def fname_to_date_tuple(fname):
'''Take a filename like m120507_0123, return 12may07'''
months = {
"01": "jan",
"02": "feb",
"03": "mar",
"04": "apr",
"05": "may",
"06": "jun",
"07": "jul",
"08": "aug",
"09": "sep",
"10": "oct",
"11": "nov",
"12": "dec"
}
if len(fname) != 17:
raise Exception("The file name '%s' is not of correct length. It "
"must be of the form mYYmmdd_nnnn.fits" % fname)
try:
fdate = fname.split("m")[1][0:6]
yr, mn, dy = "20" + fdate[0:2], fdate[2:4], int(fdate[4:6])
month = months[mn]
except:
warning("Could not parse date out of file name: %s" % (fname))
return yr, month, dy
def extract_spectra(maskname, band, interactive=True, width=10,
target='default'):
if target == 'default':
## Get objectnames from slit edges
edges = np.load('slit-edges_{}.npy'.format(band))
objectnames = [edge['Target_Name'] for edge in edges[:-1]]
eps_files = ['{}_{}_{}_eps.fits'.format(maskname, band, objectname)
for objectname in objectnames]
sig_files = ['{}_{}_{}_sig.fits'.format(maskname, band, objectname)
for objectname in objectnames]
spectrum_plot_files = ['{}_{}_{}.png'.format(maskname, band, objectname)
for objectname in objectnames]
fits_files = ['{}_{}_{}_1D.fits'.format(maskname, band, objectname)
for objectname in objectnames]
else:
objectnames = [target]
eps_files = ['{}_{}_eps.fits'.format(target, band)]
sig_files = ['{}_{}_sig.fits'.format(target, band)]
spectrum_plot_files = ['{}_{}.png'.format(target, band)]
fits_files = ['{}_{}_1D.fits'.format(target, band)]
aperture_tables = {}
if interactive:
print_instructions()
# First, iterate through all slits and define the apertures to extract
for i,eps_file in enumerate(eps_files):
objectname = objectnames[i]
eps = fits.open(eps_file, 'readonly')[0]
aperture_tables[objectname] = find_apertures(eps, width=width, title=objectname,
interactive=interactive,
maskname=maskname,
)
# Second, iterate through all slits again and perform spectral extraction
# using the apertures defined above
for i,eps_file in enumerate(eps_files):
objectname = objectnames[i]
sig_file = sig_files[i]
spectrum_plot_file = spectrum_plot_files[i]
fits_file = fits_files[i]
if len(aperture_tables[objectname]) == 0:
info('No apertures defined for {}. Skipping extraction.'.format(
objectname))
else:
info('Extracting spectra for {}'.format(objectname))
eps = fits.open(eps_file, 'readonly')[0]
sig = fits.open(sig_file, 'readonly')[0]
try:
optimal_extraction(eps, sig, aperture_tables[objectname],
fitsfileout=fits_file,
plotfileout=spectrum_plot_file,
)
except Exception as e:
warning('Failed to extract spectra for {}'.format(objectname))
warning(e)
def iterate_spatial_profile(P, DmS, V, f,\
smoothing=5, order=3, minpixels=50,\
sigma=4, nclip=2, verbose=True):
poly0 = models.Polynomial1D(degree=order)
fit_poly = fitting.LinearLSQFitter()
Pnew = np.zeros(P.shape)
for i,row in enumerate(P):
weights = f**2/V[i]
weights.mask = weights.mask | np.isnan(weights)
srow = np.ma.MaskedArray(data=signal.medfilt(row, smoothing),\
mask=(np.isnan(signal.medfilt(row, smoothing)) | weights.mask))
xcoord = np.ma.MaskedArray(data=np.arange(0,len(row),1),\
mask=srow.mask)
for iter in range(nclip+1):
nrej_before = np.sum(srow.mask)
fitted_poly = fit_poly(poly0,\
xcoord[~xcoord.mask], srow[~srow.mask],\
weights=weights[~weights.mask])
fit = np.array([fitted_poly(x) for x in xcoord])
resid = (DmS[i]-f*srow)**2 / V[i]
newmask = (resid > sigma)
weights.mask = weights.mask | newmask
srow.mask = srow.mask | newmask
xcoord.mask = xcoord.mask | newmask
nrej_after = np.sum(srow.mask)
if nrej_after > nrej_before:
if verbose:\
info('Row {:3d}: Rejected {:d} pixels on clipping '+\
'iteration {:d}'.format(\
i, nrej_after-nrej_before, iter))
## Reject row if too few pixels are availabel for the fit
if (srow.shape[0] - nrej_after) < minpixels:
if verbose:
warning('Row {:3d}: WARNING! Only {:d} pixels remain after '+\
'clipping and masking'.format(\
i, srow.shape[0] - nrej_after))
fit = np.zeros(fit.shape)
## Set negative values to zero
if np.sum((fit<0)) > 0 and verbose:
info('Row {:3d}: Reset {:d} negative pixels in fit to 0'.format(\
i, np.sum((fit<0))))
fit[(fit < 0)] = 0
Pnew[i] = fit
return Pnew
def readscitbl(path):
print(path)
hdulist = pf.open(path)
header = hdulist[0].header
try:
targs = hdulist[1].data
ssl = hdulist[2].data
msl = hdulist[3].data
asl = hdulist[4].data
except:
warning("Improper MOSFIRE FITS File: %s" % path)
hdulist.close()
return header, targs, ssl, msl, asl
def readscitbl(path):
print(path)
hdulist = pf.open(path)
header = hdulist[0].header
try:
targs = hdulist[1].data
ssl = hdulist[2].data
msl = hdulist[3].data
asl = hdulist[4].data
except:
warning("Improper MOSFIRE FITS File: %s" % path)
hdulist.close()
return header, targs, ssl, msl, asl
def polyfit_clip(xs, ys, order, nsig=2.5):
ff = np.poly1d(np.polyfit(xs, ys, order))
sd = np.std(ys - ff(xs))
if sd == 0.0:
warning('Clipping failed because stddev=0, using unclipped fit.')
result = ff
else:
r = np.abs(ys - ff(xs))
ok = r < (sd * nsig)
try:
result = np.polyfit(xs[ok], ys[ok], order)
except:
warning('Clipping failed, using unclipped fit.')
