Re: [gnugo-devel] address@hidden: computer-go: 13x13 CGoTournament 2002-03 (12)]

[Gunnar Farneback]

Dan wrote:
> Note the interesting comments about GNU Go.

Indeed.

Stefan Mertin wrote:
> Most surprising is the newer version 3.2 of GnuGo, 
> which I am very happy to have included now:
> but with level=10 (standard) the results were nearly the same 
> as with version 3.0 - so I rised the playing level to 15 for this program 
> (computing time mostly was still less than 30 min. per game) 
> and the success is incredible: GnuGo is now to be found among 
> the worlds strongest Go programs  - 
> and it's free! Big felicitations for this great work!

Increasing the level affects a lot of depth parameters. It would be
interesting to know which of those pay off best.

> I today actualized my website
> http://geocities.com/comp_go/13x13.html
> with the new results and all the 744 (!) games,

I've added the GNU Go 3.2 games to the regression/games/mertin13x13
directory. Appended is a test suite for some of the mistakes in those
games.

It's encouraging to see that current CVS already solves 26 of the 43
tests. However, increasing the level to 15 reduces the number to 23
(1, 3, and 32 no longer pass).

I recommend taking a look at the games against Many Faces. GNU Go wins
4 out of 6 and the general wildness factor is rather high.

> Meanwhile also my next tournament, the final 19x19 CGoT has started 
> and I would like to invite all programmers to let me know, 
> which versions of their program you would like me to let participate!

GNU Go 3.4 maybe. It doesn't exist yet but judging from the 13x13
tournament I'd guess the 19x19 one will take at least a year or two to
finish.

/Gunnar

# This is a collection of mistakes by GNU Go in Stefan Mertin's private
# 13x13 - Computer Go tournament in 2001-2002 (see
# http://www.geocities.com/comp_go/13x13.html).
# 
# While the games used in 13x13.tst were played by GNU Go 3.0, this test
# suite covers games played at the end of the tournament by GNU Go 3.2
# at level 15.
# 
# The opponents and results in these games are:
# 
# Game       GNU Go  Result  Opponent
# name       color
# DAGN1.sgf  white   +40.5   Dariush
# DAGN2.sgf  white   +52.5   Dariush
# DAGN3.sgf  white   + 2.5   Dariush
# GNDA1.sgf  black   +39.5   Dariush
# GNDA2.sgf  black   +34.5   Dariush
# GNDA3.sgf  black   +32.5   Dariush
# GAGN1.sgf  white   -29.5   GoAhead
# GAGN2.sgf  white   + 0.5   GoAhead
# GAGN3.sgf  white   +16.5   GoAhead
# GNGA1.sgf  black   + 2.5   GoAhead
# GNGA2.sgf  black   +67.5   GoAhead
# GNGA3.sgf  black   - 5.5   GoAhead
# GHGN1.sgf  white   +33.5   Goliath
# GHGN2.sgf  white   + 6.5   Goliath
# GHGN3.sgf  white   -48.5   Goliath
# GNGH1.sgf  black   + 5.5   Goliath
# GNGH2.sgf  black   +13.5   Goliath
# GNGH3.sgf  black   -59.5   Goliath
# GIGN1.sgf  white   +12.5   Go Intellect
# GIGN2.sgf  white   - 0.5   Go Intellect
# GIGN3.sgf  white   +21.5   Go Intellect
# GNGI1.sgf  black   +31.5   Go Intellect
# GNGI2.sgf  black   +18.5   Go Intellect
# GNGI3.sgf  black   + 2.5   Go Intellect
# GMGN1.sgf  white   + 3.5   GoeMate
# GMGN2.sgf  white   -10.5   GoeMate
# GMGN3.sgf  white   - 6.5   GoeMate
# GNGM1.sgf  black   + 9.5   GoeMate
# GNGM2.sgf  black   -12.5   GoeMate
# GNGM3.sgf  black   - 6.5   GoeMate
# GPGN1.sgf  white   - 6.5   Go Professional
# GPGN3.sgf  white   -15.5   Go Professional
# GPGN5.sgf  white   -16.5   Go Professional
# GNGP1.sgf  black   +27.5   Go Professional
# GNGP3.sgf  black   - 0.5   Go Professional
# GNGP5.sgf  black   - 1.5   Go Professional
# GXGN1.sgf  white   +25.5   Go4++
# GXGN2.sgf  white   -20.5   Go4++
# GXGN3.sgf  white   -14.5   Go4++
# GNGX1.sgf  black   -10.5   Go4++
# GNGX2.sgf  black   +11.5   Go4++
# GNGX3.sgf  black   - 4.5   Go4++
# HTGN1.sgf  white   +10.5   HandTalk
# HTGN2.sgf  white   +23.5   HandTalk
# HTGN3.sgf  white   + 6.5   HandTalk
# GNHT1.sgf  black   + 5.5   HandTalk
# GNHT2.sgf  black   -13.5   HandTalk
# GNHT3.sgf  black   +16.5   HandTalk
# MFGN1.sgf  white   +43.5   Many Faces of Go
# MFGN2.sgf  white   +35.5   Many Faces of Go
# MFGN3.sgf  white   -42.5   Many Faces of Go
# GNMF1.sgf  black   +17.5   Many Faces of Go
# GNMF2.sgf  black   +44.5   Many Faces of Go
# GNMF3.sgf  black   -18.5   Many Faces of Go
# WLGN1.sgf  white   - 1.5   Wulu
# WLGN2.sgf  white   -41.5   Wulu
# WLGN3.sgf  white   + 5.5   Wulu
# GNWL1.sgf  black   +52.5   Wulu
# GNWL2.sgf  black   +20.5   Wulu
# GNWL3.sgf  black   -13.5   Wulu
# 
# The variations in the sgf-files are due to Stefan Mertin.

