if args.mask: mshape, mwcs = enmap.read_map_geometry(args.mask) mbox = enmap.pixbox_of(mwcs, imap.shape, imap.wcs) mask = enmap.read_map(args.mask, pixbox=mbox).preflat[0] > 0 idiv *= 1-mask # Inpaint masked area, in case it contains fourier-unfriendly values imap = enmap.inpaint(imap, mask) del mask if "debug" in args.output: dump_prefix = args.odir + "/region_%02d_" % ri else: dump_prefix = None imap[~np.isfinite(imap)] = 0 idiv[~np.isfinite(idiv)] = 0 idiv[idiv<0] = 0 # Get our flux conversion factor #print "fit barea: %8.3f" % (barea*1e9) fluxconv = utils.flux_factor(barea, args.freq*1e9)/1e6 reg_cat = None local_amps = None for ci in range(len(imap)): # Build an amplitude prior from our input catalog fluxes prior = dory.build_prior(icat.flux[:,ci]/fluxconv, icat.dflux[:,ci]/fluxconv, 1/args.prior) src_pos = np.array([icat.dec,icat.ra]).T fit_inds, amp, icov, lamps = dory.fit_src_amps(imap[ci], get_div(idiv,ci), src_pos, beam, prior=prior, apod=args.apod, apod_margin=args.apod_margin, verbose=args.verbose, dump=dump_prefix, hack=args.hack, region=ri) if reg_cat is None: reg_cat = icat[fit_inds].copy() reg_cat.amp = reg_cat.damp = reg_cat.flux = reg_cat.dflux = 0 if ci == 0: local_amps = lamps reg_cat.amp[:,ci] = amp reg_cat.damp[:,ci] = np.diag(icov)**-0.5
data.tod = data.tod.astype(dtype)
# Set up our likelihood
L = Likelihood(data, srcpos[:,sids], amps[sids], filter=highpass)
# Find out which sources are reliable, so we don't waste time on bad ones
if prune_unreliable_srcs:
_, aicov = L.fit_amp()
good = amps[sids]**2*aicov[:,0,0,0,0] > args.minsn**2
sids = [sid for sid,g in zip(sids,good) if g]
nsrc = len(sids)
print("Restricted to %d srcs: %s" % (nsrc,", ".join(["%d (%.1f)" % (i,a) for i,a in zip(sids,amps[sids])])))
if nsrc == 0: continue
L = Likelihood(data, srcpos[:,sids], amps[sids], perdet=perdet, thumbs=True, N=L.N, method=args.method, filter=highpass)
beam_area = get_beam_area(data.beam)
_, uids = actdata.split_detname(data.dets) # Argh, stupid detnames
freq = data.array_info.info.nom_freq[uids[0]]
fluxconv = utils.flux_factor(beam_area, freq*1e9)
group_data.append(bunch.Bunch(data=data, sids=sids, lik=L, id=id, oid=oid, ind=ind, beam_area=beam_area, freq=freq, fluxconv=fluxconv))
if len(group_data) == 0:
print("No usable tods in group %s. Skipping" % ",".join([ids[i] for i in group]))
continue
# Set up the full likelihood
progress_thumbs = args.minimaps and verbose >= 3
chisq_wrappers = [data.lik.chisq_wrapper(thumb_path=args.odir + "/" + data.oid + "_thumb%03d.fits", thumb_interval=progress_thumbs) for data in group_data]
def likfun(off): return sum([chisq_wrapper(off) for chisq_wrapper in chisq_wrappers])
