Closed sellberg closed 11 years ago
@tjlane disabling quad_rotation removes all the rotations of quads, where as I meant to remove only one degree of freedom where all quads are rotated the same way, but keep the degrees of freedom where quad rotations cannot be described by a rotation of the full detector... does that make sense? unless you prove me wrong, I'd reopen this issue again. ;-P
after discussing, @sellberg and I agree that we just need to decide what the default should be
So it turns out that adding in the rotations does let you get to a slightly better answer, as measured by the objective function, at the cost of some time
With rotations:
Objective function: -16755986001.261101
Time to reach solution: 2.1e+03s
Without rotations:
Objective function: -16742384828.638697
Time to reach solution: 1.6e+03s
So rotations takes ~10 min longer.
I need to re-run to see if the difference is visible -- unfortunately I over-wrote the outputs on top of one another lol.
Leaning towards including rotation by default.
Here's some more stats from a scoring of the above:
With rotations
Total Residuals: 0.006233 inv. Angstroms
Residuals for each peak:
[ 0.00143242 0.00255038 0.00190585 0.00034463]
Without rotations:
Total Residuals: 0.012106 inv. Angstroms
Residuals for each peak:
[ 0.00114429 0.00339342 0.00533544 0.0022332 ]
Hard to tell the difference by eye, but I think we should keep the rotations. @sellberg, thoughts?
@tjlane I agree with you, 10 min is OK for a factor of 2 in total residuals. closing this again..
This is easy to set in the filter_params.yaml
vs.
Changed the default to be only offsets.