Could maybe be nice to have a feature for computing parallel triangulations when considering very large data sets. I don't know much about this, and haven't looked much into it beyond knowing that many algorithms exist e.g. using constrained triangulations to put the points into chunks, but at least this issue can track any developments or allow anyone who sees this to offer suggestions. Streaming methods are also interesting - https://www.cs.unc.edu/~isenburg/papers/ilss-scdt-06.pdf computes a triangulation of an amazing 1 billion points.
Considering large data sets would also probably mean more thought needs to be put into how Triangulation is designed - the Graph for example probably shouldn't exist and instead graph operations should be computed on the fly (e.g. get_neighbours can be exactly replaced by get_adjacent at the cost of some extra allocations). Maybe we should do a LazyTriangulation which stores only what is needed, and all other operations are computed on demand (Lazy sounds like the triangulation isn't computed at all though, which is of course wrong - its the topology relations etc that would be computed lazily).
Could maybe be nice to have a feature for computing parallel triangulations when considering very large data sets. I don't know much about this, and haven't looked much into it beyond knowing that many algorithms exist e.g. using constrained triangulations to put the points into chunks, but at least this issue can track any developments or allow anyone who sees this to offer suggestions. Streaming methods are also interesting - https://www.cs.unc.edu/~isenburg/papers/ilss-scdt-06.pdf computes a triangulation of an amazing 1 billion points.
Considering large data sets would also probably mean more thought needs to be put into how
Triangulationis designed - theGraphfor example probably shouldn't exist and instead graph operations should be computed on the fly (e.g.get_neighbourscan be exactly replaced byget_adjacentat the cost of some extra allocations). Maybe we should do aLazyTriangulationwhich stores only what is needed, and all other operations are computed on demand (Lazysounds like the triangulation isn't computed at all though, which is of course wrong - its the topology relations etc that would be computed lazily).