# Representaive individual lik for stuff that's common to them. This is a bit hacky.
# A single joint lik class would have been cleaner.
L = group_data[0].lik
if not args.nogrid:
pixbox=pixbox,
bbox=mybbox,
comm=comm).astype(dtype)
refmap = signal.prepare(refmap)
# Get the frequency and beam for this chunk. We assume that
# this is the same for every member of the chunk, so we only need
# to do this for one scan
scan = actscan.ACTScan(filedb.data[chunk_ids[inds[0]]])
_, dets = actdata.split_detname(scan.dets)
beam = scan.beam
freq = scan.array_info.info.nom_freq[dets[0]]
barea = planet9.calc_beam_area(scan.beam)
# Get the conversion from ref-freq flux to observed amplitude. This includes
# dilution by the beam area
flux2amp = 1 / utils.flux_factor(barea, args.fref * 1e9, utils.T_cmb)
fref2freq = utils.planck(freq * 1e9, args.Tref) / utils.planck(
args.fref * 1e9, args.Tref)
rfact = flux2amp * fref2freq * 1e3 # 1e3 for flux in mJy and amp in uK
# only work will be 3,ny,nx. The rest are scalar. Will copy in-out as necessary
work = signal.work()
rhs = area[0]
div = rhs.copy()
wrhs = signal.prepare(rhs)
wdiv = signal.prepare(div)
kvals = np.zeros(len(corr_pos), dtype)
freqs, bareas = np.zeros(ntod), np.zeros(ntod)
bleh = True and args.inject
if bleh:
# We support polarization here, but treat each component independently imap = enmap.read_map(args.imap, pixbox=reg_pad).preflat[:args.ncomp] idiv = divdiag(enmap.read_map(args.idiv, pixbox=reg_pad))[:args.ncomp] if args.mask: mshape, mwcs = enmap.read_map_geometry(args.mask) mbox = enmap.pixbox_of(mwcs, imap.shape, imap.wcs) idiv *= (1-enmap.read_map(args.mask, pixbox=mbox).preflat[0]) if "debug" in args.output: dump_prefix = args.odir + "/region_%02d_" % ri else: dump_prefix = None imap[~np.isfinite(imap)] = 0 idiv[~np.isfinite(idiv)] = 0 idiv[idiv<0] = 0 # Get our flux conversion factor beam2d = dory.calc_2d_beam(beam, imap.shape, imap.wcs) barea = dory.calc_beam_transform_area(beam2d) fluxconv = utils.flux_factor(barea, args.freq*1e9)/1e6 reg_cat = None for ci in range(len(imap)): # Build an amplitude prior from our input catalog fluxes prior = dory.build_prior(icat.flux[:,ci]/fluxconv, icat.dflux[:,ci]/fluxconv, 1/args.prior) src_pos = np.array([icat.dec,icat.ra]).T print imap.shape, idiv.shape, ci fit_inds, amp, icov = dory.fit_src_amps(imap[ci], get_div(idiv,ci), src_pos, beam, prior=prior, apod=args.apod, apod_margin=args.apod_margin, verbose=args.verbose, dump=dump_prefix, hack=args.hack, region=ri) if reg_cat is None: reg_cat = icat[fit_inds].copy() reg_cat.amp = reg_cat.damp = reg_cat.flux = reg_cat.dflux = 0 reg_cat.amp[:,ci] = amp reg_cat.damp[:,ci] = np.diag(icov)**-0.5 reg_cat.flux = reg_cat.amp*fluxconv reg_cat.dflux = reg_cat.damp*fluxconv
srcs= np.array([[srcpos[0], srcpos[1], 1, 0, 0, 70.6, 6.13, D]]).T # use close-enough value
# srcs= np.array([[srcpos[0], srcpos[1]]])
with bench.show("create pointing matrix"):
P = lib.PmatTotVar(scan, srcs, perdet=False, sys=sys)
# P = lib.PmatTot(scan, srcs[:,0], perdet=False, sys=sys)
# I should be able to use the same pointing matrix to do a search
# of pulsars by treating different period and phases as different
# sources, and estimating their amplitudes together. Let me give
# that a try below
# a factor to convert uK to mJy
beam_area = lib.get_beam_area(scan.beam)
_, uids = actdata.split_detname(scan.dets) # Argh, stupid detnames
freq = scan.array_info.info.nom_freq[uids[0]]
fluxconv = u.flux_factor(beam_area, freq*1e9)/1e3
print("fluxconv = ", fluxconv)
# prepare pointing matrix and noise model for searching
# for signal-only test I won't apply any noise model
# N = NmatTot(scan, model="uncorr", window=2.0, filter=highpass)
# srcs
# rhs (= P'N"d)
# N.apply(tod)
with bench.show("project TOD to src space"):
rhs = P.backward(tod, ncomp=1)
# div (= P'N"P)
tod[:] = 0
P.forward(tod, rhs*0+1)