pitrou commented on pull request #8879: URL: https://github.com/apache/arrow/pull/8879#issuecomment-824048390
This exposes a small precision issue in the variance kernel tests in MinGW32 on 32-bit floats (because of rounding issues perhaps?): https://github.com/pitrou/arrow/runs/2400347814?check_suite_focus=true#step:10:831 According to Numpy, the two float64 values differ by 5 ULPs: ```python >>> a 1011966660.249961 >>> b 1011966660.2499604 >>> np.testing.assert_array_max_ulp(a, b) Traceback (most recent call last): File "<ipython-input-15-48fada135e2b>", line 1, in <module> np.testing.assert_array_max_ulp(a, b) File "/home/antoine/miniconda3/envs/pyarrow/lib/python3.7/site-packages/numpy/testing/_private/utils.py", line 1643, in assert_array_max_ulp (maxulp, np.max(ret))) AssertionError: Arrays are not almost equal up to 1 ULP (max difference is 5 ULP) ``` -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
