Learner2D.plot() returns NaN

[andrenmelo]

Example:

import adaptive
adaptive.notebook_extension()
import numpy as np

bounds = [(1, 1.7), (250, 550)]
learner = adaptive.Learner2D(lambda x: x[0] * x[1], bounds=bounds)
adaptive.runner.simple(learner, goal=lambda l: l.npoints >= 2000)

plot_data = learner.plot().Image.I.data
print(np.sum(np.isnan(plot_data)))
>>> 47

The issue is the precision in Learner2D._scale (which is called by Learner2D.ip, which itself is used to make the plot):

points, values = learner._data_in_bounds()
points = learner._scale(points)
print(np.min(points[:, 0]), np.max(points[:, 0]))
>>> -0.5000000000000001 0.49999999999999983

Points lying exactly at x = 0.5 will be outside of the convex hull of points and scipy.interpolate.LinearNDInterpolator fills them with nan.

[basnijholt]

Thanks for reporting! I'll fix it tomorrow.

[basnijholt]

OK this problem only really appears for plotting because

points, values = learner._data_in_bounds()
values = learner.ip()(learner._scale(points))
np.sum(np.isnan(values))

prints

0

and happens inside the plot function because x = y = np.linspace(-0.5, 0.5, n).