result = ff
return result
def fname_to_date_tuple(fname):
'''Take a filename like m120507_0123, return 12may07'''
months = {"01": "jan", "02": "feb", "03": "mar", "04": "apr", "05": "may",
"06": "jun", "07": "jul", "08": "aug", "09": "sep", "10": "oct",
"11": "nov", "12": "dec"}
if len(fname) != 17:
raise Exception("The file name '%s' is not of correct length. It "
"must be of the form mYYmmdd_nnnn.fits" % fname)
try:
fdate = fname.split("m")[1][0:6]
yr, mn, dy = "20" + fdate[0:2], fdate[2:4], int(fdate[4:6])
month = months[mn]
except:
warning("Could not parse date out of file name: %s" % (fname))
return yr, month, dy
def load_edges(maskname, band, options):
''' Load the slit edge functions. Returns (edges, metadata) '''
if False:
path = os.path.join(options["outdir"], maskname)
fn = os.path.join(path, "slit-edges_{0}.npy".format(band))
fn = "slit-edges_{0}.npy".format(band)
try:
edges = np.load(fn)
except:
error("Cannot load slit edges file")
raise Exception("Cannot load slit edges file")
edges,meta = edges[0:-1], edges[-1]
if meta['maskname'] != maskname:
warning("The maskname for the edge file '%s' does not match "
"that in the edge file '%s'" % (maskname, meta['maskname']))
warning("Continuing")
return edges, meta
def load_edges(maskname, band, options):
''' Load the slit edge functions. Returns (edges, metadata) '''
if False:
path = os.path.join(options["outdir"], maskname)
fn = os.path.join(path, "slit-edges_{0}.npy".format(band))
fn = "slit-edges_{0}.npy".format(band)
try:
edges = np.load(fn)
except:
error("Cannot load slit edges file")
raise Exception("Cannot load slit edges file")
edges, meta = edges[0:-1], edges[-1]
if meta['maskname'] != maskname:
warning("The maskname for the edge file '%s' does not match "
"that in the edge file '%s'" % (maskname, meta['maskname']))
warning("Continuing")
return edges, meta
def load_lambdaslit(fnum, maskname, band, options):
''' Load the wavelength coefficient functions '''
if False:
path = os.path.join(options["outdir"], maskname)
fn = os.path.join(path, "lambda_solution_{0}.fits".format(fnum))
fn = "lambda_solution_{0}.fits".format(fnum)
print(fn)
ret = readfits(fn, options)
if ret[0]['filter'] != band:
error("Band name mismatch")
raise Exception("band name mismatch")
if ret[0]['maskname'] != maskname:
warning("The maskname for the edge file '%s' does not match "
"that in the edge file '%s'" % (maskname, ret[0]['maskname']))
warning("Continuing")
return readfits(fn, options)
def load_lambdaslit(fnum, maskname, band, options):
''' Load the wavelength coefficient functions '''
if False:
path = os.path.join(options["outdir"], maskname)
fn = os.path.join(path, "lambda_solution_{0}.fits".format(fnum))
fn = "lambda_solution_{0}.fits".format(fnum)
print(fn)
ret = readfits(fn, options)
if ret[0]['filter'] != band:
error ("Band name mismatch")
raise Exception("band name mismatch")
if ret[0]['maskname'] != maskname:
warning("The maskname for the edge file '%s' does not match "
"that in the edge file '%s'" % (maskname, ret[0]['maskname']))
warning("Continuing")
return readfits(fn, options)
def background_subtract_helper(slitno):
'''
Background subtraction follows the methods outlined by Kelson (2003). Here
a background is estimated as a function of wavelength using B-splines and
subtracted off. The assumption is that background is primarily a function
of wavelength, and thus by sampling the background across the full 2-d
spectrum the background is sampled at much higher than the native spectral
resolution of mosfire.
Formally, the assumption that background is only a function of wavelength
is incorrect, and indeed a "transmission function" is estimated from the 2d
spectrum. This gives an estimate of the throughput of the slit and divided
out.
1. Extract the slit from the 2d image.
2. Convert the 2d spectrum into a 1d spectrum
3. Estimate transmission function
'''
global header, bs, edges, data, Var, itime, lam, sky_sub_out, sky_model_out, band
tick = time.time()
# 1
top = np.int(edges[slitno]["top"](1024))
bottom = np.int(edges[slitno]["bottom"](1024))
info("Background subtracting slit %i [%i,%i]" % (slitno, top, bottom))
pix = np.arange(2048)
xroi = slice(0, 2048)
yroi = slice(bottom, top)
stime = itime[yroi, xroi]
slit = data[yroi, xroi]
Var[np.logical_not(np.isfinite(Var))] = np.inf
lslit = lam[1][yroi, xroi]
# 2
xx = np.arange(slit.shape[1])
yy = np.arange(slit.shape[0])
X, Y = np.meshgrid(xx, yy)
train_roi = slice(5, -5)
ls = lslit[train_roi].flatten().filled(0)
ss = slit[train_roi].flatten()
ys = Y[train_roi].flatten()
dl = np.ma.median(np.diff(lslit[lslit.shape[0] // 2, :]))
if dl == 0:
return {"ok": False}
sort = np.argsort(ls)
ls = ls[sort]
ys = ys[sort]
hpps = np.array(Filters.hpp[band])
diff = np.append(np.diff(ls), False)
OK = (diff > 0.001) & (ls > hpps[0]) & (ls < hpps[1]) & (np.isfinite(ls)) \
& (np.isfinite(ss[sort]))
if len(np.where(OK)[0]) < 1000:
warning("Failed on slit " + str(slitno))
return {"ok": False}
# 3
pp = np.poly1d([1.0])
ss = (slit[train_roi] / pp(Y[train_roi])).flatten()
ss = ss[sort]
knotstart = max(hpps[0], min(ls[OK])) + 5
knotend = min(hpps[1], max(ls[OK])) - 5
for i in range(3):
try:
delta = dl * 0.9
knots = np.arange(knotstart, knotend, delta)
bspline = II.splrep(ls[OK], ss[OK], k=5, task=-1, t=knots)
except ValueError as e:
warning('Failed to fit spline with delta = {:5f}'.format(delta))
warning(str(e))
delta = dl * 1.4
info('Trying with delta = {:5f}'.format(delta))
knots = np.arange(knotstart, knotend, delta)
try:
bspline = II.splrep(ls[OK], ss[OK], k=5, task=-1, t=knots)
except ValueError as e:
warning("Could not construct spline on slit " + str(slitno))
warning(str(e))
return {"ok": False}
ll = lslit.flatten()
model = II.splev(ll, bspline)
oob = np.where((ll < knotstart) | (ll > knotend))
model[oob] = np.median(ss[~np.isnan(ss)])
model = model.reshape(slit.shape)
output = slit - model
std = np.abs(output) / (np.sqrt(np.abs(model) + 1))
tOK = (std[train_roi] < 10).flatten() & \
np.isfinite(std[train_roi]).flatten()
OK = OK & tOK[sort]
return {
"ok": True,
"slitno": slitno,
"bottom": bottom,
"top": top,
"output": output,
"model": model,
"bspline": bspline
}
def imcombine(files,
maskname,
options,
flat,
outname=None,
shifts=None,
extension=None):
'''
From a list of files it imcombine returns the imcombine of several values.
