Add Quantization effect

[teutoburg]

Add a new effect electronic.Quantization, which does the "flooring" previously performed within ShotNoise and also sets the data type to unsigned integer (32 bit) by default. This can be changed by passing the dtype keyword argument to any dtype accepted by Numpy. A warning will be logged if the selected data type is not an integer, since usually the outcome of flooring should be some kind of integer.

Warning

As this takes the flooring out of ShotNoise, it will probably (slightly) change (increase) any previous output unless the new Quantization effect is added in the corresponding IRDB YAML file.

[codecov]

Codecov Report

Attention: 1 lines in your changes are missing coverage. Please review.

Comparison is base (0510a49) 80.28% compared to head (c400805) 80.34%.

Files	Patch %	Lines
scopesim/effects/electronic.py	92.85%	1 Missing ⚠️

Additional details and impacted files

@@              Coverage Diff               @@
##           dev_master     #308      +/-   ##
==============================================
+ Coverage       80.28%   80.34%   +0.06%     
==============================================
  Files             147      148       +1     
  Lines           14844    14898      +54     
==============================================
+ Hits            11917    11970      +53     
- Misses           2927     2928       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[hugobuddel]

Awesome! Can't wait to add these to the IRDB.

Specifying a signed dtype will 'wrap around' on overflow to negative values. Having a positive flux value be converted to negative values is probably never what the end users wants, as no detector works this way:

my_data = np.array([3.8, 8.1, 130000.2])

np.floor(my_data).astype("uint16")
# Works fine, overflow capped to maximum value:
# array([    3,     8, 64464], dtype=uint16)

np.floor(my_data).astype("int16")
# Works against users intentions, overflow wraps around to negative values:
# array([    3,     8, -1072], dtype=int16)

Note that the non-linearity effect should have already replaced values that are larger than what the detector can measure with the maximum value (that is, the saturation value). However, the non-linearity effect is not always turned on.

The simplest change would be to simply disallow using signed integer types completely.

Another option would be to explicitly replace values with the maximum and minimum, e.g.

real_dtype = np.dtype(new_dtype)
max_value = np.iinfo(real_dtype).max
min_value = np.iinfo(real_dtype).min
np.clip(obj._hdu.data, min_value, max_value, out=obj._hdu.data)

The default is uint32, so by default things will work fine, so perhaps this issue is not that important.

[teutoburg]

Thanks for considering these edge-cases! Maybe for now it would be fine to log a warning when a signed integer type is used?

[hugobuddel]

A warning would be good as well