Life In 19x19

I think it's a generally agreed on rule of thumb to say that if a move is big (or small) for your opponent it is going to be big (or small) for you (assuming you still can play there when it's your turn, because he or she opts to play elsewhere), and vice versa.

But when is that NOT the case, if ever? And if such counterexamples can be given, is there a clear pattern to them (i.e. strictly definable conditions under which said rule of thumb does not apply)?

I am asking this because I'm programming a go playing bot, and am looking at keeping a list of move candidates ordered by their approximate "size". If I find the above rule of thumb to be stronger than I expected on first glance, I might opt to share the list between both players. If not, I'd better have separate lists for each player, despite resulting in slower lookup.

The only examples I can think of are when the move is gote for you and sente for your opponent.

Click Here To Show Diagram Code

[go]$$ Classic yose example
$$ . O X . . . |
$$ . O X . X . |
$$ . O O X . . |
$$ . . O X . . |
$$ . . . a . . |
$$ -----------[/go]

The move 'a' is bigger for white than for black - assuming black doesn't need it to live. Black should try to find a sente move elsewhere. White will pretty much get that move for free.

There does eventually come a time where it is better for black to play there just to prevent white from having it in sente, but the move will likely be worth more for white to play it than for black to defend it.

moyoaji wrote:The move 'a' is bigger for white than for black

Thanks. That's an interesting case, and I definitely think that the interplay of sente and gote is key here. However, I'm not entirely convinced that move a is any smaller for black than it is for white. Not in general anyway. It think it would at least have to depend on the rest of the board state.

Could you perhaps give an example of a embedded in a full board? That might help to take away some ambiguities at least.

If I have my opponent trapped in a ladder that runs to the edge of the board, and he plays a ladder breaker, it may be a big move for him. But if I play it, it is redundant.

Big for him; not big for me.

Ladder breakers! Yes, nice counterexample; thanks.

Now I'm thinking this is going to be very hard to describe in general purpose terms (without summing over a bunch special cases such as "ladder scenario X", "yose hane Y", etc)...

Frequently, playing in the same area is big for both players, but often the exact move each player wants to play is different.

There's a chance I'm making this up and that this didn't actually happen, but I seem to recall some anecdote from GnuGo development a long time ago where someone tried to implement the "my opponent's best move is my best move" heuristic in a fairly naive way. Among other things, it caused the bot to play moves like the following:

Click Here To Show Diagram Code

[go]$$Bcm1 Backwards monkey jump
$$ | . . X . . . . .
$$ | . . . O . . . .
$$ | . . X O . . . .
$$ | . . X O . . . .
$$ | . . X O . . O .
$$ | . . X X O . . .
$$ | . . . . O . . .
$$ | . 1 . . . . . .
$$ ---------------[/go]

Okay, lightvector, you win. I think that's the clearest and most hilarious example so far.

[EDIT] Related follow-up question: Assume it's your turn, and you have a weak group that can be killed by your opponent unless you save it by playing move M (or any move in the set M of saving moves), but struck by a sudden bout of idiocy, you pass (or tenuki). Now your opponent plays move M in your stead. Is this move M (or any move in said set M of otherwise saving moves) guaranteed to kill your group?

Again, counterexamples much appreciated.

xor wrote:Okay, lightvector, you win. I think that's the clearest and most hilarious example so far.

[EDIT] Related follow-up question: Assume it's your turn, and you have a weak group that can be killed by your opponent unless you save it by playing move M (or any move in the set M of saving moves), but struck by a sudden bout of idiocy, you pass (or tenuki). Now your opponent plays move M in your stead. Is this move M (or any move in said set M of otherwise saving moves) guaranteed to kill your group?

Again, counterexamples much appreciated.

I think it depends a lot on the problem, here. Many tsumegos can be solved by just switching sides and finding where the opponent may strike/defend more effectively, but the more complex the tsumego, the harder it is to pinpoint a specific case. Also, will depend a lot on the surroundings. Not all moves are made equal, so a move in the set M by the opponent may leave some exploitable aji to break through. I'm thinking of middle-game invasion scenarios, especially.

xor wrote:Okay, lightvector, you win. I think that's the clearest and most hilarious example so far.

