Machine learning is used in gerrit build

[Baole Fang]

Hi,

My name is Baole Fang and I'm responsible for training a machine learning
model to select unit tests to run in GSoC.

Currently, the model <https://github.com/baolef/libreoffice-ci> is trained
and is integrated into gerrit_master_ml
<https://ci.libreoffice.org/job/gerrit_master_ml/>. It has replaced the
original gerrit_master <https://ci.libreoffice.org/job/gerrit_master/> to
be run when a patch is pushed to gerrit. The gerrit_master_ml
<https://ci.libreoffice.org/job/gerrit_master_ml/> job first runs the
machine learning model to predict whether the patch is likely to pass or
fail any unit tests. If it is likely to fail, then a fast track
gerrit_master_seq <https://ci.libreoffice.org/job/gerrit_master_seq/> will
be run, where gerrit_linux_clang_dbgutil
<https://ci.libreoffice.org/job/gerrit_linux_clang_dbgutil/> is first run,
followed by other builds if it passes. If the patch is likely to pass, then
the original gerrit_master <https://ci.libreoffice.org/job/gerrit_master/> is
run.

The new job saves computation when the model predicts correctly that a
patch will fail any unit tests. When it runs the fast track
gerrit_master_seq <https://ci.libreoffice.org/job/gerrit_master_seq/> and
the patch will fail some unit tests, only one build will be run instead of
all of them.

Feel free to contact me if you have any questions!

Best,
Baole Fang
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/libreoffice/attachments/20230809/31f246c8/attachment.htm>