def get_smooth_continuum(s, wvl=None):
"""
wvl : required for masking our some absorption features
"""
from libs.trace_flat import get_finite_boundary_indices
k1, k2 = get_finite_boundary_indices(s)
if k1 == k2:
r = np.empty(len(s), dtype="d")
r.fill(np.nan)
return r
sl = slice(k1, k2+1)
#s1m = ni.median_filter(np.array(s1[sl]), 150)
s1m = np.array(s[sl])
if wvl is not None:
wvl11 = np.array(wvl[sl])
for ww1, ww2 in [(1.9112, 1.9119),
(1.91946, 1.92139),
(1.92826, 1.92901),
(1.92372, 1.92457)]:
msk = (ww1 < wvl11) & (wvl11 < ww2)
s1m[msk] = np.nan
if len(s1m) > 351:
f12 = sv_iter(s1m, winsize1=351, winsize2=91)
elif len(s1m) > 25:
f12 = sv_iter(s1m, winsize1=25, winsize2=11)
else:
f12 = None
r = np.empty(len(s), dtype="d")
r.fill(np.nan)
if f12 is not None:
r[sl] = f12
return r
def get_sv_mask(self, ratio):
# use sv_iter to derive continuum level from ratio
from libs.smooth_continuum import sv_iter
from libs.trace_flat import get_finite_boundary_indices
# naive iterative approach.
ratio = ratio.copy()
for thresh in [0.05, 0.03, 0.01]: #, 0.005, 0.0001]:
i1, i2 = get_finite_boundary_indices(ratio)
sl2 = slice(i1, i2+1)
f12 = sv_iter(ratio[sl2], winsize1=251, winsize2=151)
#fl = s1_orig_n[sl2]/f12 # s1m : normalization factor
# f12 : flattening curve
# fl : flattened spectrum
new_msk = np.abs(ratio[sl2]-f12) > thresh
ratio[sl2][new_msk] = np.nan
# # recover masked area due to underestimated continuum
# ratio0 = s1_orig_n/tt # model corrected spec
# mmm = ratio0[sl2] > f12
# ratio[sl2][mmm] = ratio0[sl2][mmm]
# for thresh in [0.01]:
# i1, i2 = get_finite_boundary_indices(ratio)
# sl2 = slice(i1, i2+1)
# f12 = sv_iter(ratio[sl2], winsize1=351, winsize2=261)
# fl = (s1_orig/s1m)[sl2]/f12 # s1m : normalization factor
# # f12 : flattening curve
# # fl : flattened spectrum
# new_msk = np.abs(ratio[sl2]-f12) > thresh
# ratio[sl2][new_msk] = np.nan
msk = ~np.isfinite(ratio)
return msk
a0v_ss = a0v1/f12
a0v_ss[f12<np.nanmax(f12)*0.1] = np.nan
ax3.plot(wvl1, a0v_ss/np.nanmax(a0v_ss[100:-100]))
#ax3.set_ylim(0.7, 1.1)
#ax3.plot(wvl1[sl], a0v_ss/np.median(a0v_ss)*0.02)
#ax3.set_ylim(-0.02, 0.05)
if 0:
#for s1, a0v1, wvl1 in zip(specs, dd, wvl_sol):
#s1 = specs[ii]
#a0v1 = dd[ii]
from libs.trace_flat import get_finite_boundary_indices
k1, k2 = get_finite_boundary_indices(s1)
sl = slice(k1, k2+1)
#s1m = ni.median_filter(np.array(s1[sl]), 150)
s1m = np.array(s1[sl])
wvl11 = np.array(wvl1[sl])
for ww1, ww2 in [(1.9112, 1.9119),
(1.91946, 1.92139),
(1.92826, 1.92901),
(1.92372, 1.92457)]:
msk = (ww1 < wvl11) & (wvl11 < ww2)
s1m[msk] = np.nan
def flatten_deprecated(self, w1, s1_orig, dw_opt, gw_opt, try_small=True,
ax1=None, ax2=None):
tel_interp1d_f = self.tel_interp1d_f
s_a0v = self.a0v_interp1d(w1+dw_opt)
s1_orig = s1_orig / s_a0v
tel_interp1d = tel_interp1d_f(gw_opt)
tt = tel_interp1d(w1+dw_opt) # telluric model
if ax2:
# plot telluric model
ax2.plot(w1, tt, alpha=0.2, lw=3)
s1m = np.nanmax(s1_orig)
#s1m = 1.
s1_orig_n = s1_orig/s1m # normalize
# plot raw spec
if ax1:
ax1.plot(w1, s1_orig, alpha=0.3)
ratio = s1_orig_n/tt # model corrected spec
thresh=0.5
ratio[tt<thresh] = np.nan # blank out area with low tel-trans.