loadsgf games/mertin13x13/GNDA3.sgf 61
1 gg_genmove black
#? [M4]

loadsgf games/mertin13x13/GAGN1.sgf 12
2 gg_genmove white
#? [!B12]

# N5 is locally worse than M5.
loadsgf games/mertin13x13/GAGN1.sgf 28
3 restricted_genmove white M5 N5
#? [M5]

# Hane at K1 is not appropriate.
loadsgf games/mertin13x13/GAGN1.sgf 42
4 restricted_genmove white F4 H2 K1
#? [F4]

# Unreasonable to block at A5.
loadsgf games/mertin13x13/GAGN1.sgf 66
5 restricted_genmove white A5 B4 C4 C5
#? [!A5]

# B4 unreasonable. C4 locally the best move.
loadsgf games/mertin13x13/GAGN1.sgf 68
6 restricted_genmove white B4 C4
#? [C4]

# H11 allows a combination attack starting with K10.
loadsgf games/mertin13x13/GNGA3.sgf 67
7 gg_genmove black
#? [J11]

# F10 blocks and connects in the best way. F11 is clearly inferior.
loadsgf games/mertin13x13/GNGA3.sgf 81
8 gg_genmove black
#? [F10]

# Most other moves are better than the game move at C6.
loadsgf games/mertin13x13/GHGN1.sgf 26
9 gg_genmove white
#? [!C6]

# Most other moves are better than the game move at M6.
loadsgf games/mertin13x13/GHGN2.sgf 48
10 gg_genmove white
#? [!M6]

# Most other moves are better than the game move at M6.
loadsgf games/mertin13x13/GHGN2.sgf 52
11 gg_genmove white
#? [L6]

# J5 not joseki.
loadsgf games/mertin13x13/GHGN3.sgf 8
12 gg_genmove white
#? [!J5]

# M13 is a very odd move.
loadsgf games/mertin13x13/GHGN3.sgf 38
13 gg_genmove white
#? [!M13]

# J9 is huge.
loadsgf games/mertin13x13/GNGH2.sgf 23
14 gg_genmove black
#? [J9]

# N11 is meaningless.
loadsgf games/mertin13x13/GNGH2.sgf 71
15 gg_genmove black
#? [B5]

# H2 completely misses the point. Necessary to block at G4.
loadsgf games/mertin13x13/GNGH3.sgf 57
16 gg_genmove black
#? [G4]

