keno / celeste.jl

Scalable inference for a generative model of astronomical images

Celeste.jl

Celeste.jl finds and characterizes stars and galaxies in astronomical images. It implements variational inference for the generative model described in

Jeffrey Regier, Andrew Miller, Jon McAuliffe, Ryan Adams, Matt Hoffman, Dustin Lang, David Schlegel, and Prabhat. “Celeste: Variational inference for a generative model of astronomical images”. In: Proceedings of the 32nd International Conference on Machine Learning (ICML). 2015.

Usage

The main entry point is bin/celeste.jl. Run celeste.jl --help for detailed usage information.

Note that in the score mode, the script requires data downloaded from the CasJobs Stripe82 database in a given RA, Dec range. Here's an example query in the RA, Dec range [0, 1], [0, 1]:

select
  objid, rerun, run, camcol, field, flags,
  ra, dec, probpsf,
  psfmag_u, psfmag_g, psfmag_r, psfmag_i, psfmag_z,
  devmag_u, devmag_g, devmag_r, devmag_i, devmag_z,
  expmag_u, expmag_g, expmag_r, expmag_i, expmag_z,
  fracdev_r,
  devab_r, expab_r,
  devphi_r, expphi_r,
  devrad_r, exprad_r
into mydb.s82_0_1_0_1
from stripe82.photoobj
where
  run in (106, 206) and
  ra between 0. and 1. and
  dec between 0. and 1.

Then download the mydb.s82_0_1_0_1 table as a FITS file.

When scoring, one must use RUN, CAMCOL, FIELD combinations that are entirely within the RA, Dec range selected above. To find such fields, run the following query:

select distinct run, camcol, field
from dr8.frame
where
  rerun = 301 and
  ramin > 0 and ramax < 1 and
  decmin > 0 and decmax < 1
order by run

License

Celeste.jl is free software, licensed under version 2.0 of the Apache License.