1 (;GM[1]FF[4]CA[UTF-8]AP[CGoban:3]ST[2]
2 RU[Chinese]SZ[19]KM[7.50]TM[14400]
3 PW[StoneGrid]PB[pachi2]BR[3d]DT[2011-08-24]GC[51 +E14:bad!{ladder} pachi plays a broken ladder, which almost cost the game]PC[The KGS Go Server at http://www.gokgs.com/]C[StoneGrid [-\]: GTP Engine for StoneGrid (white): stonegrid version 0.3.4 48
4 pachi2 [3d\]: GTP Engine for pachi2 (black): Pachi Distributed Engine version 8.99 (Hakugen-devel): I'm playing the distributed engine. When I'm losing, I will resign, if I think I win, I play until you pass. Anyone can send me 'winrate' in private chat to get my assessment of the position.
6 ;B[pp]BL[14399.379]C[maproomad [-\]: pachi2: In 2367711 playouts at 64 machines, white M10 can win with 96.89% probability.
7 maproomad [-\]: I guess that is left over from the previous round
8 maproomad [-\]: pachi2: In 10750 playouts at 64 machines, black Q4 can win with 100.00% probability.
9 maproomad [-\]: maybe not
11 ;W[mg]WL[14246.492]C[BinX [-\]: black q4 can win with a 100% probability ? kind of a divine first move ?
13 ;B[cj]BL[14189.181]C[BinX [-\]: oh :)
19 ;B[pq]BL[13977.422]C[maproomad [-\]: pachi2: In 125164015 playouts at 64 machines, black Q3 can win with 48.70% probability.
22 ;B[co]BL[13775.243]C[uurtamo [-\]: but pachi2 could have checked each of the 6-ply inbetween then. :)
24 ;W[dn]WL[13637.551]C[goIngo [-\]: is this avantgarde go?
25 uurtamo [-\]: it's curious.
26 BinX [-\]: oh, pachi lose confidence with q3 already :o
28 ;B[cn]BL[13669.876]C[uurtamo [-\]: yes, to go from 100% to 48.7% in a sequence of moves all of which could be explored, means something isn't right. :)
29 blob [5k\]: sure, n13, to hell with joseki
30 uurtamo [-\]: i don't understand n13 well enough to say whether it's joseki or not.
32 ;W[do]WL[13486.901]C[blob [5k\]: n13 of course, is not joseki
33 blob [5k\]: not yet anyway.......
34 uurtamo [-\]: joseki isn't about what is in a book, it's about what makes the board equal.
36 blob [5k\]: it's what has the historical concensus of being a solid move
39 ;B[dm]BL[13564.821]C[uurtamo [-\]: it's what has the historical concensus of giving a fair outcome for both players.
40 blob [5k\]: which n13 hasn't
41 uurtamo [-\]: fair enough.
42 uurtamo [-\]: however, it might give a fair outcome for both players. all i'm saying.
43 blob [5k\]: "might".... which is why i said "not yet anyway"
46 ;B[em]BL[13457.786]C[uurtamo [-\]: you're fixated on the "historical concensus" and i'm fixated on the "fair outcome". no biggie.
47 blob [5k\]: sure, but you just misunderstand the meaning of joseki apparently
48 uurtamo [-\]: could very well be the case.
50 ;W[cp]WL[13186.961]C[blob [5k\]: the jo in joseki means "established" (among other things)
51 blob [5k\]: i.o.w., an established move
53 ;B[bm]BL[13363.95]C[blob [5k\]: ....why do these bots get so much time?
54 uurtamo [-\]: because it's the "slow play tournament"
55 blob [5k\]: it sort of rewards brute force approaches to machine go, which i think isn't really desirable
57 ;W[cc]WL[13039.78]C[uurtamo [-\]: isn't desirable in what sense?
58 blob [5k\]: is there a fast tournament?
59 uurtamo [-\]: have you ever tried to write a machine go program?