The imcombine code also estimates the readnoise ad RN/sqrt(numreads) so
that the variance per frame is equal to (ADU + RN^2) where RN is computed
in ADUs.
Arguments:
files[]: list of full path to files to combine
maskname: Name of mask
options: Options dictionary
flat[2048x2048]: Flat field (values should all be ~ 1.0)
outname: If set, will write (see notes below for details)
eps_[outname].fits: electron/sec file
itimes_[outname].fits: integration time
var_[outname].fits: Variance files
shifts[len(files)]: If set, will "roll" each file by the
amount in the shifts vector in pixels. This argument
is used when telescope tracking is poor. If you need
to use this, please notify Keck staff about poor
telescope tracking.
Returns 6-element tuple:
header: The combined header
electrons [2048x2048]: e- (in e- units)
var [2048x2048]: electrons + RN**2 (in e-^2 units)
bs: The MOSFIRE.Barset instance
itimes [2048x2048]: itimes (in s units)
Nframe: The number of frames that contribute to the summed
arrays above. If Nframe > 5 I use the sigma-clipping
Cosmic Ray Rejection tool. If Nframe < 5 then I drop
the max/min elements.
Notes:
header -- fits header
ADUs -- The mean # of ADUs per frame
var -- the Variance [in adu] per frame.
bs -- Barset
itimes -- The _total_ integration time in second
Nframe -- The number of frames in a stack.
Thus the number of electron per second is derived as:
e-/sec = (ADUs * Gain / Flat) * (Nframe/itimes)
The total number of electrons is:
el = ADUs * Gain * Nframe
'''
ADUs = np.zeros((len(files), 2048, 2048))
itimes = np.zeros((len(files), 2048, 2048))
prevssl = None
prevmn = None
patternid = None
maskname = None
header = None
if shifts is None:
shifts = np.zeros(len(files))
warnings.filterwarnings('ignore')
for i in range(len(files)):
fname = files[i]
thishdr, data, bs = IO.readmosfits(fname, options, extension=extension)
itimes[i, :, :] = thishdr["truitime"]
base = os.path.basename(fname).rstrip(".fits")
fnum = int(base.split("_")[1])
if shifts[i] == 0:
ADUs[i, :, :] = data.filled(0.0) / flat
else:
ADUs[i, :, :] = np.roll(data.filled(0.0) / flat,
np.int(shifts[i]),
axis=0)
''' Construct Header'''
if header is None:
header = thishdr
header["imfno%3.3i" % (fnum)] = (fname, "img%3.3i file name" % fnum)
list(
map(lambda x: rem_header_key(header, x), [
"CTYPE1", "CTYPE2", "WCSDIM", "CD1_1", "CD1_2", "CD2_1",
"CD2_2", "LTM1_1", "LTM2_2", "WAT0_001", "WAT1_001",
"WAT2_001", "CRVAL1", "CRVAL2", "CRPIX1", "CRPIX2", "RADECSYS"
]))
for card in header.cards:
if card == '': continue
key, val, comment = card
if key in thishdr:
if val != thishdr[key]:
newkey = key + ("_img%2.2i" % fnum)
try:
header[newkey.rstrip()] = (thishdr[key], comment)
except:
pass
''' Now handle error checking'''
if maskname is not None:
if thishdr["maskname"] != maskname:
error("File %s uses mask '%s' but the stack is of '%s'" %
(fname, thishdr["maskname"], maskname))
raise Exception(
"File %s uses mask '%s' but the stack is of '%s'" %
(fname, thishdr["maskname"], maskname))
maskname = thishdr["maskname"]
if thishdr["aborted"]:
error("Img '%s' was aborted and should not be used" % fname)
raise Exception("Img '%s' was aborted and should not be used" %
fname)
if prevssl is not None:
if len(prevssl) != len(bs.ssl):
# todo Improve these checks
error("The stack of input files seems to be of "
"different masks")
raise Exception("The stack of input files seems to be of "
"different masks")
prevssl = bs.ssl
if patternid is not None:
if patternid != thishdr["frameid"]:
error("The stack should be of '%s' frames only, but "
"the current image is a '%s' frame." %
(patternid, thishdr["frameid"]))
raise Exception("The stack should be of '%s' frames only, but "
"the current image is a '%s' frame." %
(patternid, thishdr["frameid"]))
patternid = thishdr["frameid"]
if maskname is not None:
if maskname != thishdr["maskname"]:
error("The stack should be of CSU mask '%s' frames "
"only but contains a frame of '%s'." %
(maskname, thishdr["maskname"]))
raise Exception("The stack should be of CSU mask '%s' frames "
"only but contains a frame of '%s'." %
(maskname, thishdr["maskname"]))
maskname = thishdr["maskname"]
if thishdr["BUNIT"] != "ADU per coadd":
error("The units of '%s' are not in ADU per coadd and "
"this violates an assumption of the DRP. Some new code "
"is needed in the DRP to handle the new units of "
"'%s'." % (fname, thishdr["BUNIT"]))
raise Exception(
"The units of '%s' are not in ADU per coadd and "
"this violates an assumption of the DRP. Some new code "
"is needed in the DRP to handle the new units of "
"'%s'." % (fname, thishdr["BUNIT"]))
''' Error checking is complete'''
debug("%s %s[%s]/%s: %5.1f s, Shift: %i px" %
(fname, maskname, patternid, header['filter'], np.mean(
itimes[i]), shifts[i]))
warnings.filterwarnings('always')
