Daniel Russel wrote: > After the discussion of connectivityrestraint performance, I suggest we > add a benchmarks folder to IMP. This folder will have C++ programs which > time some aspect of IMP so that we can see how speed is doing (I can't > figure out how to profile code run from python, which is necessary any > time we want to get into running times). Eventually it would be nice to > have them run periodically on a reference machine and recorded, but that > can wait as it sounds annoying to set up :-)

Sounds reasonable to me. Python code can certainly be benchmarked, but it's probably easier for now to only worry about the C++ level, where there are fewer overheads anyway.

Ben