On my way to shodan and need reviews game#2

[EdLee]

jts, nothing you said supports your claim:

A player's weaknesses are most manifest in the games that he loses.

You gave one example of a game-losing, glaring, L&D mistake.
You gave another example of a broken shape that is not a game-ending mistake.

The problem is, how do you jump from these 2 examples to your general claim?

As with hyperpape and illluck, do you have the statistics data to support your claim?

[EdLee]

The claim's intent was clearly "a player's worst mistakes are _more likely_ to show up in lost games", would you dispute that statement?

Yes, I would. Please show the statistics data.

Maybe it'll help make my position more clear if I rephrase the claim as follows:

In general, a lost game magically makes our worst mistakes more clear (manifest), than a won game.

Please provide any evidence -- theoretical or statistics data -- to remove the "magic" component of the claim.

Hint 1: I know from my own experience that my worst mistakes are just as likely to be "manifest" in a won game as in a lost game.
I have seen many games where the winner's worst mistakes were glaring (even to the winner himself);
conversely, I have seen many games where the loser's worst mistakes were very subtle and unclear even after they were pointed out by a pro.
Perhaps jts, hyperpape, and illluck -- you had vastly different experiences from mine?

Hint 2: My guess is this common misconception came from the emotions toward lost games (thus, ignorance),
and not from any vigorous theoretical framework or hard statistical evidence. <--- This is my main point.

[kwhyte]

jts, nothing you said supports your claim:

A player's weaknesses are most manifest in the games that he loses.

You gave one example of a game-losing, glaring, L&D mistake.
You gave another example of a broken shape that is not a game-ending mistake.

The problem is, how do you jump from these 2 examples to your general claim?

As with hyperpape and illluck, do you have the statistics data to support your claim?

Isn't the probabilistic argument fairly clear? In any game both players lose points by playing more that are worse than perfect play - the end results (ignoring issues over whether the usual komi is correct) then depend on whose total errors are bigger. Thus, on average, you either make more mistakes in a loss than a win or your opponent makes less (or both). In the first case you either have more or bigger mistakes to look at, and in the second your opponent makes less so the consequences of your mistakes are move easily available for discussion. I think the theoretical argument is pretty clear - to argue against it one has to suggest something very odd, like that losses tend to occur in games where you actually play better than your average but make some mistake completely atypical for you that causes your loss. Surely such losses happen, but it is hard to believe that this is the norm ( it would mean, for example, that having an early advantage is an indicator that you will likely lose the game ).

As for the practical level, I think we all probably make so many mistakes that our more typical ones are there is basically any game we look at, win or loss. So I'd agree that the above theoretical advantage to looking at losses is minimal at best. However, as many people have mentioned, the practical advantage is that you have already seen the negative consequences of more of your mistakes in a loss and so are in a better position to understand advice.

I do think one needs to be a bit careful here - for example, when I was first studying seriously I thought I was great at strategy but terrible at tactics and looked to my losses where one of my groups died as evidence. As I got stronger I started to realize that I lost these fights because they were stacked against me because of some earlier strategic failure - trying to identify one mistake in an amateur game as the "reason" for the loss can lead to learning the wrong lesson.

When I've taken small group lessons from pros they usually seem to want to see losses, not wins, and I think the reason has more to do with wanting to make sure the student is actually trying to learn from the review, not just to show off their win to the other students.

[prokofiev]

The claim's intent was clearly "a player's worst mistakes are _more likely_ to show up in lost games", would you dispute that statement?

Yes, I would. Please show the statistics data.

If you'll allow that the probability of a loss given such a mistake is higher than the probability of a loss given no such mistake, then the above is a simple consequence of Bayes Theorem.

[Mef]

I think Ed has a decent point in that, for many, whether the game is won vs. lost probably has poor correlation with the number and severity of errors. I think a more reasonable stance might be that "highly skewed" games make for poor reviews. A game where you establish an early lead and can put the game on autopilot is less likely to push you and reveal true weaknesses. Likewise, a game where you fall behind early and are desperately clawing for anything is unlikely to be fruitful for review (you can identify the one blunder early on, but after that it gets hard to properly judge what the player thought was wise vs. "necessary even if it doesn't work"). Once you remove those highly skewed games you are left with a large number of close games. At lower levels just about any reasonably close game is likely to be won or lost due to carelessness in near or during endgame, however this is likely not going to be the area most teachers focus on.

[Joaz Banbeck]

@Leyleth: The whole point of this discussion is just to demonstrate to you that go players must have tenacity, among other virtues.

[daal]

The question is: Is it better to review won games or lost games. I wonder if it's the right question. Perhaps a better one is: How do we make the most of a review?

First of all, we need to have an open mind and not let our judgment get clouded by emotions. Both the pride having won and the shame of having lost can cause us to lose focus. On the other hand, such emotions can also motivate us. In this light, it seems less important whether we present a won or lost game for review than whether we present a game that we can look at objectively.

