def __init__(self,
boresight=None,
offsets=None,
comps=None,
tod=None,
sys=None,
cut=None,
site=None,
mjd0=0,
noise=None,
dets=None,
id=""):
# Boresight will always be unwound, i.e. it will have no 2pi jumps in it.
# Time is measured in seconds since start of scan, with mjd0 indicating the MJD of the scan start.
self.boresight = np.asfarray(boresight) # [nsamp,coords]
self.offsets = np.asfarray(offsets) # [ndet,coords]
self.comps = np.asfarray(comps) # [ndet,comps]
self.cut = rangelist.Multirange(cut) # Multirange[ndet,ranges]
self.noise = noise
self.id = id # Identifier of this scan, for printing purposes
# These are needed in order to interpret the coordinates
self.sys = str(sys) # str
self.site = site
self.mjd0 = mjd0 # time basis
self.dets = np.arange(len(self.comps)) if dets is None else dets
self.dgrid = (1, np.max(self.dets) + 1)
self.hwp = None
# Not part of the general interface
self._tod = np.asfarray(tod) # [ndet,nsamp]
def find_spikes(tod, nsigma=10, width=15, padding=7, noise=None):
res = []
ftod = tod.reshape(-1, tod.shape[-1])
if noise is None: noise = estimate_white_noise(ftod)**0.5
for di, d in enumerate(ftod):
smooth = scipy.signal.medfilt(d, width)
bad = np.abs(d - smooth) > noise[di] * nsigma
bad = rangelist.Rangelist(bad)
bad = bad.widen(padding)
res.append(rangelist.Rangelist(bad))
res = rangelist.Multirange(res)
res.data.reshape(tod.shape[:-1])
return res
def find_jumps(tod,
bsize=100,
nsigma=10,
margin=50,
step=50,
margin_step=1000):
ndet = tod.shape[0]
cuts = []
for det, det_tod in enumerate(tod):
n = len(det_tod)
# Compute difference tod
dtod = det_tod[1:] - det_tod[:-1]
nsamp = dtod.size
# Find typical standard deviation
nblock = int(nsamp / bsize)
sigma = utils.medmean(
np.var(dtod[:nblock * bsize].reshape(nblock, bsize), -1))**0.5
# Look for samples that deviate too much from 0
bad = np.abs(dtod) > sigma * nsigma
bad = np.concatenate([bad[:1], bad])
# Look for steps, areas where the mean level changes dramatically on each
# side of the jump. First find the center of each bad region
steps = bad * 0
labels, nlabel = scipy.ndimage.label(bad)
centers = np.array(
scipy.ndimage.center_of_mass(bad, labels,
np.arange(nlabel + 1))).astype(int)[:,
0]
# Find mean to the left and right of each bad region
for i, pos in enumerate(centers):
m1 = np.mean(det_tod[max(0, pos -
step * 3 / 2):max(1, pos - step / 2)])
m2 = np.mean(det_tod[min(n - 2, pos +
step / 2):min(n - 1, pos + step * 3 / 2)])
if np.abs(m2 - m1) > sigma * nsigma:
steps[pos] = 1
#print centers.shape, np.sum(steps)
# Grow each cut by a margin
bad = scipy.ndimage.distance_transform_edt(1 - bad) <= margin
steps = scipy.ndimage.distance_transform_edt(1 - steps) <= margin_step
cuts.append(rangelist.Rangelist(bad | steps))
return rangelist.Multirange(cuts)
def __init__(self, fname):
self.fname = fname
with h5py.File(fname, "r") as hfile:
for k in ["boresight", "offsets", "comps", "sys", "mjd0", "dets"]:
setattr(self, k, hfile[k].value)
n = self.boresight.shape[0]
neach = hfile["cut/neach"].value
flat = hfile["cut/flat"].value
self.cut = rangelist.Multirange((n, neach, flat), copy=False)
self.cut_noiseest = self.cut.copy()
self.noise = nmat.read_nmat(hfile, "noise")
self.site = bunch.Bunch(
{k: hfile["site/" + k].value
for k in hfile["site"]})
self.subdets = np.arange(self.ndet)
self.hwp = np.zeros(n)
self.hwp_phase = np.zeros([n, 2])
self.sampslices = []
self.id = os.path.basename(fname)
self.entry = bunch.Bunch(id=self.id)
def to_rangelist(self):
ranges = self.to_ranges()
return rangelist.Multirange(
[rangelist.Rangelist(r, n=self.nsamp) for r in ranges])
"hor",
args.objname,
mjd,
site=scan.site)
visible = np.any(object_pos[1] >= margin)
if not visible:
cut = rangelist.zeros((d.ndet, d.nsamp))
else:
pmap = pmat.PmatMap(scan, mask, sys="hor:%s" % args.objname)
# Build a tod to project onto.
tod = np.zeros((d.ndet, d.nsamp), dtype=dtype)
# And project
pmap.forward(tod, mask)
# Any nonzero samples should be cut
tod = np.rint(tod)
cut = rangelist.Multirange([rangelist.Rangelist(t) for t in tod])
print "%s %6.4f %d" % (id, float(cut.sum()) / cut.size, visible)
mystats.append([ind, float(cut.sum()) / cut.size, visible])
# Write cuts to output directory
if args.persample:
files.write_cut("%s/%s.cuts" % (args.odir, id),
d.dets,
cut,
nrow=d.array_info.nrow,
ncol=d.array_info.ncol)
mystats = np.array(mystats)
stats = utils.allgatherv(mystats, comm)
if comm.rank == 0:
with open(args.odir + "/stats.txt", "w") as f:
for stat in stats:
f.write("%s %6.4f %d\n" %
def nocut(ndet, nsamp):
return rangelist.Multirange([
rangelist.Rangelist(np.zeros([0, 2], dtype=int), n=nsamp)
for i in range(ndet)
])