60 blob [5k\]: i've started one yes
61 uurtamo [-\]: there are tournaments every month, and a server where programs are continuously playing one another
62 blob [5k\]: by not desirable i mean that i find efficiency in computation intuitively more "human"
63 blob [5k\]: than monte carlo-esque stuff anyway
64 uurtamo [-\]: http://www.weddslist.com/kgs
65 blob [5k\]: thanks for that
67 ;B[kq]BL[13166.445]C[uurtamo [-\]: i think that you want computation to be a certain way that it might not be.
68 blob [5k\]: look, the status quo is that humans are stronger than machines at go, yet use less resources (energy wise)
69 blob [5k\]: i think it's natural to learn by that example somehow
70 uurtamo [-\]: that idea was very popular in the 70's.
72 uurtamo [-\]: i'm just suggesting that other approaches have been quite a bit more successful, not just in computer go, but machine learning in general and that you ought not discount their success.
73 uurtamo [-\]: there's a mistake in being fixated on what "seems natural"
74 uurtamo [-\]: computers aren't all that natural and neither is computation.
75 blob [5k\]: i am familiar with ai techniques
77 ;W[ko]WL[12892.264]C[uurtamo [-\]: there's a mailing list about computer go if you're interested in seeing what people think about.
78 blob [5k\]: i'm saying that it seems that human level of higher abstraction can lead to significant gains in efficiency
79 blob [5k\]: i know what's out there already.....
81 ;B[hq]BL[12958.564]C[blob [5k\]: there's a huge gain to be had over mere monte carlo, by clustering and classifying patterns dynamically
82 uurtamo [-\]: well, the easiest way to test your ideas is on CGOS.
83 uurtamo [-\]: http://cgos.boardspace.net
84 blob [5k\]: yeah, as soon as i get some more time on my hands ):
86 uurtamo [-\]: because otherwise you are simply fantasizing.
87 uurtamo [-\]: also, "mere monte carlo" has developed quite a bit.
88 blob [5k\]: i already have a working ML go prototype
89 uurtamo [-\]: great. you can probably hook it into cgos with very little trouble, then.
91 ;W[jp]WL[12747.122]C[blob [5k\]: probably, but i want to work on it more first
92 blob [5k\]: it's still weak
93 uurtamo [-\]: cgos will let you see its level exactly compared with other programs.
94 blob [5k\]: i'll keep that in mind, thanks
96 ;B[jq]BL[12854.702]C[maproomad [-\]: pachi2: In 76190780 playouts at 64 machines, black K3 can win with 50.28% probability.
97 blob [5k\]: btw, any existing go bots that use machine learning out there?
98 uurtamo [-\]: yes, and they all get crushed by the monte carlo bots. which is why everyone is studying that now.
99 uurtamo [-\]: just teasing.
102 ;W[hp]WL[12601.969]C[uurtamo [-\]: there are bots that implement some kinds of machine learning, but they usually also have a very strong monte-carlo component.
103 blob [5k\]: that's good news for me
104 uurtamo [-\]: because otherwise they get absolutely crushed.
105 uurtamo [-\]: but you should ask on the computer go mailing list. there are lots of helpful ideas there and people there.
106 blob [5k\]: they probably rely on monte carlo so much because their board evaluation sucks
107 blob [5k\]: and have to play out very far
108 uurtamo [-\]: be careful with your estimate of other programs you haven't written.
110 ;B[gp]BL[12750.975]C[uurtamo [-\]: the only thing you can be sure of is what you've already written.
112 blob [5k\]: have you written a bot?
113 uurtamo [-\]: for instance, many faces of go, which is very strong, not uses some MC, but didn't before.
114 uurtamo [-\]: nope, i just started writing one.
115 uurtamo [-\]: *now uses
116 uurtamo [-\]: i've written a board evaluator, but never turned it into a bot.
117 blob [5k\]: evaluation is crucial though, so that should take you a long way
118 uurtamo [-\]: or not.
119 uurtamo [-\]: words like "should" require deep experience.
121 ;W[iq]WL[12457.681]C[blob [5k\]: whatever, either agree or don't
123 ;B[ir]BL[12647.293]C[goIngo [-\]: nick, can you give pachi %?
