The original file Alex gave me is at:
pub:/home/kappes/icecube/data/IC22/Test_82_hitmap_vfit_para0/nugen_numu.000082.000000.proc.para0_S22.sel.i3.gz
The llh-map module is still being changed frequently, but it works with gulliver suite V01-00-00 and is available at:
http://code.icecube.wisc.edu/svn/sandbox/jdumm/llh-map/branches/V01-00-00a/
The script I used is available also: llh_map_pbf22.py
Once IceTray is done, the output files have to be processed with something like this script to be made into pngs (this system is changing...): visualize_llh_map.C
This script relies on the spiral.C macro.
The event-centered plots are centered around his Pandel reconstruction result. These plots then really show the llh space where paraboloid failed.
Recall that this is the overall layout of the 6-canvas plots.
| detector coords, whole sky, full llh scale | detector coords (aitoff proj) |
| reco-centered coords, whole sky, color llh scale is limited from min_llh to min_llh+6 | reco-centered coords, zoomed-in, color llh scale is limited from min_llh to min_llh+6 |
| reco-centered coords, zoomed-in and "lego2" | Event Statistics |
From these 12 events, it looks like most of paraboloid's pbfStatus=22 failures are because the likelihood space really doesn't look parabolic. Interestingly, in many cases the minimizer ends up on a slight slope, although usually not a very steep slope. In other cases, the minimum is just very broad. Although I don't see any true saddle points in the llh space, it seems plausible that some of these pbfErr values may just as well be negative as positive. It seems as though paraboloid is failing for events which we don't especially care to keep.
|
|
|
|
|
|
|
|
|
|
|
|