Re: [gnugo-devel] endgame regression

[Arend Bayer]

On Sat, 21 Sep 2002, Evan Berggren Daniel wrote:

> As I mentioned earlier on nngs, GNU Go doesn't notice that the second hane
> is gote.  It values it at about 6 points, the same as the first hane, when
> it should probably be about 3.  There are other moves on the board worth
> 4-5 points, and it looks like those are better.

If you run
gnugo -l gnugo-3.3.8-evand-200209211825.sgf -L 154 -t
gnugo -l gnugo-3.3.8-evand-200209211825.sgf -L 158 -t
you will see that both moves are valued territorially by around 3
points. T14 gets a bonus for the followup value. In the 2nd case you
see:

  P19: 3.00 - change in territory
  P19:   0.00 - total followup value, added 0.00 as territorial followup
  P19: 3.00 - added due to followup (0.00) and reverse followup values (6.00)
  P19: 0.20 - connects strings (connect value 2, shape factor 0.040)
Move generation values P19 to 6.20

This means GNU Go treats this hane as reverse sente and that is why
it values it with 6 points. (The correct value could be even higher
because after P19, Q16 is absolutely sente.)

Guessing why GNU Go makes a particular move is usually misleading. There
is a lot of information in the output of -t (and you can get even
more if you use relevant debug flags).

Arend