Re: [gnugo-devel] small plays at the end of the game

[David G Doshay]

On 2, Dec 2004, at 5:08 PM, address@hidden wrote:

> > Why is it that GNU Go has move evaluations  of 0.0 for all possible
> > moves at the end of the game, but it has a specific best move? This
> > quirk resulted in SlugGo making end game errors because I had not
> > expected that GNU Go would do that.
>
> GNU Go's policy is to fill all liberties, partly because filling
> a liberty is worth one point in Chinese rules, and partly because
> mistakes can be made while filling a liberty if backfilling is
> needed, so the best discipline is to play this part of the
> endgame out to the end.
>
> Therefore there's the function fill_liberty() whose purpose
> is to propose a move to fill a liberty. It is called from genmove:
>
>   if (val < 0.0
>       && fill_liberty(move, color)
>       && (!allowed_moves || allowed_moves[*move])) {
>     val = 1.0;
>     TRACE("Filling a liberty at %1m\n", *move);
>     record_top_move(*move, val);
>     move_considered(*move, val);
>     time_report(1, "fill liberty", NO_MOVE, 1.0);
>   }
>
> Of course it has to propose a particular move.

Yes, this is what I am seeing, situations where backfilling can be
done right or wrong. GNU Go did the right thing with the floating
point value of the right move being 0.0 and the value of the wrong
move being 0.0.

Thanks,
David