# the electrons and el_per_sec arrays are:
# [2048, 2048, len(files)] and contain values for
# each individual frame that is being combined.
# These need to be kept here for CRR reasons.
electrons = np.array(ADUs) * Detector.gain
el_per_sec = electrons / itimes
output = np.zeros((2048, 2048))
exptime = np.zeros((2048, 2048))
numreads = header["READS0"]
RN_adu = Detector.RN / np.sqrt(numreads) / Detector.gain
RN = Detector.RN / np.sqrt(numreads)
# Cosmic ray rejection code begins here. This code construction the
# electrons and itimes arrays.
standard = True
new_from_chuck = False
# Chuck Steidel has provided a modified version of the CRR procedure.
# to enable it, modify the variables above.
if new_from_chuck and not standard:
if len(files) >= 5:
print("Sigclip CRR")
srt = np.argsort(electrons, axis=0, kind='quicksort')
shp = el_per_sec.shape
sti = np.ogrid[0:shp[0], 0:shp[1], 0:shp[2]]
electrons = electrons[srt, sti[1], sti[2]]
el_per_sec = el_per_sec[srt, sti[1], sti[2]]
itimes = itimes[srt, sti[1], sti[2]]
# Construct the mean and standard deviation by dropping the top and bottom two
# electron fluxes. This is temporary.
mean = np.mean(el_per_sec[1:-1, :, :], axis=0)
std = np.std(el_per_sec[1:-1, :, :], axis=0)
drop = np.where((el_per_sec > (mean + std * 4))
| (el_per_sec < (mean - std * 4)))
print("dropping: ", len(drop[0]))
electrons[drop] = 0.0
itimes[drop] = 0.0
electrons = np.sum(electrons, axis=0)
itimes = np.sum(itimes, axis=0)
Nframe = len(files)
else:
warning("With less than 5 frames, the pipeline does NOT perform")
warning("Cosmic Ray Rejection.")
# the "if false" line disables cosmic ray rejection"
if False:
for i in range(len(files)):
el = electrons[i, :, :]
it = itimes[i, :, :]
el_mf = scipy.signal.medfilt(el, 5)
bad = np.abs(el - el_mf) / np.abs(el) > 10.0
el[bad] = 0.0
it[bad] = 0.0
electrons[i, :, :] = el
itimes[i, :, :] = it
electrons = np.sum(electrons, axis=0)
itimes = np.sum(itimes, axis=0)
Nframe = len(files)
if standard and not new_from_chuck:
if len(files) >= 9:
info("Sigclip CRR")
srt = np.argsort(electrons, axis=0, kind='quicksort')
shp = el_per_sec.shape
sti = np.ogrid[0:shp[0], 0:shp[1], 0:shp[2]]
electrons = electrons[srt, sti[1], sti[2]]
el_per_sec = el_per_sec[srt, sti[1], sti[2]]
itimes = itimes[srt, sti[1], sti[2]]
# Construct the mean and standard deviation by dropping the top and bottom two
# electron fluxes. This is temporary.
mean = np.mean(el_per_sec[2:-2, :, :], axis=0)
std = np.std(el_per_sec[2:-2, :, :], axis=0)
drop = np.where((el_per_sec > (mean + std * 4))
| (el_per_sec < (mean - std * 4)))
info("dropping: " + str(len(drop[0])))
electrons[drop] = 0.0
itimes[drop] = 0.0
electrons = np.sum(electrons, axis=0)
itimes = np.sum(itimes, axis=0)
Nframe = len(files)
elif len(files) > 5:
warning("WARNING: Drop min/max CRR")
srt = np.argsort(el_per_sec, axis=0)
shp = el_per_sec.shape
sti = np.ogrid[0:shp[0], 0:shp[1], 0:shp[2]]
electrons = electrons[srt, sti[1], sti[2]]
itimes = itimes[srt, sti[1], sti[2]]
electrons = np.sum(electrons[1:-1, :, :], axis=0)
itimes = np.sum(itimes[1:-1, :, :], axis=0)
Nframe = len(files) - 2
else:
warning("With less than 5 frames, the pipeline does NOT perform")
warning("Cosmic Ray Rejection.")
# the "if false" line disables cosmic ray rejection"
if False:
for i in range(len(files)):
el = electrons[i, :, :]
it = itimes[i, :, :]
# calculate the median image
el_mf = scipy.signal.medfilt(el, 5)
el_mf_large = scipy.signal.medfilt(el_mf, 15)
# LR: this is a modified version I was experimenting with. For the version
# written by Nick, see the new_from_chuck part of this code
# sky sub
el_sky_sub = el_mf - el_mf_large
# add a constant value
el_plus_constant = el_sky_sub + 100
bad = np.abs(el - el_mf) / np.abs(el_plus_constant) > 50.0
el[bad] = 0.0
it[bad] = 0.0
electrons[i, :, :] = el
itimes[i, :, :] = it
electrons = np.sum(electrons, axis=0)
itimes = np.sum(itimes, axis=0)
Nframe = len(files)
''' Now handle variance '''
numreads = header["READS0"]
RN_adu = Detector.RN / np.sqrt(numreads) / Detector.gain
RN = Detector.RN / np.sqrt(numreads)
var = (electrons + RN**2)
''' Now mask out bad pixels '''
electrons[data.mask] = np.nan
var[data.mask] = np.inf
if "RN" in header:
error("RN Already populated in header")
raise Exception("RN Already populated in header")
header['RN'] = ("%1.3f", "Read noise in e-")
header['NUMFRM'] = (Nframe, 'Typical number of frames in stack')
header['BUNIT'] = 'ELECTRONS/SECOND'
IO.writefits(np.float32(electrons / itimes),
maskname,
"eps_%s" % (outname),
options,
header=header,
overwrite=True)
# Update itimes after division in order to not introduce nans
itimes[data.mask] = 0.0
header['BUNIT'] = 'ELECTRONS^2'
IO.writefits(var,
maskname,
"var_%s" % (outname),
options,
header=header,
overwrite=True,
lossy_compress=True)
header['BUNIT'] = 'SECOND'
IO.writefits(np.float32(itimes),
maskname,
"itimes_%s" % (outname),
options,
header=header,
overwrite=True,
lossy_compress=True)
return header, electrons, var, bs, itimes, Nframe
def imcombine(filelist, out, options, method="average", reject="none",\
lsigma=3, hsigma=3, mclip=False,\
nlow=None, nhigh=None):
'''Combines images in input list with optional rejection algorithms.
