Initial commit of GNU Go v3.8.

[sgk-go] / regression / games / seki_exotics / README-MANNER

Some remarks concerning the files collected:

seki1.sgf

The "classical" seki. gnugo plays correctly, but the score is wrong.

seki2.sgf

Each side has an eye

seki3.sgf

Another one.

seki_stupid.sgf

Here i wanted to see, if gnugo plays the stupid move C 1. Well it doesnt.
But maybe it is good to keep the file, just to make sure that future
"improvements" dont spoil it.

sekichallenge.sgf

Thats what it is. gnugo makes mistakes and loses everything. But i get
the impression that the reading depth is just a little bit too short.

seki_score.sgf

a seki. unless i counted it wrongly, assuming an equal number of
prisoners
the japanese rules give:

territory b 96  w 97
prisoners b  0  w  0
seki      b  0  w  0

the chinese rules give:

territory b 96  w 97
stones    b 79  w 79
seki      b 8.5 w 1.5

so there is a difference of 7 points, depending on which rules you take.

twoko_seki1.sgf

A double ko seki. with the present program and playing japanese rules
the eternal loop can only be avoided, if both players have enough
intelligence to see that it is impossible to win. i said already
before that i think that the program is wrong. that is after two
passes, i thought the game is finished, but do_pass allows gnugo
to play another move, so the loop continues. if we change do_pass,
it would be good enough to have one intelligent player to avoid
the loop. Since we know nothing about the intelligence of our
opponent, the intelligent player would have to be gnugo.
So if we stick to japanese rules, this needs a detailed analysis
of the situation.
In principle the japanese rules define in their precedents
that this position is seki, should gnugo know the precedents ?
chinese rules avoid the loop and give the interesting result, that
whoever insists on playing, loses everything. maybe not everybody
knows that already by heart, so it goes like this:

w takes ko 1
b takes ko 2
w tenuki
b takes ko 1
w takes ko 2
b tenuki

finally there will be no playable point left on the board.
lets assume that b has played the last possible move.
then if w insists:

w takes ko 1
b takes ko 2
w passes
b takes ko 1

now w would like to take ko 2, but that is forbidden, because it
reproduces a position, which has already existed before.
so

w passes
b takes all.

If there is another ko somewhere else on the board, then this
double ko would be an example of what Ing calls a disturbing ko.


twoko_seki2.sgf

Another double ko seki. twoko_seki1 has two groups with an eye each,
this one has one liberty common to both groups.

twoko_live.sgf

B is alive.


twoko_dead.sgf

W is dead.


fourko.sgf

A quadruple ko. I have never heard about that, but i guess the
japanese will have to cancel the game. with chinese rules one
can play it of course, but it is rather tricky.

onadare_ko.sgf

The double ko variation of the onadare joseki.

onadare_ko2.sgf

Here i have added a ko in the lower right corner. This is completely
different from twoko_seki1.sgf plus another ko. here the black group
in the double ko is alive. however white has an unlimited supply
of ko threats, so it will win any other ko. If i understand this
correctly, the japanese do not consider this to be a triple ko
I thought that Ing should call the double ko a "fighting ko",
but apparently he doesnt. So the Ing rules would have to treat
the two cases twoko_seki1.sgf plus another ko and the present one
the same way.
gnugo does not know how to handle this.

bent4.sgf

gnugo handles this correctly. when i said maybe for the wrong reasons
Gunnar said, what do i mean. well first one should look, if it does
all possible variations correctly. But what i really was afraid of, was
maybe it treats it the same way it handles seki. only for seki it is
wrong and here it is right. anyway, whenever the seki has been fixed, 
one should look, if this is still ok.

bent4_2.sgf

Another one. This one is also handled correctly.


seki_bent4.sgf

The combination of a bent4 with a seki, which gives an unremovable
ko threat. The japanese rules kill the bent4 anyway. gnugo has
various difficulties.


doublebent4.sgf

I guess the japanese rules say that both groups are dead ???
Apart from that gnugo handles it correctly.

threeko1.sgf

The "classical" triple ko. i guess, if we play japanese rules,
gnugo should realise that the game should be cancelled.
Playing chinese rules one can usually neutralise two ko's
and then play the one which is remaining as before.

threeko2.sgf

This is a fake. white can never win this.

threeko3.sgf

Similar to threeko2.sgf, with a slight difference. If it is
white's turn it can play B 5 and then it becomes a double ko
which black cannot win.
If it is black's turn, the japanese will cancel the game and
for the chinese white can never win.

chosei.sgf

A chosei( long life / eternal life ).
Apart from the fact that the game should be cancelled,
gnugo makes a lot of mistakes and loses everything.
It seems everybody who writes about that, shows the same diagram.
Maybe this is the only one ?
If people think that this is interesting, I would have to do
some work.

junkan-ko.sgf

A junkan-ko (round robin ko). Everything I wrote about chosei
is valid here as well. i thought a little about these positions,
but i dont want to bore everybody, unless somebody really wants
to hear about it.