[EDIT] Related follow-up question: Assume it's your turn, and you have a weak group that can be killed by your opponent unless you save it by playing move M (or any move in the set M of saving moves), but struck by a sudden bout of idiocy, you pass (or tenuki). Now your opponent plays move M in your stead. Is this move M (or any move in said set M of otherwise saving moves) guaranteed to kill your group?

Again, counterexamples much appreciated.

Sorry, go is not that simple

Easiest case, a semeai where the last liberty is not shared

Click Here To Show Diagram Code

[go]$$Bcm1
$$ | . . . . . . . .
$$ | . . . . . . . .
$$ | X X X . . . . .
$$ | X O X O O . . .
$$ | . O O X O . . .
$$ | O O X X O O O .
$$ | O X O X X X O .
$$ | . X O . X . O .
$$ ---------------[/go]

xor wrote:Okay, lightvector, you win. I think that's the clearest and most hilarious example so far.

[EDIT] Related follow-up question: Assume it's your turn, and you have a weak group that can be killed by your opponent unless you save it by playing move M (or any move in the set M of saving moves), but struck by a sudden bout of idiocy, you pass (or tenuki). Now your opponent plays move M in your stead. Is this move M (or any move in said set M of otherwise saving moves) guaranteed to kill your group?

Again, counterexamples much appreciated.

Click Here To Show Diagram Code

[go]$$B
$$ +----------------
$$ | . X O . . O . .
$$ | . . . X X O . .
$$ | X X X O O O . .
$$ | O O O O . . . .
$$ | . . . . . . . .[/go]

xor wrote:I think it's a generally agreed on rule of thumb...

TIM82 wrote:Sorry, go is not that simple

xor, I agree with Tim. In my (very limited) experience so far, I've found no such thing.

But good luck, and please post your progress and discoveries here.

As has already been shown, exceptions to the heuristic that The opponent's play is my play abound. But even when it is true and the best play by either side is on the same point, that does not mean that each play gains the same (as moyoaji indicates ).







For gote, the plays gain the same for each side. For sente, the sente gains more for the sente players. For ambiguous positions, one side may have an advantage.

xor wrote:I am asking this because I'm programming a go playing bot, and am looking at keeping a list of move candidates ordered by their approximate "size". If I find the above rule of thumb to be stronger than I expected on first glance, I might opt to share the list between both players. If not, I'd better have separate lists for each player, despite resulting in slower lookup.

See http://computer-go.org/pipermail/computer-go/

Good luck!

A rule more generally applicable might be 'a move that prevents a good move by my opponent is a good move'. The monkey jump block case can be seen as transforming the valuable monkey jump into merely a point losing ko threat.

The redundant ladder breaker is bad because it prevents only a small number of good continuations your opponent has. Hence we tend to play our ladder completers close to the stone, or capture it outright, simultaneously devaluing any space further out on the ladder path for our opponent. It may still be premature to capture the stone if the ladder is yours anyways, but it's definitely still worth considering, to prevent your opponent from getting a two purpose move in.

Filling in your own liberties in a semai race is just bad, as although you remove a good move for your opponent, you replace it with an every better move for him.

The sente-gote cases are interesting in their own right, because a reverse-sente move is more valuable than a gote-gote move. There's just no guarantee that big enough for your opponent to play is also big enough for you to stop.

So rather than looking at the biggest moves as a shared list between players, you might want to use it to optimize where the AI spends its time looking. A move that removes the best response for the opponent is worth considering at least.

It might be a better idea if you replace "move" by "strategic point" or "strategic idea".

It still fails in some cases (e.g. sente moves for one player vs gote for the other) but most ideas here would still apply: if it's a good time for your opponent to capture a ladder, it's a good time for you to play a ladder breaker; if it's a good time for him to capture then it's a good time for you to live; if it's a good time for him to extend on one side it might be a good time for you to extend on the same side as well...