Args:
filelist: The list of files to imcombine
out: The full path to the output file
method: either "average" or "median" combine
options: Options dictionary
bpmask: The full path to the bad pixel mask
reject: none, minmax, sigclip
nlow,nhigh: Parameters for minmax rejection, see iraf docs
mclip: use median as the function to calculate the baseline values for
sigclip rejection?
lsigma, hsigma: low and high sigma rejection thresholds.
Returns:
None
Side effects:
Creates the imcombined file at location `out'
'''
assert method in ['average', 'median']
if os.path.exists(out):
os.remove(out)
if reject == 'none':
info('Combining files using ccdproc.combine task')
info(' reject=none')
for file in filelist:
debug(' Combining: {}'.format(file))
ccdproc.combine(filelist, out, method=method,\
minmax_clip=False,\
iraf_minmax_clip=True,\
sigma_clip=False,\
unit="adu")
info(' Done.')
elif reject == 'minmax':
## The IRAF imcombine minmax rejection behavior is different than the
## ccdproc minmax rejection behavior. We are using the IRAF like
## behavior here. To support this a pull request for the ccdproc
## package has been made:
## https://github.com/astropy/ccdproc/pull/358
##
## Note that the ccdproc behavior still differs slightly from the
## nominal IRAF behavior in that the rejection does not consider whether
## any of the rejected pixels have been rejected for other reasons, so
## if nhigh=1 and that pixel was masked for some other reason, the
## new ccdproc algorithm, will not mask the next highest pixel, it will
## still just mask the highest pixel even if it is already masked.
##
## From IRAF (help imcombine):
## nlow = 1, nhigh = 1 (minmax)
## The number of low and high pixels to be rejected by the
## "minmax" algorithm. These numbers are converted to fractions
## of the total number of input images so that if no rejections
## have taken place the specified number of pixels are rejected
## while if pixels have been rejected by masking, thresholding, or
## non-overlap, then the fraction of the remaining pixels,
## truncated to an integer, is used.
##
## Check that minmax rejection is possible given the number of images
if nlow is None:
nlow = 0
if nhigh is None:
nhigh = 0
if nlow + nhigh >= len(filelist):
warning('nlow + nhigh >= number of input images. Combining without rejection')
nlow = 0
nhigh = 0
if ccdproc.version.major >= 1 and ccdproc.version.minor >= 1\
and ccdproc.version.release:
info('Combining files using ccdproc.combine task')
info(' reject=clip_extrema')
info(' nlow={}'.format(nlow))
info(' nhigh={}'.format(nhigh))
for file in filelist:
info(' {}'.format(file))
ccdproc.combine(filelist, out, method=method,\
minmax_clip=False,\
clip_extrema=True,\
nlow=nlow, nhigh=nhigh,\
sigma_clip=False,\
unit="adu")
info(' Done.')
else:
## If ccdproc does not have new rejection algorithm in:
## https://github.com/astropy/ccdproc/pull/358
## Manually perform rejection using ccdproc.combiner.Combiner object
info('Combining files using local clip_extrema rejection algorithm')
info('and the ccdproc.combiner.Combiner object.')
info(' reject=clip_extrema')
info(' nlow={}'.format(nlow))
info(' nhigh={}'.format(nhigh))
for file in filelist:
info(' {}'.format(file))
ccdlist = []
for file in filelist:
ccdlist.append(ccdproc.CCDData.read(file, unit='adu', hdu=0))
c = ccdproc.combiner.Combiner(ccdlist)
nimages, nx, ny = c.data_arr.mask.shape
argsorted = np.argsort(c.data_arr.data, axis=0)
mg = np.mgrid[0:nx,0:ny]
for i in range(-1*nhigh, nlow):
where = (argsorted[i,:,:].ravel(),
mg[0].ravel(),
mg[1].ravel())
c.data_arr.mask[where] = True
if method == 'average':
result = c.average_combine()
elif method == 'median':
result = c.median_combine()
for key in list(ccdlist[0].header.keys()):
header_entry = ccdlist[0].header[key]
if key != 'COMMENT':
result.header[key] = (header_entry,
ccdlist[0].header.comments[key])
hdul = result.to_hdu()
# print(hdul)
# for hdu in hdul:
# print(type(hdu.data))
hdul[0].writeto(out)
# result.write(out)
info(' Done.')
elif reject == 'sigclip':
info('Combining files using ccdproc.combine task')
info(' reject=sigclip')
info(' mclip={}'.format(mclip))
info(' lsigma={}'.format(lsigma))
info(' hsigma={}'.format(hsigma))
baseline_func = {False: np.mean, True: np.median}
ccdproc.combine(filelist, out, method=method,\
minmax_clip=False,\
clip_extrema=False,\
sigma_clip=True,\
sigma_clip_low_thresh=lsigma,\
sigma_clip_high_thresh=hsigma,\
sigma_clip_func=baseline_func[mclip],\
sigma_clip_dev_func=np.std,\
)
info(' Done.')
else:
raise NotImplementedError('{} rejection unrecognized by MOSFIRE DRP'.format(reject))
def imcombine(filelist, out, options, method="average", reject="none",\
lsigma=3, hsigma=3, mclip=False,\
nlow=None, nhigh=None):
'''Combines images in input list with optional rejection algorithms.
Args:
filelist: The list of files to imcombine
out: The full path to the output file
method: either "average" or "median" combine
options: Options dictionary
bpmask: The full path to the bad pixel mask
reject: none, minmax, sigclip
nlow,nhigh: Parameters for minmax rejection, see iraf docs
mclip: use median as the function to calculate the baseline values for
sigclip rejection?
lsigma, hsigma: low and high sigma rejection thresholds.
