I remember an article in a Go World that tried to quantify how many points various (joseki?) mistakes made, and thought it would be interesting to apply that idea to a whole game and see if the sum of the mistakes ends up near the actual game score. For example, I think pretty much everyone above 10 kyu could identify this block of as a mistake in direction as there is no extension stone around the marked point yet, but I've seen dan players play this way. Assuming the rest of the game was played perfectly, how much would black win by? Two points, 5 points, 10 points? My feeling is maybe 3-4 points. Perhaps monte carlo bots could give some insight here, though their play is far from perfect .
To stimulate discussion, here are two games of mine, both ending in resignation but with a fairly easy to count position, the first me losing to a player around the same strength (in fact I posted this game for analysis at http://www.lifein19x19.com/forum/viewto ... f=4&t=8310) and another beating a weaker player by a wider margin. My suggestion is to go through the game, identifying mistakes from both sides and assigning a point value to them, and then sum up the totals to see the "mistake score" (I will use the convention that black's score is positive, so mistakes from black are negative and from white positive). Then actually count the position and see how close they are. Of course some(most?) mistakes are really hard to quantify and it's hard to know what the current score is, for example in the 2nd game is the lower side really my territory? If up until his invasion we count yes and white is leading 30 points in the mistake count, but then we realise black could have found a living sequence then that mistake is like 40 points but that doesn't mean the score at the end is white by 70. Also what about a mistake that goes unpunished? Should you count the mistake assuming the opponent punishes correctly, and then when they don't put the same amount back? I leave it up to you as this isn't meant to be rigourous, just some interesting fun.
Game 1:
My mistakes count:
TODO!
Final score:
Black +10
Game 2:
My mistakes count:
B9: -1, high better but this move has been played by pros a few times
B19: -4, maybe style and defence ok, but I think should c17 and allow attack (Guo juan lesson)
B25: -3, really hard to judge this trick play, didn't seem so awful for black until..
B33: -5, should q14 as shape hurt once white gets this.
B43: -1, not sure it's actually bad...
B45: -2, but I think should keep fighting as W happy to make strong shape
W48: +2, B found good answer, but I didn't know what better
B57: -3, slack, j8 push and cut.
B69: -1, first mistake that's basically just territory vs f16 (but also liberties of f15)
B79: -5, too early if gote, gotta try c3
B85: -2, just for obedience
B99: -4, I think black can answer and just about not die below
W106: (+5), I got away with it, but in retrospect h5 would be safer to patch lower side aji as there is non zero chance black has something in there
B123: -6, in the end black lost a few points with his failed invasion
I think that we can agree that a wedge by Black is urgent now. I like .
I also think that this position is simple enough that my influence estimates are pretty good. Given White to play and komi, I think that this position is about even.
Not that White did not block from the wrong side, but I expect that was a mistake. Better jump out, or possibly make a double kakari.
----
BTW, if you are going to add and subtract mistakes, be sure to use miai counting. Things do not add up with deiri counting.
This looks like the kind of mistake people make when they try to play black fuseki as white, but I think this particular error would be worth more than komi, since black has a substantial group in the corner in sente.
Re: How many points do mistakes lose?
Posted: Fri May 10, 2013 1:40 pm
by Joaz Banbeck
While I applaud the goal of this thread, I think that it is a futile exercise unless a few pros join L19. From most of us, it is a great accomplishment to even recognize that we made a mistake - without being able to quantify it.
At our level, some endgame mistakes are quantifiable, but the closer we get to the beginning, the fuzzier it gets. Measuring fuseki mistakes is for pros.
EDIT: But just to try to contribute to the thread...I think that allowing the avalanche in the lower left cost you 7.358 points, +/- 6.0.
Re: How many points do mistakes lose?
Posted: Fri May 10, 2013 1:54 pm
by Splatted
Correctly measuring fuseki mistakes may be impossible, but I don't think that means that time spent thinking about it is wasted. The task requires you to thoroughly investigate the effects of your mistakes, which is useful in itsef, and while the point values we come up with may be nonsense, it's basically the same as an abstract scale, which I think is useful. Realising that a mistake is more/less valuable than you thought can have a surprisingly large impact on your playing.
Re: How many points do mistakes lose?
Posted: Fri May 10, 2013 2:20 pm
by skydyr
Splatted wrote:Correctly measuring fuseki mistakes may be impossible, but I don't think that means that time spent thinking about it is wasted. The task requires you to thoroughly investigate the effects of your mistakes, which is useful in itsef, and while the point values we come up with may be nonsense, it's basically the same as an abstract scale, which I think is useful. Realising that a mistake is more/less valuable than you thought can have a surprisingly large impact on your playing.
