non-parametric Tractor

Lang and I pair-coded (him in Princeton, me in an office in Heidelberg with Foreman-Mackey, Marshall, and Price-Whelan forced to listen in) a version of the Tractor to model Herschel maps of warm dust in M31. Our model is a thermal emission model on a grid of point sources, smoothed with the (very variable with wavelength) point-spread function. We started by trying to do this by just making a huge grid of Tractor point sources with fixed positions and leaving everything up to the Tractor's excellent optimizer, but no way. So in the evening, Lang made a new Tractor model which is a non-parametric (think: lots of parameters) thermally emitting dust blanket. It worked! Now we just have to figure out how to make it fast enough that we can run it on large amounts of sky in practice. The idea of all of this is to get a Herschel-inferred dust map in M31 at the resolution of Herschel's best data, not smoothed to the resolution of Herschel's worst data.

At the other desks in the Camp Hogg office, Price-Whelan worked on object detection, photometry, and astrometry in multi-epoch data with variable conditions, Marshall worked on modeling lenses in PanSTARRS, and Foreman-Mackey worked on making The Thresher run on full-frame images, with and without drifting centroids, and on PanSTARRS data (to give something to our hosts here at MPIA).