TEST: fix medium test comparison #1688
Conversation
|
Thanks for fixing another float number issue! There are a few more exact comparisons in this test file. Change them all in this PR? |
There was a problem hiding this comment.
Thanks @dmarek-flex for fixing this!
I guess another question would be why it started complaining about this now. Like, what changed?
Well part of the test relies on a random number generation, which maybe is why it happens from time to time. Although there seems to be a constant seed used. Over the day I have been getting both passes and fails. I don't know why that would not show up on the github tests though. And then on one of the versions of the test some interpolation is used, which could result in some introduced error I guess. |
dmarek-flex
force-pushed
the
dmarek/fix_medium_test_comparison
branch
from
c9c1a54
to
6a43c71
Compare
|
I ended up using a relative tolerance which is equal to single precision epsilon. Other tests would fail if I tried to be very strict with double precision epsilon. In lieu of choosing a relative tolerance for each comparison, I just chose to relax the tolerance. |
|
@marc-flex It would be helpful if you could checkout this branch yourself and see if any test fails. |
Of course! :) I just checked and all test passed |
Resolved some issues with test that used exact comparisons. The default tolerances in np.isclose are suitable.