Repository navigation
Test failure: eventseg test_fit_shapes #550
Description
@davidt0x, have you seen this failure before?
FAILED ../../../../../../../../../Users/runner/work/brainiak/brainiak/tests/eventseg/test_event.py::test_fit_shapes - AssertionError: Segmentation from learn_events not correctly normalized
assert False
+ where False = <built-in method all of numpy.ndarray object at 0x1451ff270>()
+ where <built-in method all of numpy.ndarray object at 0x1451ff270> = array([False, False, False, False, False, False, False, False, False,\n False]).all
+ where array([False, False, False, False, False, False, False, False, False,\n False]) = <function isclose at 0x10677a030>(array([nan, nan, nan, nan, nan, nan, nan, nan, nan, nan]), array([1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]))
+ where <function isclose at 0x10677a030> = np.isclose
+ and array([nan, nan, nan, nan, nan, nan, nan, nan, nan, nan]) = <function sum at 0x10676d870>(array([[nan, nan, nan, nan, nan],\n [nan, nan, nan, nan, nan],\n [nan, nan, nan, nan, nan],\n [nan, nan... nan, nan, nan],\n [nan, nan, nan, nan, nan],\n [nan, nan, nan, nan, nan],\n [nan, nan, nan, nan, nan]]), axis=1)
+ where <function sum at 0x10676d870> = np.sum
+ and array([1., 1., 1., 1., 1., 1., 1., 1., 1., 1.]) = <function ones at 0x10676bc70>(10)
+ where <function ones at 0x10676bc70> = np.ones
https://github.com/brainiak/brainiak/actions/runs/11862458790/job/33061887263
Activity
davidt0x commentedon Nov 15, 2024
No, never saw this one, weird. Let me try to reproduce.
davidt0x commentedon Nov 16, 2024
One thing I am noticing is the test uses different random sample data each time. Maybe something is flakey with it numerically that generates NaNs on some occasions. Let me see if I can reproduce by running a lot locally. Do to the increased number of runs (os's, python versions, etc) if the test is flakey because of such an issue we will be more likely to run into it now. It would probably be a good idea to seed in this case. Let me confirm locally.
mihaic commentedon Nov 16, 2024
If it looks flakey when running locally, we can certainly mark it and deal with it later.
davidt0x commentedon Nov 16, 2024
Weird, can't reproduce locally (WSL linux), but re-running the jobs seems to have fixed it as well. Maybe it is only flakey on x64 macos?