Bug in number of groups in twopoint_difference code

[stscijgbot-jp]

Issue JP-3793 was created on JIRA by David Law:

A science collaborator reported strange performance of detector1  in versions 1.14.1 and more recent of the pipeline, which I've run down to partially related to an erroneous calculation in the number of groups within the twopoint_difference step.

Around line 164:

else:
        total_sigclip_groups = nints * (ngrps - num_flagged_grps)
 total_groups = nints * (ngrps - num_flagged_grps)
 total_diffs = nints * (ngrps - 1 - num_flagged_grps)
 total_usable_diffs = total_diffs - num_flagged_grps

I think the parentheses are misplaced here, in the case I'm looking at (jw01294008) ngroups=5, nints=15, and num_flagged_grps=30 since there are 2 flagged groups (first and last) in each of the 15 ints.  The calculated total_groups is thus -375, and likewise for the other computed terms, e.g. total_usable_diffs = -420

[stscijgbot-jp]

Comment by David Law on JIRA:

Discussion ongoing with Michael Regan about how this should be resolved, and whether the comparison of ngrps against minimum_groups should instead be the number of usable groups against minimum_groups.

Test case inspired by jw01294008001_08101_00002_mirifushort which started showing NaN slopes after the 1.14.1 change to being able to look across integrations.

[stscijgbot-jp]

Comment by David Law on JIRA:

See spacetelescope/stcal#317