Bad weight masking in 1D extractions

[gbrammer]

The weight masking functions in make_optimal_extraction occasionally corrupt the 1D extractions at very high S/N. The earlier default behavior max_wht_percentile=98 masked pixels where the weight was above the 98th percentile, in order to flag occasional pixels that had extremely high weights for some unknown reason. This would occasionally clip the centers of bright targets.

This change implemented a new masking scheme with the parameter max_med_wht_factor=10, which would mask pixels that have weights larger than 10 times the median (nonzero) weight in the 2D array. That seems to mostly fix the issue.

Here's an example comparing the weighting schemes:

import matplotlib.pyplot as plt
from msaexp import spectrum

# A bright UNCOVER object
file = 'https://s3.amazonaws.com/msaexp-nirspec/extractions/uncover-4m1-v1/uncover-4m1-v1_prism-clear_2561_42213.spec.fits'

spec = spectrum.SpectrumSampler(file)

new = spec.redo_1d_extraction(bkg_offset=6,
                              max_wht_percentile=None, # Turn off this mask
                              max_med_wht_factor=20)

_ = spec.drizzled_hdu_figure(unit='flam')

_ = new.drizzled_hdu_figure(unit='flam')

fig, ax = plt.subplots(1,1,figsize=(9,4))

ax.plot(spec['wave'], spec['flux']*spec['to_flam'], label='Original, max_wht_percentile=98', color='r', alpha=0.5)
ax.plot(new['wave'], new['flux']*spec['to_flam'], label='max_med_wht_factor=10', color='k', alpha=0.5)

ax.legend()
ax.set_xlim(0.6, 5.2)
ax.grid()
ax.semilogx()

[gbrammer]

These updates implemented in version 0.6.15.