[Magnap]

I noticed that the paper mentions constructing a triangulation of a set of initial sample points, which would allow for the optimization domain to be any convex polytope. Is this something you are planning to explore/incorporate?

EDIT: I see now that this is not an implemention of the algorithm in the paper, but its own algorithm entirely. Even so, it appears to me the same technique could be used in this case.

[chroem-]

Hi there,

Yes, I hope to add that feature soon-ish. The reason I left it out of this first release is that computing Delaunay triangulations scales very poorly with dimensionality. That was why TRIOPT was restricted to low dimensional spaces. I will have to implement a more efficient way to compute nonoverlapping triangulations. In the mean time, you can simply make the optimization domain bigger to encapsulate the shape of your problem domain.