Computational Efficiency

AdvancedPhotonSource / xdesign

Tools for designing x-ray phantoms and experiments.

https://xdesign.readthedocs.io

Other

24 stars 16 forks source link

Computational Efficiency #30

Closed carterbox closed 6 years ago

carterbox commented 8 years ago

After profiling sinogram generation, I have concluded that the following methods may decrease the run time of calculating projections:

[x] Faster solver than scipy.optimize.linprog
[x] Less calls to polytope.intersect by pre sorting the polytopes using a bounding circles.
[x] Less calls to polytope volume e.g. don't calculate the intersection of a volume that is empty
[x] prevent polytope from doing redundant bounding box calculations
[x] prevent polytope from doing redundant cheby_ball and reduce calculations

carterbox commented 8 years ago

Polytope is already efficient at calculating the empty volumes. It doesn't make a difference whether volume is called less.

carterbox commented 7 years ago

Tried to reduced calls to polytope.intersect by prescreening intersections using bounding spheres. In my test of 10,000 intersections, it only avoided 3.8% of the computations, but the run time on my machine was also reduced by around 4%. (3111ca7)

carterbox commented 7 years ago

Cut beam_intersection computation time in half by adding rotations and translations to polytope and caching the half_space instead of recreating a polytope every time the beam moves. (a25e9c2)