Status: In Progress
Fix Version/s: None
Component/s: Alert Production
Sprint:AP S20-4 (March), AP S20-5 (April), AP F20-5 (October), AP F20-6 (November), AP S21-2 (January), AP S21-3 (February)
Following up the discussion with Robert Lupton, we want to test the decorrelation afterburner and image differencing in general with the well defined Gaussian source cases.
- Create test case(s) of template, science exposures where the matching kernel is well sampled, i.e. the PSF width (sigma sqrt quadrature) difference is at least of a few pixels.
- Create test case for the close matching case, where the matching kernel is not well sampled, i.e. the PSF width (sigma sqrt quadrature) difference is about or less than one pixel.
- Check what exists in current unit tests. The Zogy code may already have bits that can be reused or adopted.
Updated description following December sprint planning:
Create unit tests for image differencing with simulated images for the template and science image. The simulated images should include gaussian PSFs for simplicity, but have different noise levels and PSF widths. To test the decorrelation afterburner, there should be tests that swap which simulated image is the template and which the science image, and verify that they produce the same result to within a predefined tolerance.
It is expected that the tests may fail when implemented properly, due to potential unresolved issues with the decorrelation afterburner. If that is the case, the tests should use @unittest.expectedFailure and a new ticket should be filed to fix the failure.
Discussed this 2020-06-29. Gabor has generated some test cases, but we are concerned that he's not really got to the issues that Robert is concerned about.
We agreed that this is not currently the highest priority, but Gabor will come and update this ticket with his results to date and suggest next steps when he has a few moments.