I'm inclined to think that it may be better to rate mistakes on a scale from a few points to slightly disadvantageous to game-breakingly bad, rather than try to assign arbitrary values to positions with far too many continuations to count. Even killing a group outright in a corner is hard to count beyond 'huge' or something like early enough, because in addition to the value of the dead group, you have to assign a value to the influence generated by the group that did the killing.
Re: How many points do mistakes lose?
Posted: Fri May 10, 2013 3:37 pm
by tapir
Not that White did not block from the wrong side, but I expect that was a mistake. Better jump out, or possibly make a double kakari.
I understand the impulse to throw in some paradox when someone starts with strong claims, here putting values on mistakes. However, ":b3: was a mistake" is too much. Any attempt to show this for a move that was quite popular among professional players (!) in the 90s has to include some lengthy sequences that invariably start with White blocking at the other side and White approaching the upper right corner afterwards.
In my opinion a good way to estimate mistakes in the opening is by how many stones they make obsolete. With komi 6.5 an obsolete stone would be 13 points, but most demonstrations of clear mistakes merely end with badly placed stones that still do some work (losing only half a stone or less). Here no stone is completely obsolete or horribly out of place, so I would be surprised if this is more than a quarter of a stone, likely less. (So "about even" isn't untrue, still it looks like a clear mistake to me - although much smaller than common endgame mistakes.)
Re: How many points do mistakes lose?
Posted: Fri May 10, 2013 4:32 pm
by Bantari
tapir wrote:
Not that White did not block from the wrong side, but I expect that was a mistake. Better jump out, or possibly make a double kakari.
I understand the impulse to throw in some paradox when someone starts with strong claims, here putting values on mistakes. However, ":b3: was a mistake" is too much. Any attempt to show this for a move that was quite popular among professional players (!) in the 90s has to include some lengthy sequences that invariably start with White blocking at the other side and White approaching the upper right corner afterwards.
I think this is very interesting, although slightly OT.
How are these kind of considerations affected by the question of 'style'? Consider: hardly anybody today plays like, for example, Takemiya used to... does that mean his moves were weak? Or strong? Can we draw such conclusions? And if yes, how about the converse: if a pro (or many pros) play something, is it by extension a strong or even perfect move? What if they only play it a lot during a certain period of time, like 'the 90s'? Does this brings us beyond 'style' and into 'fashion'? And what does it say about the objective quality of such moves?
Or can we say that all these moves are equally strong (or equally weak), just not played anymore for whatever reason? And how do we know?
Do the pros know?
That's lotsa questions... and I have no clue.
Re: How many points do mistakes lose?
Posted: Fri May 10, 2013 10:52 pm
by daal
Joaz Banbeck wrote:While I applaud the goal of this thread, I think that it is a futile exercise unless a few pros join L19. From most of us, it is a great accomplishment to even recognize that we made a mistake - without being able to quantify it.
Exactly. And to make matters even more difficult, we're not looking for your run of the mill mistakes but rather for Future British Champion mistakes. What this exercise needs is a good ddk game. Volunteers?
How many opening moves are left?
How many are sente?
How many are miai?
White has sente in the correct diagram Black has sente in the incorrect diagram.
In perfect play who would get the last opening move? (Since the last move is worth double)
Then some calculation from these numbers would give you the blunder value.
That's how I would think to approach this problem if I had complete Go understanding. But as I am now this is way over my head.
Or you can use your endgame techniques. sente is worth full, gote is worth half but in that case the block would be wrong regardless of an extension.
Re: How many points do mistakes lose?
Posted: Sat May 11, 2013 7:04 am
by skydyr
Unusedname wrote:
Or you can use your endgame techniques. sente is worth full, gote is worth half but in that case the block would be wrong regardless of an extension.
In that case, in the sente variation black's group has the ability to expand while the gote one prettymuch seals black in, so you have to put a value to that possible expansion when you are asserting that the gote variation is inevitably worse.
Re: How many points do mistakes lose?
Posted: Sat May 11, 2013 7:37 am
by snorri
I suspect this is just too hard.
I think the original Go World series on this focused on joseki and inefficient variations that could easily be compared to specific known joseki. Also, the mistakes were either self-punishing or near-optimal punishments could be suggested. This is a lot easier to judge than general moves at any point in the game. It didn't hurt that one of the reviewers was Ishida Yoshio, either.
Also, I think when you get further into the game, maybe even a pro would have a hard time judging point values. The fact that it is an amateur game makes it worse. Just like pros have more trouble remembering the moves of weak players than strong ones, they have a harder time doing a positional judgement on amateur games. The weaker the players, the harder it is.
