JP-1887: Refactoring ols_ramp_fit into smaller more manageable functions
#5819
Conversation
Codecov Report
@@ Coverage Diff @@
## master #5819 +/- ##
==========================================
+ Coverage 75.66% 76.74% +1.08%
==========================================
Files 406 406
Lines 36323 37965 +1642
==========================================
+ Hits 27484 29137 +1653
+ Misses 8839 8828 -11
*This pull request uses carry forward flags. Click here to find out more.
Continue to review full report at Codecov.
|
| nreads: int | ||
| Number of reads | ||
|
|
||
| ngroups: int |
There was a problem hiding this comment.
The variable groups is gotten from input_model.meta.exposure.ngroups in the RampModel passed to ramp_fit. It is not part of the shape of data.
There was a problem hiding this comment.
Be very careful with the use and meaning of nreads vs ngroups. In the instruments, individual detector reads can be averaged on-board and then downlinked as a single group. When the data hit the ground they are 4-dimensional: [ncols, nrows, ngroups, nints] in conventional (e.g. FITS) notation, which is [nints, ngroups, nrows, ncols] in python array notation. The data in the ngroups dimension can consist of 1 or more detector reads. So in this particular use in the ramp_fit step I'm guessing nreads refers to the actual number of detector readouts that took place in the instrument, which is not necessarily the same as ngroups (because, as mentioned above, 1 or more reads can get averaged into a group). So nreads can't reliably be determined from the shape of the 4-D data array. Only ngroups can.
There was a problem hiding this comment.
Upon clarification, the use of nreads and ngroups are the same values taken from two different locations in the RampModel class. in effect, these two variables are redundant that makes the code confusing. I will remove this redundancy and use the correct ngroups variable name.
jdavies-st
changed the title
ols_ramp_fit into smaller more manageable functionsols_ramp_fit into smaller more manageable functions
|
Ran the regtests on this PR and all pass. I think this is good to merge. |
Addresses #5701 / JP-1887
This is one in a series of PR's designed to refactor ramp fitting to make it more readable, maintainable, and reusable.
This PR is used to break down the function
ols_ramp_fit. It is over 700 lines long, but has four distinct steps, which have been encapsulated in four different functions thatols_ramp_fitcalls.ramp_fit_miri_multigroups(), deals with the special case of MIRI data with multiples groups in an integration, resizing the data for discarded groups.ramp_fit_first_pass(), does the "first pass" over the data to compute the initial values related to slope computation.ramp_fit_second_pass(), does the "second pass" over the data using the intermediate results from the first pass to compute values related to variances of estimates.ramp_fit_overall_slopes_and_variances(), uses the values computed in the first and second passes over the date to compute the overall slopes and variances for each pixel.