"Snowballing your advantage" in Go?

[Knotwilg]

I think I understand the "snowballing" concept and I don't think it applies to Go.

In Chess for example, an early capture of a piece (say a knight) can be leveraged to capture another piece and so on. At the least one can start an aggressive exchange strategy so that the advantage becomes relatively bigger.

An early advantage in Go cannot be leveraged to continuously increase the advantage. There's no way of stopping the opponent from losing by the margin obtained in the beginning, except for outsmarting him in another battle. Only if the opponent wants to come back, his risk taking can be punished as far as knocking him out.

There are however some tactics that can be applied to harnass the win. One of my favorites is "lose a ko to win the game" http://senseis.xmp.net/?LoseAKoToWinTheGame. Most of these tactics are a variation on the theme of furikawari (exchange). In that sense, they resemble the exchange strategy in chess as described above.

Maybe the closest you can come to the snowballing effect is when you build strong influence, then use it to attack, the outcome being another piece of strong influence, which you again use to attack ... etcetera. It's a chain reaction of sorts but not necessarily one that increases the margin of victory.

I would even say that actively trying to snowball an initial advantage is usually a bad strategy. Top players will rather narrow down the options and accept small local losses to secure the win than trying to knock out the opponent while increasing the risk.

In fact, that's what I like so much about go. There are too many games where an initial advantage can be leveraged and Go is not one of them.

It's a nice image though and an interesting concept to explore.

[Knotwilg]

As for the alternative thread (rant) about the importance of the opening, despite the convincing mathematical arguments by Uberdude and others, I sympathize with Often. It is not that I question the statistical effect of a good opening (or komi, for that matter) on the end result. It is because the opening is nearly always the subject of such debate and not other aspects of the game which are equally or more important. Sometimes the arguments to prove the statistical significance of the opening are suggestive to the point that the opening is really very important.

For example, I would like to see the statistical impact of "not resigning": we set up a group of 100 6k players and tell them never to resign. We select 100 6k players who we advise to resign when they think the game is unsurmountably lost. And we filter out those 200 from the overall pool of 6k players. (If we don't have enough 6k, expand to a range, but that's besides the point.)

My guessestimate is that the pool who doesn't resign will on average win at least 5% more games than the pool that does and 10% more than the control group who's blissfully unaware of any experiment. This because the "resign group" will at least count the score and be aware of the chances. The "don't resign" group will win because they give themselves all chances to make the one-but-last blunder.

However, this scientific argument is never explored, not even in thought experiments, because "not resigning" is by far a less popular topic than the opening. Popularity says little about effectiveness though. I think this sort of awareness leads Often to fight a lost battle against the proponents of opening importance.

[ez4u]

As for the alternative thread (rant) about the importance of the opening, despite the convincing mathematical arguments by Uberdude and others, I sympathize with Often. It is not that I question the statistical effect of a good opening (or komi, for that matter) on the end result. It is because the opening is nearly always the subject of such debate and not other aspects of the game which are equally or more important. Sometimes the arguments to prove the statistical significance of the opening are suggestive to the point that the opening is really very important.

For example, I would like to see the statistical impact of "not resigning": we set up a group of 100 6k players and tell them never to resign. We select 100 6k players who we advise to resign when they think the game is unsurmountably lost. And we filter out those 200 from the overall pool of 6k players. (If we don't have enough 6k, expand to a range, but that's besides the point.)

My guessestimate is that the pool who doesn't resign will on average win at least 5% more games than the pool that does and 10% more than the control group who's blissfully unaware of any experiment. This because the "resign group" will at least count the score and be aware of the chances. The "don't resign" group will win because they give themselves all chances to make the one-but-last blunder.

However, this scientific argument is never explored, not even in thought experiments, because "not resigning" is by far a less popular topic than the opening. Popularity says little about effectiveness though. I think this sort of awareness leads Often to fight a lost battle against the proponents of opening importance.

While the group that does resign plays 20% more games and shares way more fun with their fellow Go players.

[Knotwilg]

While the group that does resign plays 20% more games and shares way more fun with their fellow Go players.

THAT I will not deny. If we start a new thread on how to have fun with Go, I can offer a few more suggestions from local breweries.

[Uberdude]

... because "not resigning" is by far a less popular topic than the opening. Popularity says little about effectiveness though.

Well, one of Francis Roads' (a venerable British player) favourite sayings is "no one ever won a game by resigning".

Also I agree with often that the middle-game is more important than the opening in deciding who wins a game of Go*, but to claim as he did that an opening advantage (e.g. of 10 points) has no, rather than a small (55% was my guess) statistical effect on the outcome of the game was quite frankly ludicrous. Perhaps he thought he was arguing against people who thought a 10 point lead meant you would win 99% of the games. Hence why I put some numbers on it, and raised the issue of the dependence on rank of the players and how much thinking time they have.

*But if someone wants to study the opening more because they find it interesting and fun and they value that more than increasing their winning chances that's fine by me.

And yeah I don't think Go is snowbally as Knotwilg said. Controlling the game and preserving an advantage for sure is a style of play for when you are ahead, but trying to increase it can lead to risks and counterpunches.

[Knotwilg]

Well, one of Francis Roads' (a venerable British player) favourite sayings is "no one ever won a game by resigning".

One of my favorite BGA editors!

