good ideas over beer

Three good ideas came up in discussion with Foreman-Mackey and Fadely today. First, we formulated the likelihood-free inference for cosmological parameters given large-scale structure data (that I mentioned two days ago at CtU 2015) and concluded that it might be possible to perform this inference inexpensively with data and simulations in hand (that is, if we can convince some of the BOSS people to help out). Second, we discussed expanding Fadely's work on modeling the SDSS imaging catalog output to an infinite mixture of Gaussians using something like the Dirichlet process. Fadely pointed out that this might be impossible because we are doing the XD thing of convolving each Gaussian with each individual data point's individual uncertainty variance tensor. We agreed to at least look at the math, or maybe employ one of our math friends to help. Third, we talked about finding planets in K2 data by simultaneously fitting each lightcurve with a transit model plus a Gaussian-process-combined set of other-star lightcurves. That is, what we have been thinking about for Kepler search but (a) fitting the nuisances simultaneously and marginalizing out, and (b) predicting that all this will be even more valuable for K2 data than it was in the Kepler data.