tjlane / pypad

Intuitive, high-level interface to the CSPAD

Other

2 stars 1 forks source link

improve the optimization algorithm in accuracy and speed #12

Closed sellberg closed 11 years ago

sellberg commented 11 years ago

We should be able to speed up the optimization algorithm by first decouple the various degrees of freedom for a rough alignment, the to the fine optimization in the whole multi-dimensional space.

Suggestion: align gap roughly (rely on pixel center from genfilter), the align pixel center, then rotate quads, then optimize all

this scheme could most likely be improved over time..

tjlane commented 11 years ago

If we know the current algorithm works, we could get a boost by parallelizing the current optimization. @sellberg @rkirian what do you guys think?

sellberg commented 11 years ago

24 #25 #26 #27 are constructive suggestions of improvement from Lee-Ping that may solve this issue. Lee-Ping uses trust-radius (i.e. small step) Newton Raphson with regularization (i.e. shot total distance) for the optimization of his force field parameters.

@tjlane parallelizing may be a good idea if the algorithm is accurate but too slow, but i think currently we lack in accuracy...

tjlane commented 11 years ago

@sellberg agreed. I will get cracking on some of this stuff.

tjlane commented 11 years ago

closing b/c I think we split this up into many smaller more specific parts