user-defined interaction kernels would be useful

[bhaller]

It would be great to be able to define an interaction kernel with an interaction type of 's' (like a type 's' mutation type), where the user supplies a function that translates distance into interaction strength. That could be pre-evaluated and cached by SLiM, at some high resolution (say, 1024 steps between 0 and maxDistance, or even more), so it would be extremely fast to evaluate – probably faster than most of the built-in kernels, in fact (and maybe the built-in kernels should be recast into this implementation for more speed). This would get rid of one common usage of interaction() kernels, which are very slow.

@petrelharp Might be of interest. What resolution do you think would suffice? (I suppose that could even be user-specified, but maybe that's excessively complicated.) The only issue really is memory usage, and the memory usage for, say, 1024 doubles is trivial.

[bhaller]

Perhaps this would be a good improvement to postpone until such time as the SpatialKernel class is visible in Eidos, if that ever happens. The up-front computational cost of this kernel type would be best done just once, not once per call, which (for many uses) would mean needing to construct and keep a permanent SpatialKernel instance. Tagging long-term.

[bhaller]

This would notably be useful for the circle-intersection interaction function of @samchamper.