Allow bounds to be specified by convex hull of points for Leaner2D

[serhiy-yevtushenko]

In some cases, the input area, where function to be interpolated is defined is not a rectangular one, and function is not defined outside of the definition area. It would be good, if one would be able to pass a convex hull of points instead (or just boundary points of the convex hull), which would allow more flexible definition of area, where points should be sampled.

[basnijholt]

AFK now, but LearnerND supports this.

[serhiy-yevtushenko]

Hi Bas, thank you very much for the tip. I believe it will be useful to add a tip for the documentation then, that if one is interested in learning 2D interpolation on function with complex shape of input domain, then one should look at LearnerND.

[serhiy-yevtushenko]

See as well issue #468 for issue on to_dataframe()

[serhiy-yevtushenko]

One more point to mention (I think it worth documenting) - when using LearnerND with convex hull, it will first to try to sample on convex hull boundary points. If the shape of convex hull is complex (i.e., it contains many thousands points), it may be not desirable from user perspective, and then it makes sense for user to sample some points on boundary before applying adaptive (in order to get good enough approximation inside domain)