Also I agree with often that the middle-game is more important than the opening in deciding who wins a game of Go*, but to claim as he did that an opening advantage (e.g. of 10 points) has no, rather than a small (55% was my guess) statistical effect on the outcome of the game was quite frankly ludicrous. Perhaps he thought he was arguing against people who thought a 10 point lead meant you would win 99% of the games. Hence why I put some numbers on it, and raised the issue of the dependence on rank of the players and how much thinking time they have.

You've laid it out very convincingly.

*But if someone wants to study the opening more because they find it interesting and fun and they value that more than increasing their winning chances that's fine by me.

"This" as it seems to be the hype to say online these days. However, when I suggest not resigning as a way to have more fun, it meets with less sympathy it seems

[Bill Spight]

For example, I would like to see the statistical impact of "not resigning": we set up a group of 100 6k players and tell them never to resign. We select 100 6k players who we advise to resign when they think the game is unsurmountably lost. And we filter out those 200 from the overall pool of 6k players. (If we don't have enough 6k, expand to a range, but that's besides the point.)

My guessestimate is that the pool who doesn't resign will on average win at least 5% more games than the pool that does and 10% more than the control group who's blissfully unaware of any experiment. This because the "resign group" will at least count the score and be aware of the chances. The "don't resign" group will win because they give themselves all chances to make the one-but-last blunder.

As you know, I advise DDKs to never resign. And to fill all the dame during play. There are too many chances to reverse the game at the end. By the time they reach 6 kyu, however, they have only about a 1% chance to reverse an even game at the dame stage, and they will mostly miss it. How often the never resign group can win from a position that they would otherwise resign depends on what those positions are. I don't know, but my guesstimate for reversals is at most half that of yours.

[Kirby]

Okay, I've gotta wonder, now...

Take 100 6k players and have them join arguments on L19, and take another 100 6k players, and tell them to refrain... What will the results be?

[Knotwilg]

(...) By the time they reach 6 kyu, however, they have only about a 1% chance to reverse an even game at the dame stage, and they will mostly miss it. How often the never resign group can win from a position that they would otherwise resign depends on what those positions are. I don't know, but my guesstimate for reversals is at most half that of yours.

Let me do this: I'll take a random set resigned games in the 6k-1d range and do a positional judgment
- what's the distribution of the margin
- how many games were not yet in the endgame stage

If we agree that good endgame, let alone blunders, can swing a game as many as 15 points, then I'm quite positive we'll find a high percentage of resigned games that could have been turned around. I have only anecdotal evidence but there a quite a few games where I was surprised to see my opponent resign and the ones I turned around since I've become more conservative in resigning are not negligible either.

[Bill Spight]

(...) By the time they reach 6 kyu, however, they have only about a 1% chance to reverse an even game at the dame stage, and they will mostly miss it. How often the never resign group can win from a position that they would otherwise resign depends on what those positions are. I don't know, but my guesstimate for reversals is at most half that of yours.

Let me do this: I'll take a random set resigned games in the 6k-1d range and do a positional judgment
- what's the distribution of the margin
- how many games were not yet in the endgame stage

If we agree that good endgame, let alone blunders, can swing a game as many as 15 points, then I'm quite positive we'll find a high percentage of resigned games that could have been turned around. I have only anecdotal evidence but there a quite a few games where I was surprised to see my opponent resign and the ones I turned around since I've become more conservative in resigning are not negligible either.

Oh, I think that a goodly percentage of 6k vs. 6k games that are resigned could have been turned around. If you replace the resigner by a shodan. The main reason for my lower guesstimate for reversals is that I don't think that the average 6 kyu could manage to reverse many of those resigned games.

I will say, however, that a 6 kyu who fights to the end will have a better chance of reversing than the average 6 kyu. A lot of kyu players think that the endgame is just about points.

[Kirby]

A lot of kyu players think that the endgame is just about points.

What is it about? Presumably fighting?

[Bill Spight]

A lot of kyu players think that the endgame is just about points.

What is it about? Presumably fighting?

Yes. There are quiescent endgames, but endgame skill requires fighting strength. If you lose your fighting spirit in the endgame you will miss opportunities and lose games you should not.

[Uberdude]

Hi often, in case you missed it I would be interested to hear your thoughts on the below.

Often, let's say we pick 200 random KGS 5ks. I then construct a board position after the opening in which black is around 10 points ahead (according to me or Bill or some pro or you or a monte carlo bot or whoever). We then pair up those 200 5ks and get them to play 100 games and record how many black wins. Is your position that the opening is so irrelevant that the answer will follow a binomial distribution with mean 50, in other words the same as if they just played all those 100 games from a blank board? My guessestimate is the mean would be perhaps 55-60. Probably rather similar to if 100 5ks were playing as black against 100 6ks. If we did the same experiment with 200 3ds I expect black would win perhaps a mean of 65 due to the higher skill and consistency of those dan players and their ability to maintain a lead throughout the game and the smaller sizes and frequencies of their blunders. With pros, perhaps 90+?

[Loons]

When you are behind, you should try a riskier strategy to get ahead. If you are ahead and your opponent is trying something risky, you will probably win.

[emeraldemon]

I think the idea of "10 point lead in the opening" is misleading. After the opening, I get groups and shapes that are worth something, or at least I have to try to make them worth something. Maybe with perfect evaluation I could tell you how much that weak group or big corner is worth, but in a game I just have to guess who is winning and try to make what I can with what I have.