns the name of the output file.
Note: sd_map and int_map can be of any provenance as long as the SB units
are Jy/bm. Results have only been validated for the case that the sd_map
pixels are larger than the int_map pixels however.
You should probably multiply the result by the INT PB before using it in TCLEAN
"""
regridded_sd_map = sd_map + '.TMP.regrid'
out_sd_map = sd_map + tag + '.regrid.jyPix'
imregrid(imagename=sd_map,
template=int_map,
output=regridded_sd_map,
overwrite=True)
sd_header = imhead(regridded_sd_map)
int_header = imhead(int_map)
rad_to_arcsec = 206264.81
twopi_over_eightLnTwo = 1.133
flux_conversion = (rad_to_arcsec * np.abs(sd_header['incr'][0])) * (
rad_to_arcsec * np.abs(sd_header['incr'][1])
) / (twopi_over_eightLnTwo * sd_header['restoringbeam']['major']['value'] *
sd_header['restoringbeam']['minor']['value'])
print("====================")
print("THESE SHOULD BE ARCSEC:" +
sd_header['restoringbeam']['major']['unit'] + " " +
sd_header['restoringbeam']['minor']['unit'])
print("THESE SHOULD BE RADIANS: " + sd_header['axisunits'][0] + " " +
sd_header['axisunits'][1])
print("if not the unit conversions were wrong....")
print("====================")
flux_string = "(IM0 * %f)" % flux_conversion
rmtables(out_sd_map)
immath(imagename=regridded_sd_map,
expr=flux_string,
outfile=out_sd_map,
mode='evalexpr')
new_unit = 'Jy/pixel'
imhead(imagename=out_sd_map, mode='put', hdkey='BUNIT', hdvalue=new_unit)
if (len(pb_map) > 0):
rmtables('placeholder.im')
try:
immath(imagename=[out_sd_map],
mode='evalexpr',
expr='IM0*1.0',
outfile='placeholder.im')
except:
print(" Problem creating placeholder table ")
else:
rmtables(out_sd_map)
immath(imagename=['placeholder.im', pb_map],
expr='IM0*IM1',
outfile=out_sd_map)
finally:
rmtables('placeholder.im')
# clean up after self-
rmtables(regridded_sd_map)
return out_sd_map