You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Nov 17, 2023. It is now read-only.
This was observed during the development of #18424. I've developed a fix which I've added via commit 29286c3. I observed that the failures occurred when the sample was near one or both ends of the unit normal distribution for the samples data input. Repro commands provided below.
@DickJC123
It looks like your commits have solved this issue. Thanks for that.
This reason behind, is the heavy-tailed property of Cauchy, when the input for the quantile/icdf function is close to 1 or 0, the output will blow up, causing flaky tests.
Description
This was observed during the development of #18424. I've developed a fix which I've added via commit 29286c3. I observed that the failures occurred when the sample was near one or both ends of the unit normal distribution for the samples data input. Repro commands provided below.
Occurrences
MXNET_TEST_SEED=783333771 pytest --verbose -s tests/python/unittest/test_gluon_probability_v2.py::test_gluon_cauchy
MXNET_TEST_SEED=1447693795 pytest --verbose -s tests/python/unittest/test_gluon_probability_v2.py::test_gluon_cauchy
MXNET_TEST_SEED=952434416 pytest --verbose -s tests/python/unittest/test_gluon_probability_v1.py::test_gluon_cauchy_v1
The text was updated successfully, but these errors were encountered: