Add a benchmark

[mwestphal]

@christianbrinch @smaret
Is there any way to test lime functionnality ?

I've been trying to compare results (image0.fits) but at each execution , the results are different.

is there a minimal example wich lime would resolve always the same way ?

[christianbrinch]

Lime will never give you the exact same result due to its stochastic nature. If you want it to give the exact same result in two consecutive runs, you should fix the random number seeds. Mind you that a random number sequence is initialized in several places.

I used to have a test suite, but I left that behind at my old job. At a very minimum, LIME should reproduce the van Zadelhoff (2002) benchmark. It is also useful to have a continuum benchmark test, mainly for the ray-tracer.

[mwestphal]

@smaret I may need your help here.

[smaret]

I agree that a test suite would be useful for future developments. I can have a look at this but probably not before the beginning of January though.

[christianbrinch]

Maybe another benchmarking workshop would be in order. After all, the original Leiden benchmark is 15 years old...

[smaret]

@christianbrinch Maybe... But let first make sure than we still pass the original benchmark.

[smaret]

For reference, the original benchmark problems can be found here.