Re: Advice on evaluation

[cesczx@xxxxxxxxxxxxxxxxxx (Chen Zhixing)]

David Fotland wrote:
>
>I know that Go4++ evaluates always pessimistically for the side to
>move, so the evaluation would be different if colors were reversed,
>and it seems to work quite well.  
>
   In my opinion, the pessimistical estimation leads to conservative style
of moving. There are two aspects about this.
   1. GO4++ conservatively estimates the territory of its own side, this
leads to a conservative style in fuseki, but it can get small territory in
mid-game.  Thus, it underestimate the value of a moyo, so GO4++ is not good
to form big moyo.  On the other hand, other programs may overestimate the
value of influence, so they may move on some points getting no or very few
real points but increasing relative influence, while GO4++ does not move to
get such unclear 'territory', it will rather to get some small territory.
As a result, GO4++ often lags behind in the earlier stage when it plays vs.
other strong programs such as Handtalk, but it can get a lot of small
territory in the midgame and may become ahead. 
   2. GO4++ exaggerates the danger of its own group, this often leads to
abandon some of its own group even when it can easily be saved.  In 1996 Ing
Cup GO4++ made a tenuki when its opponent, Wulu, attacked a group of GO4++,
leading to death of the group.  Fortunately, Wulu was not so strong in that
year and lost many point in the endgame stage.  Eventually GO4++ won by a
few points.  In 1998 Ing Cup, the situation was almost reproduced in the
game between GO4++ and Wulu--GO4++ lost a group by tenuki. Meanwhile Wulu
had got much stronger, so GO4++ lost the game.
   In my opinion, the conservative style of GO4++ leads to good results in
tournament, but GO4++ cannot be a good merchandized software of go.  It
always plays too simple and too monotonical, not leading to complicated
positions which let the players enjoy more.  I eagerly hope that Reiss will
improve his GO4++ in the style, and I also eagerly hope that David Fotland
will not let his MFGO become conservative. 

   I have another advice to David Fotland.  I know that MFGO treats single
jump as breakable.  I heard a comment by David last year that the upper
white group like the following figure is estimated dead, because there are
many break-points at the b's, then the upper white group cannot be linked to
the downward group.  I think a single jump may be better treated as
unbreakable, so that all the white stones in the following figure is in one
group.

....................
............X...OOX.
...............O.X..
............X..b..X.
...............O....
............X..b.X..
...............O....
............X..b....
...............O.X..
..............O.....
.............O...X..
....................
....................
...

Chen Zhixing
www.wulu.com