Opening a PR on behalf of @dmargala for final review. This PR adds test comparisons to specter.extract.ex2d.ex2d_patch, and adds options for whether the R-matrix calculation should be only internal to individual fibers, or include decorrelation across fibers. Thanks for the updates that make these within np.allclose level of agreement.
Briefly thinking about it more, my description that the decorrelation is for noise vs. signal isn't quite right either. It's more about whether the calculation should decorrelate across neighboring fibers or not. The default in this branch is good (only within a single fiber), and I like the structure of how the option is used, but I'd like to think a bit more about what the option names should be. I'm booked for the rest of today, but will try to comment more tonight.
To make progress in the meantime, I'm ok with either:
branch from this branch and work on full ex2d comparisons with specter (e.g. by running spex vs. desi_extract_spectra with same input data and wavelength options); then rebase that against master after this is merged but before it is pushed to github
go ahead and merge this branch and then start working on the full ex2d comparison, with the expectation that the decorrelate option names may change.
Opening a PR on behalf of @dmargala for final review. This PR adds test comparisons to
specter.extract.ex2d.ex2d_patch
, and adds options for whether the R-matrix calculation should be only internal to individual fibers, or include decorrelation across fibers. Thanks for the updates that make these withinnp.allclose
level of agreement.Briefly thinking about it more, my description that the decorrelation is for noise vs. signal isn't quite right either. It's more about whether the calculation should decorrelate across neighboring fibers or not. The default in this branch is good (only within a single fiber), and I like the structure of how the option is used, but I'd like to think a bit more about what the option names should be. I'm booked for the rest of today, but will try to comment more tonight.
To make progress in the meantime, I'm ok with either:
decorrelate
option names may change.