Returning np.nan breaks the 1D learner

[tlaeven]

From commit: 4381a9e onward,
returning (a np.array of) np.nan cancels the runner.
Example code, to compare the working commit, f38d20f,
with the not working one, 4381a9e.

import adaptive
adaptive.notebook_extension()
import numpy as np
def f(x):
 if 0 < x < 1:
 return x
 else:
 return np.nan

learner = adaptive.Learner1D(f, 0.1*np.array([-1, 1]))
runner = adaptive.Runner(learner)
runner.live_info()

[basnijholt]

Thanks for reporting this!

This is definitely a bug, previously we have decided that when a NaN is returned that it is ignored and that the learner should use the interpolated value, it should definitely not crash the runner!