124 maproomad [-\]: The more people know about computer Go, the less they pontificate about it :)
126 maproomad [-\]: pachi2: In 130351958 playouts at 64 machines, black J2 can win with 51.04% probability.
128 klausP [3d?\]: i prefer black here
129 blob [5k\]: very few ppl "know" about computer go
132 ;B[gq]BL[12553.873]C[blob [5k\]: too slow for me.... good nigh
133 uurtamo [-\]: zen has a bye?
135 ;W[lq]WL[12220.093]C[maproomad [-\]: yes - Zen has a bye at last. gghideki will use the time to isntall its new hardware
137 ;B[lr]BL[12449.955]C[maproomad [-\]: pachi2: In 115171212 playouts at 64 machines, black M2 can win with 52.80% probability.
138 uurtamo [-\]: cool, new hardware.
146 ;W[go]WL[11653.706]C[uurtamo [-\]: do you know what the hardware is?
152 ;B[ce]BL[11471.174]C[GoRoGoRo [9k\]: Zen still playing in its old clothes
156 ;W[cf]WL[11098.126]C[goIngo [-\]: why to buy new ones, when the old fit?
160 ;B[cd]BL[11184.929]C[maproomad [-\]: pachi2: In 178944910 playouts at 64 machines, black C16 can win with 56.76% probability.
165 ;B[eg]BL[10983.134]C[maproomad [-\]: e13 looks like a losing ladder
166 uurtamo [-\]: everyone wants to look stylin'.
167 maproomad [-\]: pachi2: In 126328940 playouts at 64 machines, black E13 can win with 57.39% probability.
169 ;W[eh]WL[10584.286]C[uurtamo [-\]: oh, this could be a game ender.
170 uurtamo [-\]: i wonder how far into the ladder pachi2 will notice and tenuki.
172 ;B[fg]BL[10862.716]C[maproomad [-\]: pachi2: In 148112137 playouts at 64 machines, black F13 can win with 55.81% probability.
173 uurtamo [-\]: i guess not yet, then. :)
175 ;W[gg]WL[10448.687]C[ARPAnet [?\]: cool, lee sedol's ladder game :)
177 ;B[fh]BL[10727.682]C[maproomad [-\]: pachi2: In 164712506 playouts at 64 machines, black F12 can win with 51.06% probability.
180 ;B[mo]BL[10544.165]C[maproomad [-\]: pachi2: In 170439444 playouts at 64 machines, black N5 can win with 44.51% probability.
186 ;B[gh]BL[10162.79]C[maproomad [-\]: pachi2: In 82988479 playouts at 64 machines, black G12 can win with 40.76% probability.
187 goIngo [-\]: hmm, pachi in downfall?
189 ;W[hh]WL[9923.536]C[klausP [4d?\]: jup
191 ;B[gn]BL[10004.306]C[maproomad [-\]: pachi2: In 122189851 playouts at 64 machines, black G6 can win with 38.62% probability.
198 ;B[gd]BL[9484.417]C[maproomad [-\]: pachi2: In 71631949 playouts at 64 machines, black G16 can win with 38.29% probability.
201 ;B[pg]BL[9335.446]C[klausP [3d?\]: poor pachi
218 ;B[rg]BL[8198.341]C[maproomad [-\]: pachi2: In 169314007 playouts at 64 machines, black S13 can win with 42.44% probability.
221 ;B[pj]BL[8074.903]C[klausP [3d?\]: pachi fighting for a comeback, so is gomorra
222 maproomad [-\]: pachi2: In 142840643 playouts at 64 machines, black Q10 can win with 43.64% probability.
227 ;B[js]BL[7835.596]C[maproomad [-\]: pachi2: In 157538848 playouts at 64 machines, black K1 can win with 44.47% probability.