# now we blank out area where derivative varies significantly.
# derivative of ratio
ratiom = ni.gaussian_filter1d(ratio, 3, order=1)
# do median filtering to derive a baseline for derivative.
# Because of nan values of original spec. doing it with
# single value of filter size does not seem to work well.
# Instead, we first filter with large filter size, then
# with smaller one. Then replaces nan of former with
# result of latter.
dd = ni.median_filter(ratiom, 11)
dd_small = ni.median_filter(ratiom, 5)
dd_msk = ~np.isfinite(dd)
dd[dd_msk] = dd_small[dd_msk]
# subtract baseline from derivative.
dd1 = (ratiom - dd)
# mask out areas of high deriv.
# The threshold is arbitrary but seem to work for most of orders.
msk = np.abs(dd1) > 0.0005
#msk = np.abs(dd1) > 0.0003# for H
# do binary-closing for mask
msk = ni.binary_closing(msk, iterations=1) | ~np.isfinite(dd1)
dd1[msk] = np.nan
if ax1:
# plot mask
ax1.plot(w1, dd1)
if ax2:
# plot mask
ax2.plot(w1, dd1)
# now, mask out msk from original ratio.
ratio[msk] = np.nan
ratio[:4] = np.nan
ratio[-4:] = np.nan
# use sv_iter to derive continuum level from ratio
from libs.smooth_continuum import sv_iter
from libs.trace_flat import get_finite_boundary_indices
# naive iterative approach.
for thresh in [0.05, 0.03, 0.01]:
i1, i2 = get_finite_boundary_indices(ratio)
sl2 = slice(i1, i2+1)
f12 = sv_iter(ratio[sl2], winsize1=251, winsize2=151)
fl = (s1_orig/s1m)[sl2]/f12 # s1m : normalization factor
# f12 : flattening curve
# fl : flattened spectrum
new_msk = np.abs(ratio[sl2]-f12) > thresh
ratio[sl2][new_msk] = np.nan
# recover masked area due to underestimated continuum
ratio0 = s1_orig_n/tt # model corrected spec
mmm = ratio0[sl2] > f12
ratio[sl2][mmm] = ratio0[sl2][mmm]
for thresh in [0.01]:
i1, i2 = get_finite_boundary_indices(ratio)
sl2 = slice(i1, i2+1)
f12 = sv_iter(ratio[sl2], winsize1=351, winsize2=261)
fl = (s1_orig/s1m)[sl2]/f12 # s1m : normalization factor
# f12 : flattening curve
# fl : flattened spectrum
new_msk = np.abs(ratio[sl2]-f12) > thresh
ratio[sl2][new_msk] = np.nan
if try_small:
i1, i2 = get_finite_boundary_indices(ratio)
sl2 = slice(i1, i2+1)
f12 = sv_iter(ratio[sl2], winsize1=151, winsize2=91)
fl = (s1_orig/s1m)[sl2]/f12
# f12 : flattening curve
# fl : flattened spectrum
new_msk = np.abs(ratio[sl2]-f12) > thresh
ratio[sl2][new_msk] = np.nan
if ax1:
# show model corrected raw spec
ax1.plot(w1, ratio*s1m, lw=3, color="0.8")
# show fitted continuum
ax1.plot(w1[sl2], f12*s1m)
#plot(w1, s1_new)
if ax2:
# plot flattened spec
ax2.plot(w1[sl2],
np.ma.array(fl,mask=f12<0.05).filled(np.nan))
f12_0 = np.zeros_like(s1_orig)
f12_0[sl2] = f12
return s1m*f12_0, np.isfinite(ratio)
