Re: [gnugo-devel] arend_1_25.3: less (??) experimental influence

[Daniel Bump]

Arend wrote:

> After this patch, --experimental-influence produces 58 PASSes and 53 FAILs.
> which is quite dubious, of course. However, my personal impression was
> that the current state should already be an improvement in playing strength
> compared to standard influence.

I'm running twogtp to test this and so far the results look good.

Both programs are current CVS + Arend's patch. W is compiled
with --enable-experimental-influence --enable-alternate-connections
--enable-experimental-connections and B is compiled omitting
the experimental influence. (I actually didn't need to recompile
because there's command line options --experimental-influence
and --standard-influence.)

Currently W has won 8 out of 9 games:

Result according to W: W+33.5
Result according to W: B+43.5
Result according to W: W+13.5
Result according to W: W+15.5
Result according to W: W+12.5
Result according to W: W+9.5
Result according to W: W+39.5
Result according to W: W+33.5
Result according to W: W+16.5

I suppose I'll run it about another day then collect the
games. We can get some information from individual games
by replaying them to find out when the two versions
differ on particular moves.

Dan

PS another game just finished and W won by 29.5 points.