Re: [gnugo-devel] tsumego Evan Berggren Daniel

[Evan Berggren Daniel]

On Thu, 2 Oct 2003, Wolfgang Manner wrote:

> Evan Berggren Daniel was kind enough to send me the under the stones
> problems which he scanned in from yutopian. i excluded 1 example
> (aoc9) and looked at all the others.

You're welcome :)

> definitions:
> m0 move not found
> m1 move found with low priority (not highest move value)
> m2 move found with highest move value in the corner
> m3 move found

I take it from this you're asking gnugo to generate a move for the
position?  is there a reason not to use the owl status of the group?  No
one ever poses tsumego as whole board problems normally, so it would make
sense to evaluate these as "black to play and live" or equivalent, not
"black to make a whole board play."  I believe I included a .tst file that
runs all the positions in all the problems in the tar file I sent you;
from the regression/ directory, run ./eval.sh aoc.tst.

> i have annotated the files, but since Evan didnt send the files around
> for copy-right reasons, i wont do that either. However, if he agrees,
> i can send it to people individually.

Fine by me.

> so i think one should work very hard to improve that.
> (i guess you have already invested about 100 man years in
> this business)

Not quite that much, I think :)

However, there are some known problems that are particularly highlighted
by this problem set, mostly related to strange shapes (under the stones
shapes are unlike most others; you'll find that people who haven't looked
at the problems are particularly bad at them as well, I think).  The other
major problem is cutting points in eyespaces, like in the position you
posted.  The first can be solved by the addition of patterns or some more
sophisticated technique, the second probably requires major work on the
owl code.  The third problem also relates to eye valuation, as in the
following position (from problem #1):

|OOO
|XXO
|XXO
|.OX
+---

How many eyes is this for white?  Gnugo currently believes it is one eye.
This is clearly wrong, but the position is more subtle than immediately
obvious, because if black is asked to prove it, then he has to capture
from the outside first, otherwise white can get a ko (black takes the
stone, white takes the black stones):

|OOO
|..O
|21O
|.OX
+---

Now white has a ko for two eyes.

Actually, this is probably equivalent to the cutting points problem in
many ways, but is probably different enough to be worth considering.

Evan Daniel