228 maproomad [-\]: creeping back towards 50%
231 ;B[mh]BL[7719.925]C[goIngo [-\]: unbelievable
235 ;W[lg]WL[7944.056]C[goIngo [-\]: pachi %, please
237 ;B[li]BL[7500.636]C[maproomad [-\]: pachi2: In 141731142 playouts at 64 machines, black M11 can win with 47.00% probability.
239 maproomad [-\]: if I don't res[pond, it's probably becasue I am not here
242 ;W[ok]WL[7830.385]C[HYamashita [3k\]: pachi ran from ladder... if StoneGrid lose, 66th G11 is maybe lose move :-)
244 ;B[pl]BL[7314.172]C[goIngo [-\]: KlausP also thought so
245 maproomad [-\]: pachi2: In 175825257 playouts at 64 machines, black Q8 can win with 48.13% probability.
250 ;B[oj]BL[7099.234]C[maproomad [-\]: pachi2: In 156331002 playouts at 64 machines, black P10 can win with 50.68% probability.
259 ;B[cl]BL[6616.812]C[maproomad [-\]: pachi2: In 87049581 playouts at 64 machines, black C8 can win with 54.53% probability.
261 ;W[po]WL[7078.102]C[goIngo [-\]: wow, turning the table
262 maproomad [-\]: seems optimistic - W has a huge top, and B's central group is weak
264 ;B[qo]BL[6432.123]C[maproomad [-\]: pachi2: In 136842147 playouts at 64 machines, black R5 can win with 54.94% probability.
309 ;B[od]BL[4276.269]C[maproomad [-\]: pachi2: In 45121745 playouts at 64 machines, black P16 can win with 81.12% probability.
310 maproomad [-\]: maybe it thinks the bottom left is dead
314 ;W[oe]WL[4731.432]C[GoRoGoRo [9k\]: isn't it?
315 GoRoGoRo [9k\]: well ... W d2 may be a start?
316 maproomad [-\]: W d2?
323 ;B[eq]BL[3931.121]C[maproomad [-\]: pachi2: In 66792603 playouts at 64 machines, black E3 can win with 83.32% probability.
326 ;B[rq]BL[3841.046]C[maproomad [-\]: pachi2: In 53617646 playouts at 64 machines, black S3 can win with 84.08% probability.
331 ;B[rr]BL[3662.657]C[maproomad [-\]: pachi2: In 53837927 playouts at 64 machines, black S2 can win with 84.37% probability.
334 ;B[hd]BL[3595.694]C[maproomad [-\]: pachi2: In 112272634 playouts at 64 machines, black H16 can win with 82.43% probability.
353 ;B[nl]BL[2960.247]C[maproomad [-\]: pachi2: In 93360296 playouts at 64 machines, black O8 can win with 85.51% probability.
354 jlg [-\]: I should also report the dynamic komi. pachi is currently giving 26 points komi
364 ;W[fm]WL[3159.724]C[Winnetou [?\]: The moves are very strange. After pachi lost the ladder the game was terrible.
366 ;B[ps]BL[2579.106]C[jlg [-\]: yes this is a known ladder bug indeed awful. But now pachi is trying to maximize the score (komi 35 points)
368 ;W[jk]WL[3088.298]C[Winnetou [?\]: It seems that it is not pachi who maximizes the win but Stonegrid who maximizes the loss ;)
370 ;B[be]BL[2503.785]C[maproomad [-\]: pachi2: In 93549867 playouts at 64 machines, black B15 can win with 89.42% probability.
372 ;W[na]WL[3018.609]C[reticent [5k\]: go seigen would either love or shake his head at the fuseki in this game
373 reticent [5k\]: not nearly strong enough to know which
375 ;B[ho]BL[2432.144]C[reticent [5k\]: and what the hell was the ladder pachi2 played out?
376 jlg [-\]: known ladder bug
377 reticent [5k\]: right
378 reticent [5k\]: anyway, looks like pachi2 spotted stonegrid fifty points and went on to win anyway
383 ;B[bc]BL[2293.487]C[jlg [-\]: 92%, komi 54
390 ;B[mk]BL[2221.186]C[salty [2d\]: i hope w isn`t planing to win that ^