In a game of go, practically every move that we make has advantages and disadvantages. Moves in which the advantages are overwhelming are "the only move." Moves in which the disadvantages are overwhelming are "bad." As we get stronger, the ratio for each move of advantage to disadvantage diminishes. Playing a bad move or missing an only move is a "mistake," and it should be obvious that such mistakes occur in every game, including those of pros.

Mistakes are a broad field. Perhaps some of us tend to undervalue thickness. Others misjudge connectivity. Some of us do not recognize the advantage of a honte move. Others do not see the disadvantages inherent in failing to strengthen a weak group.

In an effective review, these tendencies to misjudge the merits of certain moves in particular situations are revealed to us. To a good and appropriate reviewer, these tendencies are apparent in every game, thus the criteria for choosing a game to present should not be whether or not one has won or lost, but simply whether or not one has truly tried to do their best.

[EdLee]

What Mef said.

I think we all probably make so many mistakes that our more typical ones are there is basically any game we look at, win or loss.
So I'd agree that the above theoretical advantage to looking at losses is minimal at best.

Yes.

If you'll allow that the probability of a loss given such a mistake is higher than the probability of a loss given no such mistake

I'm sorry I'm ignorant of Bayes' theorem. So may I ask why should we allow it (especially for kyu level games; see below)?
According to http://en.wikipedia.org/wiki/Bayes%27_theorem

Under this interpretation, Bayes' theorem links confidence before and after observing evidence.

And it seems we're back to observed evidence.
My observation, especially in kyu level games, is any crazy thing can happen,
at any move, in any stage of the game, won or lost.
Thus far, I have seen no evidence for jts' claim. (And jts, I most certainly meant what I said. )

[hyperpape]

Kwhyte has summarized it pretty clearly. You don't need technical concepts to understand the point--I just said "statistics and probability" because I was in a hurry, and that was the only one sentence summary I could think of. I think I mostly accept what Mef said, but I still think losses can be more profitable.

Here's my concrete example: as far as I can tell, I'm good at playing with a small-medium lead, playing patiently and reasonably to convert that into an eventual advantage. And I'm decent at endgame when I exert myself, I think. Maybe the difference is that I exert myself more than most players at my level--lots of them seem to think the endgame is a chore to be finished as fast as possible. So a common pattern is for me to take a lead and play a dull game afterwards.

I don't think I'm good at complicating a game or profiting from complicated fighting. More often than not, I just play insane garbage and end up dead everywhere (btw: this is why I disagree with people who tell kyu players not to sweat the fuseki--it can't save you in every case, but it really does make a difference for many of us). If I feel that I'm at a disadvantage after the fuseki, that seems to compound itself into a true disaster.

Now, which game will show me more of my mistakes? The game where I take an early lead and preserve it until the end of the game, or the game where I get behind and struggle? Which game will feature more mistakes? The latter, on both counts. It will show me the times that my fuseki is weak, and it will show me all the missed opportunities to get back in the game. There will be many mistakes in the other game, and I should try to review them from time to time. But I think it's clear where the bulk of my effort should go.

[prokofiev]

I agree with Mef's comment about highly skewed games likely not being great for reviews for the most part. Heavy losses will have at least some key lessons to be drawn, though, while a heavy win may not have much there.

If you'll allow that the probability of a loss given such a mistake is higher than the probability of a loss given no such mistake

I'm sorry I'm ignorant of Bayes' theorem. So may I ask why should we allow it (especially for kyu level games; see below)?

What I asked you to allow was meant to be something easier to accept than what we were discussing. Bayes Theorem then allows us to conclude from that that the probability of such a mistake in a lost game is higher than the probability of such a mistake in a won game. If you don't accept the statement I asked you to allow, then the point is moot.

According to http://en.wikipedia.org/wiki/Bayes%27_theorem

Under this interpretation, Bayes' theorem links confidence before and after observing evidence.

And it seems we're back to observed evidence.

I'm using Bayes Theorem in a more direct manner. It relates the probability of event A given event B has happened to the probability of event B given event A has happened. Here event A is a particular type of mistake being made in the game and event B is a game loss.

My observation, especially in kyu level games, is any crazy thing can happen,
at any move, in any stage of the game, won or lost.

Surely you accept that if two given 5 kyus play with komi the winning percentage of white will be higher (perhaps just slightly) than if they play without komi? This is basically all I meant by asking you to allow that the probability of a loss given a that the player makes particular type of mistake in the game is higher than the probability of a loss given no such mistake. Surely in some small percentage of the cases the game is actually decided by the advantage afforded by that mistake.

If your claim is that the difference in winning percentages is so small as to not particularly matter, perhaps you're correct.

