[MRG+1] reduce decimal restriction in test_singular_values to avoid non-deterministic test failure #9162
Conversation
|
I am okay with this kind of fix. I think 6dp is too big a tolerance. Without looking at the failure I would have thought 9 or so. The difference occurs at the 11th dp and the test would, for instance, easily pass This is all despite the fact that I'm still uncomfortable that there is some unknown source of indeterminacy. I suppose it's possible that AppVeyor uses subtly different machines?? |
|
@jnothman Thanks. I follow your instruction and that seems enough to avoid this problem. I have modified the main content of the pull request to mark that it only avoid the problem temporarily, or maybe we need to remain the issue open. |
|
LGTM |
|
This LGTM. I'm merging this one. Are we okay with just modifying this specific test. There seem to be quite a few other tests comparing at 12 DP. |
|
@raghavrv Thanks. To solve current issue, it seems enough since the wrong result in three known cases(please refer to the issue) are exactly the same. It only occurs occasionaly in a particular statement and similar issues are not reported(according to my search). But it is definitely better to know the exact reason behind it. So if it is necessary, maybe we can keep the issue open and change the milestone. |
…on-deterministic test failure (scikit-learn#9162) * reduce decimal restriction * change restriction
…on-deterministic test failure (scikit-learn#9162) * reduce decimal restriction * change restriction
…on-deterministic test failure (scikit-learn#9162) * reduce decimal restriction * change restriction
…on-deterministic test failure (scikit-learn#9162) * reduce decimal restriction * change restriction
…on-deterministic test failure (scikit-learn#9162) * reduce decimal restriction * change restriction
…on-deterministic test failure (scikit-learn#9162) * reduce decimal restriction * change restriction
…on-deterministic test failure (scikit-learn#9162) * reduce decimal restriction * change restriction
…on-deterministic test failure (scikit-learn#9162) * reduce decimal restriction * change restriction
…on-deterministic test failure (scikit-learn#9162) * reduce decimal restriction * change restriction
Reference Issue
Fixes #7893 (temporarily avoid it)
What does this implement/fix? Explain your changes.
Please refer to the issue.
test_singular_valuessometimes fail and it seems unrelated to scikit-learn.(CI become green after an empty commit)
For example in pull request #9108 :
https://ci.appveyor.com/project/sklearn-ci/scikit-learn/build/1.0.14629/job/t8udw2qky46u5ewm (fail)
https://ci.appveyor.com/project/sklearn-ci/scikit-learn/build/1.0.14695/job/8kovdaea238us3u2 (pass)
Before we find the reason, it may be better to reduce the
decimalrestriction inassert_array_almost_equalto avoid unecessary confusion.Any other comments?