The test hp.inference.tests.test_prior.TestBoundedGaussian.test_sample failed at least once on Travis in the context of #358.
Here is @briandleahy's plan of attack pasted from a comment in that thread:
IIRC there was no seed test in the test, and it was only testing for 10 samples.
To me this seems like an edge case that happens once in a while (samples are generated outside the bounds).
I would say "someone" should do the following:
Find exact failing test. [done]
Raise an issue that this test is finicky. [done]
Set a seed and increase the number of samples to a few thousand to see if the test fails and therefore what the bug is.
Fix the bug.
The test
hp.inference.tests.test_prior.TestBoundedGaussian.test_sample
failed at least once on Travis in the context of #358.Here is @briandleahy's plan of attack pasted from a comment in that thread: