23/05/2007
DC lamp installed, animation here. Still 3-5% variation.
.
16/03/2007
Commited revision 4965, results are as follows:
Playing around in 640x480 8bit mode. Flopcal.py currently works ok in static mode (LSM fit only), but the LSM fit is proving to be tricky. Here's what I noted:
fitting a gaussian + support is not robust enough, solver gets lost. Possibly because we don't have a good initial estimate for the supports.
fitting polc background (2nd degree) separately, then LSM separately seems to work. Background converges in 4 iterations (very quickly) to Chisq~12. LSM converges in 9 iterations, to chisq~90. This produces good-looking residuals. Options used:
background_order = 2 initial_fit_fwhm = 3.0 filename_pattern = test.pgm fit_bg_lsm = 0 masking_radius = 10 ideal_fwhm = extract_sources = 1 polc_background = 2 static_mode = 1 directory_name = . scaling_factor = 1 initial_fit_flux = 5000.0
fitting LSM sources + polc background simultaneously does not converge. But that's ok, since the background fit is so fast anyway.
Note that I've currently got the system linked against /aips++/prod, so the old LSQFit is used. The new one was not so encouraging last time I tried it, but perhaps I should try it again some time.