[Shootout-list] scoring discrepancy
Brent Fulgham
bfulg@pacbell.net
Tue, 6 Jul 2004 15:00:12 -0700 (PDT)
--- Stephen Weeks <sweeks@sweeks.com> wrote:
>
> I wrote some code to compute the scores using the
> data in
> .craps.table. For the default case, the scores that
> my code computes
> are different than the scores computed by PHP that
> appear on the web
> page.
What does your code do? Perhaps I have a typo in my
implementation.
The "make_craps" program does the following:
1. It reads through the data file for each benchmark,
finding the lowest non-zero performer (perhaps this is
the problem?) and uses this as its "best" case.
Removing the test for 0-value scores doesn't seem to
move our scores closer to agreement...
2. For all entries, it computes the score as:
1 / (1 + SCALE * log(n/best)/log(2))
> As a simple sanity check, one can count by
> hand the number of tests missing from clean. I
> count 3, which agrees with what my code
> computes, and disagrees with the web page, which
> says 4.
This is because I neglected to check in the Clean
source file, though the "marked-up" HTML version is in
the archive.
What algorithm are you using to compute the score?
-Brent