I remember one pro looking at an opening between two kyu players and saying it's hard to judge because so many stones from both sides were in the wrong place.
If such a project were possible, I'd suspect there would be a lot of pros who would have tried it in their reviews. They would have no objection to breaking things down like that if it was the best way to analyze. You'll see some comments like that in the late endgame in some pro reviews, where the calculation is in a more restricted position, but not just anywhere.
Also, it's not comparing individual moves---it's comparing settled results, because unsettled positions are hard to count.
Re: How many points do mistakes lose?
Posted: Sat May 11, 2013 8:04 am
by Bill Spight
snorri wrote:I suspect this is just too hard.
I think the original Go World series on this focused on joseki and inefficient variations that could easily be compared to specific known joseki. Also, the mistakes were either self-punishing or near-optimal punishments could be suggested. This is a lot easier to judge than general moves at any point in the game. It didn't hurt that one of the reviewers was Ishida Yoshio, either.
Ishida "The Computer" Yoshio!
As I recall from reading his book, "How many points is this move worth?" (IIRC), he seems to use the old value of 10 pts. for a move in isolation, a value which is surely too low, given komi of 6.5 - 7.5. When he wrote the book, komi was 5.5, so that was a reasonable estimate. Still, he was underestimating the value of some plays and positions.
As an endgame play, the pros have figured out its approximate value. That value assumes that all stones currently on the board are immortal. OC, that is not true in the opening. In the opening it is worth more, but how much more?
As an endgame play, the slide gains about 9 points. Because it affects the strength of the groups, its value should be higher in the opening.
9 pts. almost the traditional value of 10 points, so the pros of not too many years ago could simply take its value as 10 points. But now we know that a play in an empty corner is worth at least 13 points. (It could be more because of miai.) Is the slide worth that much?
Here is my guesstimate, based upon my study of influence. I estimate the value of a 4-4 stone on the empty board as 14.5, but that is probably a bit high, because it assumes that the stone is immortal. But it is close.
Black might pincer instead, but this follow-up is normal. Evaluating this position is trickier, because plainly the assumption that the White stone is immortal is wrong. To get a rough estimate, I assumed that it was half dead, and took the average of this position (assuming White immortal) and the one below (assuming White dead).
We add 1 pt. for the dead White stone in this diagram.
The average estimate for these two diagrams is 28.4. As a check, the average of these two estimates should be near the estimated value of the 4-4 stone. The average is 15.4, which is close. It is almost 1 pt. high, but that fits with the idea that after each player has played two stones in the corner, Black has gained more than 2 pts.
Anyway, our estimate for how much the slide gains is 13 pts., 4 pts. more than its value as an endgame play.
That is 1.5 pts. less than the estimate for playing in an open corner, so it seems like the slide is a second tier move for the opening. (Note that now pros often tenuki after it, which was not the case a few decades ago.)
Re: How many points do mistakes lose?
Posted: Sat May 11, 2013 3:02 pm
by Splatted
skydyr wrote:
Splatted wrote:Correctly measuring fuseki mistakes may be impossible, but I don't think that means that time spent thinking about it is wasted. The task requires you to thoroughly investigate the effects of your mistakes, which is useful in itsef, and while the point values we come up with may be nonsense, it's basically the same as an abstract scale, which I think is useful. Realising that a mistake is more/less valuable than you thought can have a surprisingly large impact on your playing.
I'm inclined to think that it may be better to rate mistakes on a scale from a few points to slightly disadvantageous to game-breakingly bad, rather than try to assign arbitrary values to positions with far too many continuations to count. Even killing a group outright in a corner is hard to count beyond 'huge' or something like early enough, because in addition to the value of the dead group, you have to assign a value to the influence generated by the group that did the killing.
I guess this is a case of each to their own, because if I set myself the task of making a vague qualitative judgement I tend to rely on my intuition and make a decision after only a cursory look, but if I require myself to quantify that judgement it encourages me to explore the issue more thouroughly, even if it's an impossible task. I can see that for anyone who doesn't need that motivation it would seem like a pointless complication though.
as a mistake in direction as there is no extension stone around the marked point yet, but I've seen dan players play this way. Assuming the rest of the game was played perfectly, how much would black win by? Two points, 5 points, 10 points? My feeling is maybe 3-4 points. Perhaps monte carlo bots could give some insight here, though their play is far from perfect 
.
. 
was a mistake. Better jump out, or possibly make a double kakari.
What this exercise needs is a good ddk game. Volunteers?
and 
from an even position.