HLA-1362: Modified the minimum RMS computation used as a discriminant for choos…

[mdlpstsci]

Resolves HLA-1362

Closes #

Modified the minimum RMS computation used as a discriminant for choosing the
background computation algorithm for source detection. The update is for the source code only as any detector-dependent configuration values have not yet been tuned.

Access to the updated source code is so statistical tests can be run for further evaluation of the change and data which can be used to tune the detector-dependent configuration values.

Checklist for maintainers

• added entry in CHANGELOG.rst within the relevant release section
• updated or added relevant tests
• updated relevant documentation
• added relevant label(s)
• ran regression tests, post a link to the Jenkins job below.
  https://plwishmaster.stsci.edu:8081/job/RT/job/Drizzlepac-Developers-Pull-Requests/156/

[rlwastro]

This looks good to me. Very clean implementation. You can certainly ignore my suggestion if you want!

[rlwastro]

Suggested change

	minimum_rms = np.sqrt(np.clip(bkg_median * self.keyword_dict['texpo_time'], a_min=1.0, a_max=None)) / self.keyword_dict['texpo_time']
	minimum_rms = np.sqrt(
	(bkg_median * self.keyword_dict['texpo_time']).clip(min=1.0)
	) / self.keyword_dict['texpo_time']

I like using the ndarray.clip() method rather than the np.clip function. But if you have some reason to prefer the np.clip function version, your code will also work. In either case, you could omit the max value (or a_max, which seems the same as max), since it defaults to None.

[mdlpstsci]

I have no preference for either function/method, but since the function is already coded I will continue to use it :). I will note that one does have to specify both a_min and a_max (TypeError: clip() missing 1 required positional argument: 'a_max') as I discovered when I initially implemented the code!

[rlwastro]

OK, sounds OK to me. (You can omit the max value with ndarray.clip.)

I have one other suggestion. I think the version of the rms that is used in the "excessive zero background" case should also have the minimum applied. The code is a little simpler because the median value is known to be zero. Here's a patch:

index e9ceb8fe..dbadf50f 100755
--- a/drizzlepac/haputils/catalog_utils.py
+++ b/drizzlepac/haputils/catalog_utils.py
@@ -289,6 +289,12 @@ class CatalogImage:
 
                 is_zero_background_defined = True
                 log.info(f"Input image contains excessive zero values in the background. Median: {self.bkg_median:.6f} RMS: {self.bkg_rms_median:.6f}")
+                # Compute a minimum rms value based upon information directly from the data
+                minimum_rms = 1. / self.keyword_dict['texpo_time']
+                if np.nan_to_num(self.bkg_rms_median) < minimum_rms:
+                    self.bkg_rms_median = minimum_rms
+                    log.info(f"Minimum RMS of input based upon the total exposure time: {minimum_rms:.6f}")
+                    log.info(f"Median RMS has been updated - Median: {self.bkg_median:.6f} RMS: {self.bkg_rms_median:.6f}")
 
         # BACKGROUND COMPUTATION 2 (sigma_clipped_stats)
         # If the input data is not the unusual case of SBC "excessive zero background", compute

I included code that handles the issue you ran across where the computed rms is NaN.