[daniel_the_smith]

I think the statistical question is quite murky, unless you make some simplifying assumptions (which then make it meaningless).

It's trivially obvious that the loser of a game lost more points than the winner. If you count the severity of a mistake as the number of points it lost compared to perfect play, then that tells you... not much, because the loser could have made 100 1-point mistakes while the winner made 9 10-point mistakes. The average size of the loser's mistakes was much smaller, but they were more numerous.

If the mistakes of the two players have the same frequency and size distributions, then I think (with low confidence) that given a sufficient sample size, it is true that the average mistake size of the loser will be larger than the average mistake size of the winner. I'm not at all certain that the loser's largest mistake would (on average) be bigger than the winner's largest mistake-- I think that would depend on the exact frequency and size distributions. I could write a computer program to simulate this but I don't think I'm quite that curious.

And anyway, that "same distribution" requirement is almost certainly false given any two particular players.

Given all the above, I think some of you are blaspheming the holy name of Bayes. You're saying, if I read correctly, "Given that I experienced a loss, Bayes says we should expect my mistakes must have been bigger". In isolation, yes. But you're not done, you also have to run some other hypotheses through, like the one that "my mistakes must have been more numerous", and the one that, "my opponent's mistakes were fewer and/or less severe". You can't use Bayes unless your evidence distinguishes between those hypotheses, i.e., it has to actually be evidence. Without knowing the player's mistake frequency and size distributions, I don't think the fact that there was a loss favors any of those explanations.

For those of you wanting explanations of Bayes' theorem, try the below:

Video explanation: http://www.youtube.com/watch?v=HHIz-gR4xHo

Text explanation: http://yudkowsky.net/rational/bayes

Alternate (shorter) text explanation: http://hemorrhagingsanity.com/?p=155

[cyclops]

There are two types of lies: true lies and statistics
Let's do game analysis instead!
And if you prefer leave the won games to Ed and me.

[daal]

Here's my concrete example: as far as I can tell, I'm good at playing with a small-medium lead, playing patiently and reasonably to convert that into an eventual advantage. And I'm decent at endgame when I exert myself, I think. Maybe the difference is that I exert myself more than most players at my level--lots of them seem to think the endgame is a chore to be finished as fast as possible. So a common pattern is for me to take a lead and play a dull game afterwards.

I don't think I'm good at complicating a game or profiting from complicated fighting. More often than not, I just play insane garbage and end up dead everywhere (btw: this is why I disagree with people who tell kyu players not to sweat the fuseki--it can't save you in every case, but it really does make a difference for many of us). If I feel that I'm at a disadvantage after the fuseki, that seems to compound itself into a true disaster.

Now, which game will show me more of my mistakes? The game where I take an early lead and preserve it until the end of the game, or the game where I get behind and struggle? Which game will feature more mistakes? The latter, on both counts. It will show me the times that my fuseki is weak, and it will show me all the missed opportunities to get back in the game. There will be many mistakes in the other game, and I should try to review them from time to time. But I think it's clear where the bulk of my effort should go.

Well, another way to look at it would be to say: I'm particularly good at fuseki and endgame, so these are my strengths and this is where my aptitude seems to lie. Because of this, I have a good basis for improvement in these areas. You might be inclined to think that in a won game you played well enough, but why not strive to get even better. Isn't it easier to understand the fine points of something that you're already good at than to understand part of the game in which you tend to play insane garbage? Why not play to your strengths?

[Mef]

This thread has given me an idea for what might be a fun exercise - take a few game records from varying levels, cut them off right at the start of endgame....try to have people decide "Would this be a good game to review for black? Would this be a good game to review for white? Who do you think won, and by how much?"....if I get the time I may try to do this (would it be easiest as a poll question?)...

[hyperpape]

@daal That's an interesting idea. But I don't think it holds. Playing to your strengths is true or false depending on the level of focus. At the most macro level, it makes perfect sense. I am terrible at drawing. I can do it for fun with my daughter, but no amount of drawing I do will make something I'm proud of, much less money or fame.

Zoom in, and it becomes false. I'm not so good at fighting, invading or playing from behind, but how good can I get at Go without improving there? I think I'm pretty near my peak. My fuseki could be better to avoid the times when I leave the fuseki at a disadvantage, my endgame could get better, but only so far*. I will _have_ to play from behind, and I will have to kill groups, or make groups live in complicated situations. If I want to improve at this game, I have to work on those subjects.

* I'm not saying that I'm near perfect or anything. In the grand scheme of things, there are players much much better at those areas than me. But Go is often wholistic: I couldn't learn to play the endgame of a 9 dan while retaining my current fighting abilities.

** Also, I wasn't actually saying I'm good at fuseki. I'm not sure that I am. Just that my games are very skewed based on how I do in the fuseki--I think I'm better at preserving an advantage than most of my opponents and worse at compensating for a disadvantage.