Open markBowman opened 17 hours ago
This is time boxed to 2 points. The current background technique does ok in the blue, but in red orders fails to the point of concern. I expect to try two more methods for the smooth background component:
dilate the traces mask and use small subsections to fit a smooth b-spline Try a SVGP using gpflow https://gpflow.readthedocs.io/en/master/notebooks/advanced/gps_for_big_data.html. This may also need to use a subset of the data to be feasible.
As an NRES developer, I would like to be able to better subtract the background to remove data artifacts due to scattered light so that we can produce reliable radial velocities.