Returns:
None
Side effects:
Creates the imcombined file at location `out'
'''
assert method in ['average', 'median']
if os.path.exists(out):
os.remove(out)
if reject == 'none':
info('Combining files using ccdproc.combine task')
info(' reject=none')
for file in filelist:
debug(' Combining: {}'.format(file))
ccdproc.combine(filelist, out, method=method,\
minmax_clip=False,\
iraf_minmax_clip=True,\
sigma_clip=False,\
unit="adu")
info(' Done.')
elif reject == 'minmax':
## The IRAF imcombine minmax rejection behavior is different than the
## ccdproc minmax rejection behavior. We are using the IRAF like
## behavior here. To support this a pull request for the ccdproc
## package has been made:
## https://github.com/astropy/ccdproc/pull/358
##
## Note that the ccdproc behavior still differs slightly from the
## nominal IRAF behavior in that the rejection does not consider whether
## any of the rejected pixels have been rejected for other reasons, so
## if nhigh=1 and that pixel was masked for some other reason, the
## new ccdproc algorithm, will not mask the next highest pixel, it will
## still just mask the highest pixel even if it is already masked.
##
## From IRAF (help imcombine):
## nlow = 1, nhigh = 1 (minmax)
## The number of low and high pixels to be rejected by the
## "minmax" algorithm. These numbers are converted to fractions
## of the total number of input images so that if no rejections
## have taken place the specified number of pixels are rejected
## while if pixels have been rejected by masking, thresholding, or
## non-overlap, then the fraction of the remaining pixels,
## truncated to an integer, is used.
##
## Check that minmax rejection is possible given the number of images
if nlow is None:
nlow = 0
if nhigh is None:
nhigh = 0
if nlow + nhigh >= len(filelist):
warning(
'nlow + nhigh >= number of input images. Combining without rejection'
)
nlow = 0
nhigh = 0
if ccdproc.version.major >= 1 and ccdproc.version.minor >= 1\
and ccdproc.version.release:
info('Combining files using ccdproc.combine task')
info(' reject=clip_extrema')
info(' nlow={}'.format(nlow))
info(' nhigh={}'.format(nhigh))
for file in filelist:
info(' {}'.format(file))
ccdproc.combine(filelist, out, method=method,\
minmax_clip=False,\
clip_extrema=True,\
nlow=nlow, nhigh=nhigh,\
sigma_clip=False,\
unit="adu")
info(' Done.')
else:
## If ccdproc does not have new rejection algorithm in:
## https://github.com/astropy/ccdproc/pull/358
## Manually perform rejection using ccdproc.combiner.Combiner object
info(
'Combining files using local clip_extrema rejection algorithm')
info('and the ccdproc.combiner.Combiner object.')
info(' reject=clip_extrema')
info(' nlow={}'.format(nlow))
info(' nhigh={}'.format(nhigh))
for file in filelist:
info(' {}'.format(file))
ccdlist = []
for file in filelist:
ccdlist.append(ccdproc.CCDData.read(file, unit='adu', hdu=0))
c = ccdproc.combiner.Combiner(ccdlist)
nimages, nx, ny = c.data_arr.mask.shape
argsorted = np.argsort(c.data_arr.data, axis=0)
mg = np.mgrid[0:nx, 0:ny]
for i in range(-1 * nhigh, nlow):
where = (argsorted[i, :, :].ravel(), mg[0].ravel(),
mg[1].ravel())
c.data_arr.mask[where] = True
if method == 'average':
result = c.average_combine()
elif method == 'median':
result = c.median_combine()
for key in list(ccdlist[0].header.keys()):
header_entry = ccdlist[0].header[key]
if key != 'COMMENT':
result.header[key] = (header_entry,
ccdlist[0].header.comments[key])
hdul = result.to_hdu()
# print(hdul)
# for hdu in hdul:
# print(type(hdu.data))
hdul[0].writeto(out)
# result.write(out)
info(' Done.')
elif reject == 'sigclip':
info('Combining files using ccdproc.combine task')
info(' reject=sigclip')
info(' mclip={}'.format(mclip))
info(' lsigma={}'.format(lsigma))
info(' hsigma={}'.format(hsigma))
baseline_func = {False: np.mean, True: np.median}
ccdproc.combine(filelist, out, method=method,\
minmax_clip=False,\
clip_extrema=False,\
sigma_clip=True,\
sigma_clip_low_thresh=lsigma,\
sigma_clip_high_thresh=hsigma,\
sigma_clip_func=baseline_func[mclip],\
sigma_clip_dev_func=np.std,\
)
info(' Done.')
else:
raise NotImplementedError(
'{} rejection unrecognized by MOSFIRE DRP'.format(reject))
def imcombine(files, maskname, options, flat, outname=None, shifts=None,
extension=None):
'''
From a list of files it imcombine returns the imcombine of several values.
The imcombine code also estimates the readnoise ad RN/sqrt(numreads) so
that the variance per frame is equal to (ADU + RN^2) where RN is computed
in ADUs.
Arguments:
files[]: list of full path to files to combine
maskname: Name of mask
options: Options dictionary
flat[2048x2048]: Flat field (values should all be ~ 1.0)
outname: If set, will write (see notes below for details)
eps_[outname].fits: electron/sec file
itimes_[outname].fits: integration time
var_[outname].fits: Variance files
shifts[len(files)]: If set, will "roll" each file by the
amount in the shifts vector in pixels. This argument
is used when telescope tracking is poor. If you need
to use this, please notify Keck staff about poor
telescope tracking.
Returns 6-element tuple:
header: The combined header
electrons [2048x2048]: e- (in e- units)
var [2048x2048]: electrons + RN**2 (in e-^2 units)
bs: The MOSFIRE.Barset instance
itimes [2048x2048]: itimes (in s units)
Nframe: The number of frames that contribute to the summed
arrays above. If Nframe > 5 I use the sigma-clipping
Cosmic Ray Rejection tool. If Nframe < 5 then I drop
the max/min elements.
Notes:
header -- fits header
ADUs -- The mean # of ADUs per frame
var -- the Variance [in adu] per frame.
bs -- Barset
itimes -- The _total_ integration time in second
Nframe -- The number of frames in a stack.
