Attached is a 9x9 game by Mogo ( Black ) versus an amateur shodan ( Joe C.,
White ).
>From the log output, Mogo was unaware that it was waaay behind until move 60.
>Oddly, move 60 was a pass by White; there may be a bug in this particular
>version, since this would have been a good time to either pass or resign. Mogo
>continued to move 68, by which time it was about 20 points behind.
I'm only 8 kyu AGA, but I can see that at move 35, Mogo was lost with 100%
probability, call the undertakers.
I offer this in support of the notion that, without an understanding of
life-and-death, MC algorithms can and do err in estimating the probability of
winning. This isn't a matter of choosing 85% of this versus 80% of that; it's a
matter of incorrectly believing that one has good prospects of winning a game
which is actually 100% lost. When I do this, it is called "hallucination" -
which I do often enough to hinder my AGA rank, unfortunately.
The log file just before move 59:
0.564769(3137%) || 215467/482517(45%)(32%)(99%/0.44)|| 9274 |46 || C9 C3 B3 C2
B1 B4 H1 C4 G2 J5 ||SSP:C9 A5 J1 A9 C3 ||PSP:C9 A5 J1 A9 C3
0.564766( 0%) || 215468/482518(45%)(32%)(99%/0.44) || 9274 |46 || C9 C3 B3 C2
B1 B4 H1 C4 G2 J5 ||SSP:C9 A5 J1 A9 C3 ||PSP:C9 A5 J1 A9 C3
100002 simulations(average length:0) done, time used:
72.47 seconds.( 1379.9 games/sec)
If the algorithm could have correctly determined the actual status of the game
by move 35, then it could presumably have chosen better alternative plays
furher back. Existing algorithms apparently make large errors in scoring
settled positions. That suggests a possible
avenue for improvement.
Thanks, all!
-- Terry
____________________________________________________________________________________
Looking for last minute shopping deals?
Find them fast with Yahoo! Search.
http://tools.search.yahoo.com/newsearch/category.php?category=shopping(;FF[4]CA[UTF-8]AP[GoGui:1.0.3]SZ[9]
KM[6.5]PB[MoGo]DT[2007-12-06]RE[W+Resign]
;B[ed];W[cc];B[ef];W[gf];B[gd];W[fh];B[cd];W[dc];B[hg];W[hf]
;B[bd];W[fb];B[gb];W[fc];B[gc];W[bc];B[fg];W[gg];B[eh];W[ff]
;B[eg];W[fd];B[gh];W[hh];B[he];W[gi];B[dd];W[hb];B[ha];W[ic]
;B[ec];W[eb];B[id];W[ge];B[ac];W[ie];B[if];W[ig];B[ee];W[fe]
;B[ib];W[ia];B[ga];W[fa];B[hd];W[ib];B[bh];W[ab];B[ad];W[bb]
;B[ba];W[da];B[dh];W[ei];B[ai];W[di];B[ci];W[fi];B[ca];W[]
;B[ae];W[ie];B[hc];W[ib];B[db];W[cb];B[hi];W[ii])
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
