Goals

• get used to some of this ugly buzzword tools as they are used in a broad audience
• read out bugzilla bug description and try to find out, if the initial description (comment) has any deeper information of the bug

Bugzilla

The script 'py-bug.py' reads out the public bugs of bugzilla.opensuse.org, one by one, and writes the * first comment (bug description) * Summary * number of comments * creation time * time of last comment to a json file. Unforntunately I could not get the area/type of the bug, so something like 'kernel', 'yast'...

Tensorflow

The script 'json-reader.py' reads in the json file of the bugs and tries to learn if the inital bug description could be linked to the 'duration' (time of las comment - creation time) or the number of comments. For this the neuronal net could be modifed by commandline parameters

Lessons learned

• accelerated docker containers are not easy to install, had to use the pip package instead
• my GPU (1050Ti) is not so much faster than my CPU (Xeon E3-1231v3)
• could not train the modell to get any useful information, so no automatic bug resolution

Github Repo

https://github.com/mslacken/ml-bugs