# F1 gives seki.
loadsgf games/mertin13x13/GNGH3.sgf 111
17 gg_genmove black
#? [F1]

# Necessary to defend the bottom.
loadsgf games/mertin13x13/GIGN2.sgf 18
18 gg_genmove white
#? [F3]

# One critical point left at B5. K10 is dame.
loadsgf games/mertin13x13/GIGN2.sgf 60
19 gg_genmove white
#? [B5]

# K10 is unreasonable.
loadsgf games/mertin13x13/GIGN3.sgf 92
20 gg_genmove white
#? [J10]

# J13 fails right out.
loadsgf games/mertin13x13/GIGN3.sgf 94
21 gg_genmove white
#? [!J13]

# A6 bad shape.
loadsgf games/mertin13x13/GIGN3.sgf 100
22 gg_genmove white
#? [B6|B7]

# J1 gives ko. K1 kills unconditionally.
loadsgf games/mertin13x13/GNGI1.sgf 95
23 gg_genmove black
#? [K1]

# N11 is clearly the worst way to end the ko. M13 and M9 are both better.
loadsgf games/mertin13x13/GNGI3.sgf 63
24 gg_genmove black
#? [M9|M13]

# M13 lives independently. K13 gives a semeai.
loadsgf games/mertin13x13/GNGI3.sgf 65
25 gg_genmove black
#? [M13]

# Must connect at D3.
loadsgf games/mertin13x13/GMGN2.sgf 30
26 gg_genmove white
#? [D3]

# B2 inefficient way to live.
loadsgf games/mertin13x13/GMGN2.sgf 92
27 gg_genmove white
#? [C4]

# H9 can be cut off.
loadsgf games/mertin13x13/GMGN3.sgf 30
28 gg_genmove white
#? [!H9]

# Must connect at F3.
loadsgf games/mertin13x13/GNGM3.sgf 11
29 gg_genmove black
#? [F3]

loadsgf games/mertin13x13/GPGN1.sgf 62
30 gg_genmove white
#? [N8]

loadsgf games/mertin13x13/GPGN1.sgf 64
31 gg_genmove white
#? [M7]

# G5 is dame. H4 is the point needed to win the game.
loadsgf games/mertin13x13/GNGP3.sgf 55
32 gg_genmove black
#? [H4]

# F6 clearly better than E6.
loadsgf games/mertin13x13/GNGP5.sgf 73
33 restricted_genmove black E6 F6
#? [F6]

# E5 is just cut off and captured.
loadsgf games/mertin13x13/GXGN2.sgf 48
34 gg_genmove white
#? [!E5]

# D10 captures on too small scale.
loadsgf games/mertin13x13/GXGN2.sgf 52
35 gg_genmove white
#? [!D10]

# Low enclosure better.
loadsgf games/mertin13x13/GXGN3.sgf 12
36 restricted_genmove white J3 J4 H3 H4
#? [J3|H3]

# Must follow up A6 with A4, which in any case is sente.
loadsgf games/mertin13x13/GNGX1.sgf 51
37 gg_genmove black
#? [A4]

# E8 clearly bigger than E7.
loadsgf games/mertin13x13/HTGN3.sgf 102
38 restricted_genmove white E7 E8
#? [E8]

# F10 needed for seki.
loadsgf games/mertin13x13/MFGN3.sgf 58
39 gg_genmove white
#? [F10]

# Must connect at F8.
loadsgf games/mertin13x13/WLGN2.sgf 74
40 gg_genmove white
#? [F8]

loadsgf games/mertin13x13/GNWL1.sgf 61
41 gg_genmove black
#? [J6]

# L7 way overconcentrated.
loadsgf games/mertin13x13/GNWL2.sgf 35
42 restricted_genmove black L7 L13
#? [L13]

# J3 is of course bigger than J1.
loadsgf games/mertin13x13/GNWL3.sgf 109
43 restricted_genmove black J1 J2 J3
#? [J3]