Viewing the gallery of events where 1-iter was >30deg from truth but 32-iter was within 3deg from truth.
These events are Simulation 2.0 dataset 651 IC22 neutrino-generator events, reconstructed with gulliver (ipdfGConvolute, 1 iteration, is the Reco result shown in plots) and paraboloid.
These events were selected because they are an odd scenario where:
1) The Gulliver 1-iter result was >30deg from the truth.
2) The Gulliver 32-iter results was within 3deg of truth.
You can browse the .i3 file containing these events, or look at them in the event viewer by downloading the .i3 file (complete with GCD) here.
Remember, pushing 't' in the shovel should eliminate those pesky TrayInfo frames (vertical red bars).
The goal of these maps is to try to understand why the 32-iter result found such a good solution while the 1-iter was so far off.
If these events are recovered by running more iterations, are they good?
For each event, I give event statistics and 5 plots of the llh map but with different coordinates, projections, zoom levels, or draw options, following the layout of this table:
| detector coords, whole sky | detector coords (aitoff proj) |
| reco-centered coords, whole sky, limited color scale | reco-centered coords, zoomed-in, limited color scale |
| reco-centered coords, zoomed-in and "lego2" | Event Statistics |
Clearly there is some problem with my function for calculating the # of muons.
The white X is the 1-iter result and the red X is the MC truth. (The 32-iter result is not shown, but it is always closer to the mc truth than the 1-iter).
