def compute_lincor_model(hdu, pix_x, pix_y, ndegree=3):
"""
Compute the lineary correction model
Parameters
----------
hdu : astropy.fits.hdu
header data unit
pix_x, pix_y: int
x, y pixel coordiates
Returns
-------
cormod : astropy.modeling.model
model the linearity correction with x = measured DN
"""
# for fitting
x = []
y = []
# get all the diffs
for k in range(nints):
gnum, ydata = get_ramp(hdu, pix_x, pix_y, k)
ggnum, gdata, aveDN, diffDN = get_good_ramp(gnum, ydata)
# accumulate data
x.append(aveDN)
y.append(diffDN)
# find the ramp that spans the largest range of DN
# and save some info -> needed for creating the correction
# if (gdata.max() - gdata.min()) > mm_delta:
# mm_delta = gdata.max() - gdata.min()
if k == 3:
max_ramp_k = k
# max_ramp_gdata = gdata
max_ramp_aveDN = aveDN
# fit the aveDN versus diffDN combined data from all integrations
x = np.concatenate(x)
y = np.concatenate(y)
mod = fit_diffs(x, y, ndegree=ndegree - 1)
# determine the startDN for all ramps (all not actually needed)
startDNvals = np.arange(0.0, 20000.0, 200.0)
chival = np.zeros((nints, len(startDNvals)))
ints = range(nints)
for i, startDN in enumerate(startDNvals):
(DN_exp, cor, cor_mod) = calc_lincor(mod[1],
max_ramp_aveDN,
startDN,
ndegree=ndegree)
for k in ints:
gnum, ydata = get_ramp(hdu, pix_x, pix_y, k)
ggnum, gdata, aveDN, diffDN = get_good_ramp(gnum, ydata)
# correct the ramps
ycor = cor_mod(gdata)
gdata_cor = gdata * ycor
# calculate the chisqr for each integration set of differences
# from the expected flat line
diffDN = np.diff(gdata_cor)
aveDN = 0.5 * (gdata[:-1] + gdata[1:])
cindxs, = np.where(aveDN > 10000.0)
chival[k, i] = np.sum((diffDN[aveDN > 15000.0] - mod[1].c0)**2)
minindx = np.zeros((nints), dtype=int)
for k in ints[1:]:
minindx[k] = np.argmin(chival[k, :])
# only use the startDN from one ramp
startDN = startDNvals[minindx[max_ramp_k]]
# get the correction
(DN_exp, cor, cor_mod) = calc_lincor(mod[1], max_ramp_aveDN, startDN)
return cor_mod
# plot the 2pt diffs versus average DN
# ax[1].plot(aveDN, diffDN, label=f"Int #{k+1}", color=pcol[k])
# if k == 0:
# ax[1].set_ylim(0.9 * min(diffDN), 1.4 * max(diffDN))
# accumulate data for fitting
x.append(aveDN)
y.append(diffDN)
if z == 0:
# fit the aveDN versus diffDN combined data from all integrations
# e.g., derive the non-linearity correction
x = np.concatenate(x)
y = np.concatenate(y)
mod = fit_diffs(x, y, noexp=noexp)
if noexp:
polymod = mod
else:
polymod = mod[2]
lincormod = polymod
# more plots
ints = range(nints)
# setup ramp fit
line_init = Linear1D()
fit_line = LinearLSQFitter()
mult_comp = False
intslopes = np.zeros((nints))
x.append(aveDN)
y.append(diffDN)
# find the ramp that spans the largest range of DN
# and save some info -> needed for creating the correction
# if (gdata.max() - gdata.min()) > mm_delta:
# mm_delta = gdata.max() - gdata.min()
if k == 3:
max_ramp_k = k
max_ramp_gdata = gdata
max_ramp_aveDN = aveDN
# fit the aveDN versus diffDN combined data from all integrations
x = np.concatenate(x)
y = np.concatenate(y)
mod = fit_diffs(x, y)
polymod = mod[2]
# plot the model
mod_x = np.linspace(0.0, 65000.0, num=100)
# for k in range(nints):
# gnum, ydata = get_ramp(hdu[0], pix_x, pix_y, k)
# ggnum, gdata, aveDN, diffDN = get_good_ramp(gnum, ydata)
# (DN_exp, cor, cor_mod) = calc_lincor(mod[1], gdata, args.startDN)
# ax[3].plot(DN_exp, cor, "--", label=f"Int #{k+1}")
startDNvals = np.arange(0.0, 20000.0, 200.0)
chival = np.zeros((nints, len(startDNvals)))
ints = range(nints)