Thus the number of electron per second is derived as:
e-/sec = (ADUs * Gain / Flat) * (Nframe/itimes)
The total number of electrons is:
el = ADUs * Gain * Nframe
'''
ADUs = np.zeros((len(files), 2048, 2048))
itimes = np.zeros((len(files), 2048, 2048))
prevssl = None
prevmn = None
patternid = None
maskname = None
header = None
if shifts is None:
shifts = np.zeros(len(files))
warnings.filterwarnings('ignore')
for i in range(len(files)):
fname = files[i]
thishdr, data, bs = IO.readmosfits(fname, options, extension=extension)
itimes[i,:,:] = thishdr["truitime"]
base = os.path.basename(fname).rstrip(".fits")
fnum = int(base.split("_")[1])
if shifts[i] == 0:
ADUs[i,:,:] = data.filled(0.0) / flat
else:
ADUs[i,:,:] = np.roll(data.filled(0.0) / flat, np.int(shifts[i]), axis=0)
''' Construct Header'''
if header is None:
header = thishdr
header["imfno%3.3i" % (fnum)] = (fname, "img%3.3i file name" % fnum)
list(map(lambda x: rem_header_key(header, x), ["CTYPE1", "CTYPE2", "WCSDIM",
"CD1_1", "CD1_2", "CD2_1", "CD2_2", "LTM1_1", "LTM2_2", "WAT0_001",
"WAT1_001", "WAT2_001", "CRVAL1", "CRVAL2", "CRPIX1", "CRPIX2",
"RADECSYS"]))
for card in header.cards:
if card == '': continue
key,val,comment = card
if key in thishdr:
if val != thishdr[key]:
newkey = key + ("_img%2.2i" % fnum)
try: header[newkey.rstrip()] = (thishdr[key], comment)
except: pass
''' Now handle error checking'''
if maskname is not None:
if thishdr["maskname"] != maskname:
error("File %s uses mask '%s' but the stack is of '%s'" %
(fname, thishdr["maskname"], maskname))
raise Exception("File %s uses mask '%s' but the stack is of '%s'" %
(fname, thishdr["maskname"], maskname))
maskname = thishdr["maskname"]
if thishdr["aborted"]:
error("Img '%s' was aborted and should not be used" %
fname)
raise Exception("Img '%s' was aborted and should not be used" %
fname)
if prevssl is not None:
if len(prevssl) != len(bs.ssl):
# todo Improve these checks
error("The stack of input files seems to be of "
"different masks")
raise Exception("The stack of input files seems to be of "
"different masks")
prevssl = bs.ssl
if patternid is not None:
if patternid != thishdr["frameid"]:
error("The stack should be of '%s' frames only, but "
"the current image is a '%s' frame." % (patternid,
thishdr["frameid"]))
raise Exception("The stack should be of '%s' frames only, but "
"the current image is a '%s' frame." % (patternid,
thishdr["frameid"]))
patternid = thishdr["frameid"]
if maskname is not None:
if maskname != thishdr["maskname"]:
error("The stack should be of CSU mask '%s' frames "
"only but contains a frame of '%s'." % (maskname,
thishdr["maskname"]))
raise Exception("The stack should be of CSU mask '%s' frames "
"only but contains a frame of '%s'." % (maskname,
thishdr["maskname"]))
maskname = thishdr["maskname"]
if thishdr["BUNIT"] != "ADU per coadd":
error("The units of '%s' are not in ADU per coadd and "
"this violates an assumption of the DRP. Some new code "
"is needed in the DRP to handle the new units of "
"'%s'." % (fname, thishdr["BUNIT"]))
raise Exception("The units of '%s' are not in ADU per coadd and "
"this violates an assumption of the DRP. Some new code "
"is needed in the DRP to handle the new units of "
"'%s'." % (fname, thishdr["BUNIT"]))
''' Error checking is complete'''
debug("%s %s[%s]/%s: %5.1f s, Shift: %i px" % (fname, maskname, patternid,
header['filter'], np.mean(itimes[i]), shifts[i]))
warnings.filterwarnings('always')
# the electrons and el_per_sec arrays are:
# [2048, 2048, len(files)] and contain values for
# each individual frame that is being combined.
# These need to be kept here for CRR reasons.
electrons = np.array(ADUs) * Detector.gain
el_per_sec = electrons / itimes
output = np.zeros((2048, 2048))
exptime = np.zeros((2048, 2048))
numreads = header["READS0"]
RN_adu = Detector.RN / np.sqrt(numreads) / Detector.gain
RN = Detector.RN / np.sqrt(numreads)
# Cosmic ray rejection code begins here. This code construction the
# electrons and itimes arrays.
standard = True
new_from_chuck = False
# Chuck Steidel has provided a modified version of the CRR procedure.
# to enable it, modify the variables above.
if new_from_chuck and not standard:
if len(files) >= 5:
print("Sigclip CRR")
srt = np.argsort(electrons, axis=0, kind='quicksort')
shp = el_per_sec.shape
sti = np.ogrid[0:shp[0], 0:shp[1], 0:shp[2]]
electrons = electrons[srt, sti[1], sti[2]]
el_per_sec = el_per_sec[srt, sti[1], sti[2]]
itimes = itimes[srt, sti[1], sti[2]]
# Construct the mean and standard deviation by dropping the top and bottom two
# electron fluxes. This is temporary.
mean = np.mean(el_per_sec[1:-1,:,:], axis = 0)
std = np.std(el_per_sec[1:-1,:,:], axis = 0)
drop = np.where( (el_per_sec > (mean+std*4)) | (el_per_sec < (mean-std*4)) )
print("dropping: ", len(drop[0]))
electrons[drop] = 0.0
itimes[drop] = 0.0
electrons = np.sum(electrons, axis=0)
itimes = np.sum(itimes, axis=0)
Nframe = len(files)
else:
warning( "With less than 5 frames, the pipeline does NOT perform")
warning( "Cosmic Ray Rejection.")
# the "if false" line disables cosmic ray rejection"
if False:
for i in range(len(files)):
el = electrons[i,:,:]
it = itimes[i,:,:]
el_mf = scipy.signal.medfilt(el, 5)
bad = np.abs(el - el_mf) / np.abs(el) > 10.0
el[bad] = 0.0
it[bad] = 0.0
electrons[i,:,:] = el
itimes[i,:,:] = it
electrons = np.sum(electrons, axis=0)
itimes = np.sum(itimes, axis=0)
Nframe = len(files)
if standard and not new_from_chuck:
if len(files) >= 9:
info("Sigclip CRR")
srt = np.argsort(electrons, axis=0, kind='quicksort')
shp = el_per_sec.shape
sti = np.ogrid[0:shp[0], 0:shp[1], 0:shp[2]]
electrons = electrons[srt, sti[1], sti[2]]
el_per_sec = el_per_sec[srt, sti[1], sti[2]]
itimes = itimes[srt, sti[1], sti[2]]
# Construct the mean and standard deviation by dropping the top and bottom two
# electron fluxes. This is temporary.
mean = np.mean(el_per_sec[2:-2,:,:], axis = 0)
std = np.std(el_per_sec[2:-2,:,:], axis = 0)
drop = np.where( (el_per_sec > (mean+std*4)) | (el_per_sec < (mean-std*4)) )
info("dropping: "+str(len(drop[0])))
electrons[drop] = 0.0
itimes[drop] = 0.0
electrons = np.sum(electrons, axis=0)
itimes = np.sum(itimes, axis=0)
Nframe = len(files)
elif len(files) > 5:
warning( "WARNING: Drop min/max CRR")
srt = np.argsort(el_per_sec,axis=0)
shp = el_per_sec.shape
sti = np.ogrid[0:shp[0], 0:shp[1], 0:shp[2]]
electrons = electrons[srt, sti[1], sti[2]]
itimes = itimes[srt, sti[1], sti[2]]
electrons = np.sum(electrons[1:-1,:,:], axis=0)
itimes = np.sum(itimes[1:-1,:,:], axis=0)
Nframe = len(files) - 2
else:
warning( "With less than 5 frames, the pipeline does NOT perform")
warning( "Cosmic Ray Rejection.")
