2 ms·
> As you can see, the calculations are absolutely fair and transparent. ...if you can grant them the premise that the 3.3% sampled represents the project as a
by wyldfire 9y ago
> As you can see, the calculations are absolutely fair and transparent.
...if you can grant them the premise that the 3.3% sampled represents the project as a whole so that the ratio will project out to 27k overall.
- ChrisLomont 9y agoSampling 3.3% of lines in a 72 million line codebase, especially across so many sub projects, is statistically significant enough to make claims like this with decent certainty. That's a few million sample lines tested.
- ordu 9y agoOh... We can ask about way sub projects were selected, and are they can be called a representative sample. Nevertheless I agree with you: there are no obvious way to do a better estimation, so words "fair" and "transparent" seem right in place there. Of course only if Karpov does not try to mislead his readers deliberately by choosing sub projects by some criterion that correlates with error rate.
- mnarayan01 9y agoEven if the sampling was rigorously random and no projects were dropped after starting, there would still likely be a bias towards small, "long tail", projects. It seems plausible that these types of projects would have more errors (or vice versa).
- MikusR 9y agoDid they include Linux kernel or other 3rd party code in those 72 million lines? If yes then that is one buggy kernel.