Details
-
Type:
Bug
-
Status: Done
-
Resolution: Done
-
Fix Version/s: None
-
Component/s: afw
-
Labels:
-
Story Points:3
Description
Sometimes the afw test testGausisianProcess.py can fail:
...
|
======================================================================
|
FAIL: testTooManyNeighbors (__main__.GaussianProcessTestCase)
|
----------------------------------------------------------------------
|
Traceback (most recent call last):
|
File "./tests/testGaussianProcess.py", line 62, in testTooManyNeighbors
|
gg.selfInterpolate(sigma, -1, nData-1)
|
AssertionError: Exception not raised
|
----------------------------------------------------------------------
|
Ran 11 tests in 1.992s
|
FAILED (failures=1)
|
This is a very rare failure. It ran 102 times on my laptop before it stopped with this failure. So far we have only seen reports of this failure on Python 3.
Attachments
Issue Links
- relates to
-
DM-7207 Modify afw tests to support pytest
- Done
Activity
I just saw this failure on my machine with Python 2, but my next rebuild passed. It wasn't with the master afw but I don't think I did changes related to this test (however I don't know afw well). Searching on Jira led me to this ticket so I think I'd report it.
Hsin-Fang Chiang
added a comment - I just saw this failure on my machine with Python 2, but my next rebuild passed. It wasn't with the master afw but I don't think I did changes related to this test (however I don't know afw well). Searching on Jira led me to this ticket so I think I'd report it. Thank you for the first Python 2 report. The code is very broken and needs an overhaul. I think we are just getting lucky with python 2 memory layouts.
This pull request adds tests to all of the user-facing methods in GaussianProcess.cc. These tests ensure (or ought to) that exceptions are thrown if users pass in invalid inputs.
I also took the opportunity to flesh-out testGaussianProcess.py so that the functionality of GaussianProcess.cc is tested more methodically.
I added some methods to the GaussianProcess class to allow users to inspect its contents (specifically methods to return the data points and function values underlying the GaussianProcess).
reviewed by Josh Meyers
Good to know. Thanks.
I think that also explains the randomness of the failures. Some small fraction of the time, the memory right before the interpolation points happens to say 0.0, and then neither the intended nor the bonus exception (the "Nearest neighbor search in selfInterpolate did not find self\n") get triggered, which triggers the assertRaises unit test.