Currently, we are skipping the test test_send_warnings_to_log() in test_output.py on Windows because we cannot figure out why it fails even though capturing sklearn warnings works in general.