Question: plot_trisurf (matplotlib) directly from qhull about adaptive

[orbisvicis]

I am able to plot the adaptive triangulated mesh of other functions. I'm not sure why the triangulation of this particular adaptive surface is so incorrect while the resulting interpolated grid (interpolated_on_grid) presents a smooth surface. You're looking at the sum of several shifted step approximation functions.

[basnijholt]

Hi @orbisvicis, thanks for your interest in Adaptive.

I am having trouble understanding the exact problem. Would it be possible to post a full running example that reproduces the problem?

[orbisvicis]

The problem was that the domains differ in scale by about 1e4 while interpolated_on_grid creates a scaled interpolator but unscales the result. So I extract my data from adaptive using _data_in_bounds() - perhaps this should be a public method - and scale, triangulate, then unscale the triangulation before plotting. The result is very good; the triangulation before scaling consisted of nearly-parallel lines.

Question. Since the learner builds an incremental triangulation and the default loss function depends on the triangle area, would I get better results if I scale my bounds before running the learner? For example, scaling [0,10] to [0,1] but reverting the scale within the function to learn. Or is this done automatically?