Balancing learner does not work with Integrator learner about adaptive

[basnijholt]

originally posted by Joseph Weston (@jbweston) at 2018-07-13T11:16:03.002Z on GitLab

More details please

[basnijholt]

originally posted by Jorn Hoofwijk (@jorn) at 2018-07-13T11:30:09.676Z on GitLab

I told him to report this, however, here is an actual example:

import numpy as np
import adaptive
adaptive.notebook_extension()

def f(x):
    return x

learner = adaptive.IntegratorLearner(f, (0,1), tol=1e-8)
runner = adaptive.Runner(learner, lambda l: l.done()) # runs perfect
runner.live_info()

learner1 = adaptive.IntegratorLearner(f, (0,1), tol=1e-8)
learner2 = adaptive.IntegratorLearner(f, (0,1), tol=1e-8)

balancer = adaptive.BalancingLearner([learner1, learner2])

runner = adaptive.Runner(balancer, lambda l: l.loss() < 0.1) # fails
runner = adaptive.Runner(balancer, lambda l: l.done())       # fails
runner = adaptive.Runner(balancer, lambda bl: all(l.done() for l in bl.learners)) # fails
runner.live_info()

especially the last one, i don't know why it fails yet

[basnijholt]

originally posted by Jorn Hoofwijk (@jorn) at 2018-07-13T11:33:02.163Z on GitLab

got it:

Traceback (most recent call last):
  File "test5.py", line 14, in <module>
    runner = adaptive.BlockingRunner(balancer, lambda l: l.loss() < 0.1) # fails
  File "/home/jorn/natk/bep/adaptive/adaptive/runner.py", line 152, in __init__
    self._run()
  File "/home/jorn/natk/bep/adaptive/adaptive/runner.py", line 175, in _run
    points, _ = self.learner.ask(n_new_tasks)
  File "/home/jorn/natk/bep/adaptive/adaptive/learner/balancing_learner.py", line 89, in ask
    return self._ask_and_tell(n)
  File "/home/jorn/natk/bep/adaptive/adaptive/learner/balancing_learner.py", line 73, in _ask_and_tell
    n=1, add_data=False)
TypeError: ask() got an unexpected keyword argument 'add_data'

[basnijholt]

originally posted by Bas Nijholt (@basnijholt) at 2018-07-18T18:20:14.013Z on GitLab

Thanks for reporting, I've added some code to raise a NotImplementedError when add_data=False in the IntegratorLearner here: https://gitlab.kwant-project.org/qt/adaptive/commit/8640b8d58006b634e278709fd185a6ed2d05bfcb.

This doesn't fix the issue but at least makes it clear as to why it happens.

[basnijholt]

originally posted by Jorn Hoofwijk (@jorn) at 2018-09-25T11:47:55.034Z on GitLab

we could do it similarly as with the LearnerND: add a with restore(self) to allow not adding the data.

Also this would be solved if we would switch to a new BalancingLearner: !113