SG-G2, SG-3, SG-4¶
Check star-to-galaxy ratios are known within 1% in the overall sample against truth values (VVDS,SG-2), against a detailed simulation (DES-DC, SG-3), dividing the sample in redshift bins (SG-4).
1 sq.deg. of full-depth, coadded data, at least in i-band, with spectroscopically identified galaxies and stars (O(40000)). In addition, or alternatively, a detailed simulation of the area (need Trilegal-like sim).
Divide area into subareas of 100 sq.arcmin. Compute in each sub-area |R1-R2|, where R1 is N_true_stars/N_true_galaxies from the external catalog (see below) and R2 is N_classified_stars/N_classified_galaxies.
SG-G2: external catalog is another real dataset (VVDS)
SG-G3: external catalog is detailed simulation (DES-DCXX truth table), useful if we need to extrapolate to data with no truth
SG-G4: external catalog is another real dataset, divided in redshift bins
Note that actual matching is not needed. Needs tuning of cut parameter for classifier.
The average of the distribution <|R1-R2|> < 0.01.
Tools: python script, not developed yet.
Perform test with matching to understand whether the problem is in stars or galaxies; use other datasets; bin in magnitudes to check where the matching breaks down