Compile a test suite with which to evaluate the different jump/ramp-fitting algorithms options
stscijgbot-jp opened this issue · comments
Issue JP-3577 was created on JIRA by David Law:
https://jira.stsci.edu/browse/JP-3576 described a requested pipeline implementation of a new likelihood-based approach to jump detection and ramp fitting in the JWST calwebb_detector1 pipeline. Once ready, there should be an extensive set of tests performed to compare the performance (both in terms of accuracy and speed) of this new approach to the existing and default approach currently in the JWST pipeline.
This ticket is to start collecting input for useful ordinary and edge cases to include in this testing suite. Key areas to include will be those with very short ramps, multiple integrations, TSO data, etc. Simulated data in which 'truth' is known would also be useful for this.
Action to David Law Michael Regan Timothy Brandt Tyler Pauly Howard Bushouse Kenneth MacDonald Jane Morrison Karl Gordon Nadia Dencheva Eddie Schlafly to add potential test cases as comments here, to be consolidated into a TBD testing effort. When adding a test case, please comment as well on the conceptual area of parameter space to be explored by the test.
Comment by Karl Gordon on JIRA:
Cases for simulated data testing. Ramps without and with Cosmic rays between specific frames (including inside multiple frame groups).
- Ramps with single frames per group. Groups from 4 to 100s.
- Ramps with multiple frames per group. Appropriate frames per group from 2 to 8 (or more?)
- Ramps with 4 groups with multiple frames per group. Appropriate frames per group from 1 to 8 (or more?)
- Ramps with 3 groups with multiple frames per group. Appropriate frames per group from 1 to 8 (or more?)
- Ramps with 2 groups with multiple frames per group. Appropriate frames per group from 1 to 8 (or more?)
- Ramps with 2-4 groups with multiple integrations. Appropriate frames per group from 1 to 8 (or more?) Appropriate number of integrations from 2 to 100s.
- Ramps with disjoint segments. E.g. multiple ramp jumps that cause multiple subsequent groups to be flagged as do_not_use.
Comment by Nadia Dencheva on JIRA:
Suggestion from an external user who found the paper on arXiv:
Section 6.2 discusses a NIRCam long exposure using DEEP8 readout, but this is for an image with NINT=7. Usually people don't use that for deep imaging in order to have better spatial sampling and dodging bad pixels. It would have been nice to see this for a "typical"
{}NIRCam DEEP8 exposure (5 or 6 groups of DEEP8, NINT=1){}.