| | 8 | * worked on psphot linear fit weighting. inspired by Doug Finkbeiner in January, I implemented a model-based weighting scheme. This week, I set up a series of tests to examine the faint and bright end behavior of the 3 weighting schemes (constant, image, model) in cases where the real PSF and the fit PSF are both similar and substantially different. In the end, I concluded that (a) model and constant weights both eliminate the faint end bias; (b) when PSFs differ greatly (eg real is Gauss, fit is PS1_V1), there is a strong bright end bias for image, a modest bright end bias for model, and no bright end bias for constant. I was unable to confirm the expected higher noise of the constant weight using simulated data. I concluded that we would be better off using the constant weight and possibly suffer a small increase in noise at the bright end, rather than have the faint-end bias. |
