search

MrUrq / LatinHypercubeSampling.jl

Julia package for the creation of optimised Latin Hypercube Sampling Plans
Other
36 stars 8 forks source link

Periodic objective breaks subLHCoptim #11

Closed eelregit closed 4 years ago

eelregit commented 4 years ago

by the following line https://github.com/MrUrq/LatinHypercubeSampling.jl/blob/85e6f5ae54265aac2de2bc607dacfd0c86f719e8/src/AudzeEglaisObjective.jl#L12 Instead n of the superset should be used here.

eelregit commented 4 years ago

BTW is the optimization parallelized? With the periodic function, the optimization seems to plateau much later. So I will have to wait for weeks if the optimization is done serially. Thanks.

MrUrq commented 4 years ago

Thanks for opening the issue. I have been short on time lately but I've now implemented the fix in master. Please have a look and I will bump the version if you think it is fine.

Currently there is no parallelization in place. This would of course be very desirable. I did some experiments with it several Julia versions ago but the overhead was too large and did not give any speedup for the moderately sized plans I tried. With current Julia versions, and the improvements that have been made to the threaded side of Julia, I think this would be possible without too much effort.

I have tried to pick off the major low hanging fruits performance-wise in this package but there is still some room for single-threaded performance as well.

If you want to take a stab at it, I would appreciate pull requests of either single-threaded performance improvements or multi-threading!

eelregit commented 4 years ago

Thanks! The fix looks great. I can look into the parallelization when I know julia better ;)