Show HN: My implementation of Google's bug prediction algorithm
Hi, everyone!
I'm a high-school student. I saw that blog post about bug prediction on Google Engineering Tools blog (<http://google-engtools.blogspot.com/2011/12/bug-prediction-at-google.html>), and today I decided to implement it in Python, just for fun.
It's buggy on some repos due to issues with GitPython.
<http://pypi.python.org/pypi/bugspots>
Anyway, what do you think?
I know, weird title: character limiting.
4 comments
[ 3.2 ms ] story [ 20.8 ms ] threadfirst it's not bad! much better code than when i was your age.
second, you have a huge assumption, i think, at line 73 where your code has "repo_age = int(time.time()) - first_commit_time". this assumes that the repository has seen edits recently, but think about a repository that has been dormant for a few months. the score of any bugfix will go down but there's no reason for it to: did bugs get fixed - and less "hot" - when the repo was dormant? no. what you need to do is to find the total timespan of the repository, looking at the min and max times of the commits, then base your scores on that.
keep up the good work!
About your concern: I too think it's strange but they put it clearly in their blog post:
"The timestamp used in the equation is normalized from 0 to 1, where 0 is the earliest point in the code base, and 1 is now (where now is when the algorithm was run)."
I could do `repo_age = commits[0]["time"] - commits[-1]["time"]` but I'd like to stick to what the blog post says.
For Google, this makes sense, although I see now it's a weird idiosyncrasy outside of the Googleplex :)
Google has changes all year round, so "now" always represents a day of some changes. The algorithm is run on the entire code base, so now is a perfectly normal spot to go with. It didn't occur to me that really what now represents is "last time of changes". This is a good spot, and I'll put it in the paper I'm writing! Thanks!
A reasonable/desirable modification is probably for now to be changed to represent the last change, just like the first change is when the normalization window begins. (0 = time of first change, 1 = time of last change).
HTH.