We have a few tests (naive bayes certainly) that use sklearn datasets that are auto-downloaded. We've seen these downloads time out recently and fail CI, so they present a chance for a flaky test. We should either eliminate use of these datasets or find a place to check them in / cache them reliably.
We have a few tests (naive bayes certainly) that use sklearn datasets that are auto-downloaded. We've seen these downloads time out recently and fail CI, so they present a chance for a flaky test. We should either eliminate use of these datasets or find a place to check them in / cache them reliably.