[gnugo-devel] Twogtp test replay result

[SP lee]

I have let the current cvs replayed (as white) the attached game between gnugo 3.4 (white) and gnugo 3.5.4 and got the following result. I have to say that for most moves, i.e. move 38, 44, 50, 52, 58, 96, 100, 124, 138 and 218 gnugo 3.4 got the better one. The rest of the difference in moves are not so significant.

 

Ruleset:      Japanese
Game Name:    TWOGTP~1
Game Comment: Engines disagreed on results:
White claimed W+83.5
Black claimed W+131.5
Black Player: GNU Go 3.5.4 (random seed 1075129494)
White Player: GNU Go 3.4 (random seed 1075129493)
Board Size:   19
Handicap      0
Komi:         6.5
Move Number:  0
To Move:      black
Computer player: White
Move 2 (white): GNU Go plays D17 (75.00) - Game move D16 (74.00)
Move 4 (white): GNU Go plays D4 (75.00) - Game move D3 (74.00)
Move 6 (white): GNU Go plays O17 (75.00) - Game move O3 (74.00)
Move 26 (white): GNU Go plays N16 (24.12) - Game move R14 (24.13)
Move 38 (white): GNU Go plays R3 (23.88) - Game move Q6

 

Q6 is a 'must', otherwise black escapes easily. R3 is too weak.

Move 44 (white): GNU Go plays G3 (16.14) - Game move Q3 (3.57)

 

Q3 got the corner cleanly, although not so efficient. If black plays Q3 instead, there is much space to thrash.

Move 50 (white): GNU Go plays S10 (14.43) - Game move M13 (12.14)

 

M13 reduces black territory, which is much bigger than S10. S10 might make the single black stone more harder to live locally, but there is a too wide escape route.

Move 52 (white): GNU Go plays R10 (10.28) - Game move K15 (8.26)

 

Again K15 is a good territory reducing move. R10 is still a weak move.

Move 58 (white): GNU Go plays R9 (16.82) - Game move Q8 (14.76)

 

Q8 is of a good and secure shape for the corner. R9 is less efficient.

Move 86 (white): GNU Go plays F14 (8.15) - Game move D11 (7.67)
Move 94 (white): GNU Go plays F12 (6.95) - Game move F14 (6.94)
Move 96 (white): GNU Go plays D19 (6.20) - Game move B9 (5.40)

 

Although the difference is not much, but B9 has a much larger followup value.

Move 100 (white): GNU Go plays D19 (6.20) - Game move E2

 

Again D19 is gote and E2 prevents black to squeeze the white area.

Move 124 (white): GNU Go plays J2 (9.70) - Game move M1 (9.51)

 

M1 saves two white stones while J2 only saves one.

Move 130 (white): GNU Go plays Q18 (4.37) - Game move O13 (4.21)
Move 138 (white): GNU Go plays S10 (4.32) - Game move S15 (3.89)

 

S15 kills black while S10 doesn't.

Move 218 (white): GNU Go plays G18 (34.52) - Game move H18 (14.00)

Again G18 kills and H18 doesn't.

 

By the way I failed to find the source of the problem. I had tried to disable the break-in feature, the result is about the same.

 

 

SP Lee

twogtp-2-004.sgf
Description: Binary data