Ingo Althöfer wrote:
2 x vs MoGo
Game Black wins by resignation: W4 is violating my copyright:) (The move is perfectly ok, except that a position like up to 8 is much easier to play with reversed colours.) 10 is terrible. 14 is good shape but the moyo value is a but too small because D9 is high and B LR group stable so that W moyo can be reduced easily enough even if W gets N4. 16 is ok. 17, 19 are mistakes. 30 is good. 32+34 are not the best choice. 46 is a typical timing mistake. I do not mention some obvious mistakes. Game White wins by resignation: Although 6 is a kind of move I have tried only in free handicap settings, it is pretty meaningful by setting up miai between F3 moyo construction and Q6. 8 is very consistent by maintaining the miai. 14 gets full marks: threatens F6 center moyo, erases much of B influence even after W R17 at some time soon, it assists the 3 lonely W stones, and creates good center moyo construction options in all directions. Note that 14 played one line to the right could be attacked by B from the outside on the shoulder, 14 played one line above by a boshi from below, one line to the left and the center moyo could not be constructed so easily, one line below and the center moyo when constructed by moves like F6 might become overconcentrated more easily. When following such a strategy, I would also have chosen exactly this intersection. It is tempting to play K12 instead because of the better shape relation to C11 and the better assistance to R10 but the solid three bottom stones would be too near then. Fuego should be promoted a rank:) 18 is better than J16 because then W formation would face solid N16 and B could reduce center from F15/16. 20, 22, 26 are overcentrated though. 24, 30 are premature; W must first reduce R. 25 is very good.
2 x vs ManyFaces
Game with 259 moves: 1, 3, 5 are just common-sense sum-style. 7 creates the good option to attack C6 more harshly while one corner is still open; so 7 is good. 8 helps B intention. 9 does not follow the dividing line L9-R11 intention but enters a different strategy. If W P10, then R11 has good extensions in two directions while W stones are a bit on dame. What if W R12? R13? It looks like W would get thick shape there while B retains chance for speedy global fuseki. So maybe 9 is possible. 11 continues to invite W to take what he wants. Then B has good options, too, of which taking the empty corner is not the best, IMO. IOW, sum-style at its best: Let the opponent deceive himself to have won the game by just taking all corners.:) 12 is a mistake, OC. 13: Note that W cannot play Q10! W would just falling into B's trap. 14 helps B to create thickness out of influence stones. 15 is possible but not the only choice. Some mistakes, then 21 looks too timid; better L14. Game with 263 moves: 1 and 3 are sum-style in reverse move order. 5 takes the corner in good direction because N11 has UR kakari as follow-up; then B gets good moyo on R. Alternatively B can control LR quadrant. 6 is a mistake: It helps B to use N11 well. For the same reason, 7 is a mistake. And 8. Sigh. Both miss the topic. Here you see that on average MC programs (nor ManyFaces) don't surpass 3d level in opening yet even if they are not being in the mode of playing games with stupid moves in the opening. 9, 11 go against human convention but it is hard to prove precisely that those moves are suboptimal. 13: Yes, tenuki creation was the obvious intention of 9, 11. 13 is an all-purpose move and as such ok. 14 is a mistake. 15 is bad timing. 17 was expected as what is obviously Fuego's style.
1 x vs Aya
Both programs make many mistakes, also early in the game. Especially timing mistakes like taking more influence when currently very big territory threats are active. Move 28 is a move that looks sum-style but I would not have played it because of its apparent overconcentration. It is better though that one would think at first glance: It also assists the two weaker W groups, defends a bit the O5 aji indirectly, threatens a moyo boundary to the left group and sort of creates miai of E/F4 and R17. So after some thinking, I must admit that I might have played it, too, if only I had invested some thinking before possibly choosing it if it were in my own game. Note that B's LR group is safe at this moment. *** Overall comments on Fuego opening moves: It is a bit like expert system programs in that very good moves or bad moves can occur. Like a kyu player trying to learn a style similar to sum-style and always avoiding mistakes in timing, etc. Most of the so called random moves are good to very good though in my judgement on my own style based on a reasoning approach. It does not surprise me though that the MC programs create those moves empirically because I have also used many thousands of games to calibrate my win-probability style aspect for all those fake random moves. I think MC could learn to play consistent high dan sum-style opening quality if the empirical evaluation were not based on winning the game but on winning dynamically updated sets of opening aims. Thereby more calculation power could be concentrated within the opening, similar to human opening thinking. I.e., early game phase analysis would become thicker. Needless to say, choosing good sets of opening aims, which then can be "won" quite like winning a scoring game, for MC tuning will be the difficulty. Obviously, we do not need Cazenave's aim proof search yet but something very rough with rough aims on a high abstraction level. Let the MC programs think about sum-style opening as if it were Nine Men's Morris, i.e., the structure of opening development suffices, maybe then combined with Local Move Selections and ladder conditions. *** BTW, I don't know. Who is Fuego's author? -- robert jasiek _______________________________________________ computer-go mailing list [email protected] http://www.computer-go.org/mailman/listinfo/computer-go/
