[gnugo-devel] Re: Timing of paul_3_15.5

[Paul Pogonyshev]

> The original time that I got when I tested paul_3_15.5 was
> apparently way off because I was running gnome and various
> things (netscape, emacs, texmacs, and ssh connections to
> math and match). The regressions took 10610 seconds with
> the patch, 9582 seconds without the patch.
>
> So I ran them again at night on an unloaded machine. The results
> were very different. The second runs gave 9183 seconds with the
> patch, 8964 seconds without it. The extra 219 seconds is about
> 2.44 %.

it seems to be too large anyway. maybe the best way to understand
what's going on is to check how often the patch changes topological
eye valuations. i'll probably do this tonight. i think that 2.5%
for a patch that solves only one test (though quite important as
it seems to me) is too much.

Paul