# the "if false" line disables cosmic ray rejection"
if False:
for i in range(len(files)):
el = electrons[i,:,:]
it = itimes[i,:,:]
# calculate the median image
el_mf = scipy.signal.medfilt(el, 5)
el_mf_large = scipy.signal.medfilt(el_mf, 15)
# LR: this is a modified version I was experimenting with. For the version
# written by Nick, see the new_from_chuck part of this code
# sky sub
el_sky_sub = el_mf - el_mf_large
# add a constant value
el_plus_constant = el_sky_sub + 100
bad = np.abs(el - el_mf) / np.abs(el_plus_constant) > 50.0
el[bad] = 0.0
it[bad] = 0.0
electrons[i,:,:] = el
itimes[i,:,:] = it
electrons = np.sum(electrons, axis=0)
itimes = np.sum(itimes, axis=0)
Nframe = len(files)
''' Now handle variance '''
numreads = header["READS0"]
RN_adu = Detector.RN / np.sqrt(numreads) / Detector.gain
RN = Detector.RN / np.sqrt(numreads)
var = (electrons + RN**2)
''' Now mask out bad pixels '''
electrons[data.mask] = np.nan
var[data.mask] = np.inf
if "RN" in header:
error("RN Already populated in header")
raise Exception("RN Already populated in header")
header['RN'] = ("%1.3f" , "Read noise in e-")
header['NUMFRM'] = (Nframe, 'Typical number of frames in stack')
header['BUNIT'] = 'ELECTRONS/SECOND'
IO.writefits(np.float32(electrons/itimes), maskname, "eps_%s" % (outname),
options, header=header, overwrite=True)
# Update itimes after division in order to not introduce nans
itimes[data.mask] = 0.0
header['BUNIT'] = 'ELECTRONS^2'
IO.writefits(var, maskname, "var_%s" % (outname),
options, header=header, overwrite=True, lossy_compress=True)
header['BUNIT'] = 'SECOND'
IO.writefits(np.float32(itimes), maskname, "itimes_%s" % (outname),
options, header=header, overwrite=True, lossy_compress=True)
return header, electrons, var, bs, itimes, Nframe
def background_subtract_helper(slitno):
'''
Background subtraction follows the methods outlined by Kelson (2003). Here
a background is estimated as a function of wavelength using B-splines and
subtracted off. The assumption is that background is primarily a function
of wavelength, and thus by sampling the background across the full 2-d
spectrum the background is sampled at much higher than the native spectral
resolution of mosfire.
Formally, the assumption that background is only a function of wavelength
is incorrect, and indeed a "transmission function" is estimated from the 2d
spectrum. This gives an estimate of the throughput of the slit and divided
out.
1. Extract the slit from the 2d image.
2. Convert the 2d spectrum into a 1d spectrum
3. Estimate transmission function
'''
global header, bs, edges, data, Var, itime, lam, sky_sub_out, sky_model_out, band
tick = time.time()
# 1
top = np.int(edges[slitno]["top"](1024))
bottom = np.int(edges[slitno]["bottom"](1024))
info("Background subtracting slit %i [%i,%i]" % (slitno, top, bottom))
pix = np.arange(2048)
xroi = slice(0,2048)
yroi = slice(bottom, top)
stime = itime[yroi, xroi]
slit = data[yroi, xroi]
Var[np.logical_not(np.isfinite(Var))] = np.inf
lslit = lam[1][yroi,xroi]
# 2
xx = np.arange(slit.shape[1])
yy = np.arange(slit.shape[0])
X,Y = np.meshgrid(xx,yy)
train_roi = slice(5,-5)
ls = lslit[train_roi].flatten().filled(0)
ss = slit[train_roi].flatten()
ys = Y[train_roi].flatten()
dl = np.ma.median(np.diff(lslit[lslit.shape[0]//2,:]))
if dl == 0:
return {"ok": False}
sort = np.argsort(ls)
ls = ls[sort]
ys = ys[sort]
hpps = np.array(Filters.hpp[band])
diff = np.append(np.diff(ls), False)
OK = (diff > 0.001) & (ls > hpps[0]) & (ls < hpps[1]) & (np.isfinite(ls)) \
& (np.isfinite(ss[sort]))
if len(np.where(OK)[0]) < 1000:
warning("Failed on slit "+str(slitno))
return {"ok": False}
# 3
pp = np.poly1d([1.0])
ss = (slit[train_roi] / pp(Y[train_roi])).flatten()
ss = ss[sort]
knotstart = max(hpps[0], min(ls[OK])) + 5
knotend = min(hpps[1], max(ls[OK])) - 5
for i in range(3):
try:
delta = dl*0.9
knots = np.arange(knotstart, knotend, delta)
bspline = II.splrep(ls[OK], ss[OK], k=5, task=-1, t=knots)
except ValueError as e:
warning('Failed to fit spline with delta = {:5f}'.format(delta))
warning(str(e))
delta = dl*1.4
info('Trying with delta = {:5f}'.format(delta))
knots = np.arange(knotstart, knotend, delta)
try:
bspline = II.splrep(ls[OK], ss[OK], k=5, task=-1, t=knots)
except ValueError as e:
warning("Could not construct spline on slit "+str(slitno))
warning(str(e))
return {"ok": False}
ll = lslit.flatten()
model = II.splev(ll, bspline)
oob = np.where((ll < knotstart) | (ll > knotend))
model[oob] = np.median(ss[~np.isnan(ss)])
model = model.reshape(slit.shape)
output = slit - model
std = np.abs(output)/(np.sqrt(np.abs(model)+1))
tOK = (std[train_roi] < 10).flatten() & \
np.isfinite(std[train_roi]).flatten()
OK = OK & tOK[sort]
return {"ok": True, "slitno": slitno, "bottom": bottom, "top": top,
"output": output, "model": model, "bspline": bspline}