Learner1D fails with extremely narrow features

[basnijholt]

(original issue on GitLab)

opened by Bas Nijholt (@basnijholt) at 2018-06-08T01:15:34.930Z

Traceback for <Task finished coro=<AsyncRunner._run() done, defined at /gscratch/home/t-banijh/adaptive/adaptive/runner.py:365> exception=KeyError((-0.18000000000275307, -0.18000000000275285),)> (most recent call last):
KeyError: (-0.18000000000275307, -0.18000000000275285)
  File "/gscratch/home/t-banijh/adaptive/adaptive/runner.py", line 321, in status
    return 'failed'
  File "/gscratch/home/t-banijh/adaptive/adaptive/runner.py", line 321, in status
    return 'failed'
  File "/gscratch/home/t-banijh/adaptive/adaptive/runner.py", line 417, in _run
    self.end_time = time.time()
  File "/gscratch/home/t-banijh/adaptive/adaptive/learner/learner1D.py", line 308, in ask
    self.tell(points, itertools.repeat(None))
  File "/gscratch/home/t-banijh/adaptive/adaptive/learner/base_learner.py", line 40, in tell
    self._tell(x, y)
  File "/gscratch/home/t-banijh/adaptive/adaptive/learner/learner1D.py", line 244, in _tell
    self.update_losses(x, real)
  File "/gscratch/home/t-banijh/adaptive/adaptive/learner/learner1D.py", line 159, in update_losses
    loss = self.losses[x_lower, x_upper]

Here the key (-0.18000000000275307, -0.18000000000275285) is spaced only 2e-16 so probably it's reaching machine precision.

[basnijholt]

originally posted by Joseph Weston (@jbweston) at 2018-06-08T21:13:55.488Z on GitLab

I don't understand this problem. Why does it raise an exception in this case?

[basnijholt]

originally posted by Bas Nijholt (@basnijholt) at 2018-07-16T18:09:20.173Z on GitLab

Closed by gitlab:!73