(feature) add anisotropic meshing to LearnerND

[basnijholt]

(original issue on GitLab)

opened by Jorn Hoofwijk (@Jorn) at 2018-07-07T16:35:34.237Z

like this:

from this source

Still need to figure out whether de stretching should depend on the derivative or the second derivative

[basnijholt]

originally posted by Joseph Weston (@jbweston) at 2018-07-07T17:08:30.027Z on GitLab

Nice idea!

I think its the derivative: imagine plotting a flat plane at an angle. To get the same triangles at different tilt angles we need to stretch differently, but in all cases the second derivative is 0

[basnijholt]

originally posted by Anton Akhmerov (@anton-akhmerov) at 2018-07-07T20:05:51.668Z on GitLab

Nice reference. I have thought about particle-based methods, but decided in favor of triangulation because I thought it is simpler when we don't have a luxury of being able to query the function at will.

With having the triangulation we should be able to do just that using the transform argument to the triangulation.

[basnijholt]

originally posted by Jorn Hoofwijk (@Jorn) at 2018-07-08T11:03:01.973Z on GitLab

I agree that particle based solutions are not very useful for adaptive. I just figured this image to be a perfect example of what we want to achieve.