Endgame: Lost Points

RobertJasiek · #1

How many points are lost by mistakes that occur during the endgame or can be avoided by endgame knowledge? I guessed that a 10 kyu (EGF) would lose more than 100 points in a typical scored game.

In the random game below and for each endgame(-like) mistake, I have stated the lost points as (a bit roughly) estimated per move values.

Total:

Black = 367 points
White = 291 points

These numbers are greater than I expected because the same local shapes can be accounted for several mistakes, as long as the players play elsewhere. Another study would be needed to count the points of each shape only once. Maybe then values would be closer to my guess, but maybe it was too low. Another reason can be that kyu games can be simplistic in that a high fraction of moves is like endgame moves.

Considering that 14 points equal one rank, even the most conservative estimate suggests that kyus of similar ranks can improve 8 or more ranks solely by studying the endgame. However, part of the endgame mistakes is caused by a combination of violating basic endgame principles and insufficient reading. IOW, to really improve ca. 10 ranks, endgame study must be accompanied by improving one's reading skill.

Conclusion: if you want to improve much, do not neglect endgame and reading!

logan · #2

Dear Robert,

I don't believe you rounded and properly valued the endgame mistakes at various parts of the game accurately enough. I've gone ahead and corrected the game record. While by your count Black was only able to improve 367/14-pt.-per-rank = 26.2 ranks to 10p and White 291/14-pt.-per-rank = 20.8 ranks to 6p, in mine they were able to improve 421p and 413p respectively.

All the best,
logan

RobertJasiek · #3

For the sake of relative simplicity, I have rounded any fractions. Each value is open for discussion because I have often not taken enough time to be as accurate as possible (it took me already hours to find the approximated guesses) and it is not already possible during the opening and middle game to get accurate losses in terms of miai values (miai value theory hardly exists then; all we know reasonably well is the early moves' 14 points, which I have used as a reference for some values of losses).

I have refrained from dividing my total values by 14 to get a number of ranks of improvement in this way because it is still an open question how to relate late opening, middle game and endgame miai values (and losses) to the game start's relation of 14 points of miai value per rank.

Your joke numbers are not helpful at all. Instead, you should appreciate my values of losses in relation to the actually made endgame(-like) mistakes. E.g., see move 66, which I have not classified as a mistake because it creates the follow-up S3 with a possible snapback connection, thus endangering the entire black group. (I would, however, suspect that the players were not aware. If all White wanted was to capture two stones, the move would have been a big mistake from his level of understanding.)

Sennahoj · #4

In what sense are moves in the very early game endgame-like? E.g. 6,9,10,12

I think it is an interesting idea to try to estimate sum of all endgame mistakes, but starting the count that early seems like estimating something else.

Krama · #5

I don't understand what you two are doing.

Do you estimate the difference in the score or does the number you write after few moves tell us the amount of points a player lost by not playing the best move?

RobertJasiek · #6

Sennahoj, endgame or endgame-like principles are also applicable during the opening:

6, 9: choose the wider gap
10 (at P4): choose the locally optimal endgame
12, 16: defend an unsettled important group (this principle applies during the opening, middle game and endgame; therefore, if one has learnt this principle for the endgame, one can apply one's endgame knowledge also during the opening)
13: attack an unsettled important group
14: this indirect defense of the unsettled important group is slightly less efficient than a direct defense
15: I have not considered this an endgame-like mistake, but it is a connection shape mistake.
17: combination of 'choose the wider gap' and 'attack an unsettled important group'

The advantage of endgame-like thinking during the opening is that understanding of the opening is often simplified.

***

Krama: logan is just joking. I assess how many points for the territory count (later: score) a move loses compared to the correct move. E.g., if a move leads to the score X and the correct move leads to the score X+L, then the mistake loses L points for the moving player. For some moves, calculation can be a bit advanced: e.g., if there is a double sente, its value is included in L together with the value increment for the correct move after the double sente sequence.

The lost values are accounted for immediately when a player loses them by making the mistake. There is no delay.

RobertJasiek · #7

Estimated values of the losses per endgame(-like) mistakes. The same local shape is counted only once per player. This allows an estimate of minimal rank improvement if the players avoid all their endgame(-like) mistakes.

Total:

Black = 237 points
White = 140 points

I do not have an explanation yet why the Black - White difference is much greater than the game score.

Judging from only this one game and using the relation 14 points = 1 rank, the players could improve this many ranks by studying endgame and reading:

Black = 17 ranks
White = 10 ranks

It is possible that the KGS ranks of these players relate to, e.g., real world 20 kyu and 13 kyu. The rank improvements would then make them ca. 3 kyu. Although this is pretty much possible, we have to recall that other games of these players could be fighting games with less relevance of the endgame (but still relevance of their reading skills). Therefore, the rough estimate of the minimal rank improvements due to the one game would be an upper bound for the rank improvement in all their games.

Sennahoj · #8

There is something dodgy with this counting methodology still. First, here's a loose description of the quantity that I think you're trying to estimate (for Black): holding White's playing strength constant, how many additional points would Black have made if all endgame-like mistakes were avoided?

This is the number that could be transformed into rank differences (if one accepts the premises of such rank<->score transformations). Then the number Black = 237 points seems very high --- do you really think that Black would have won by around 200 points against this White opponent, if all the endgame-like mistakes were avoided?

I suspect that there is still some implicit double counting, or that one side gets "credit" for mistakes that are not punished by the opponent in a biased way, and that this is the explanation for the large discrepancy between 237-140 and the actual game score difference. Will see if I can think of some concrete example...

Uberdude · #9

I know you want to sell your endgame book, but this analysis is so flawed. Nevermind the fact you are classifying opening and middle-game direction mistakes as endgame, if a player eliminates 40 points of actual endgame mistake that doesn't mean they can now beat a player 3 ranks higher. It doesn't even mean if they played the same player then the score would move 40 points in their favour (the opponent's play could change too*) but if we generously assume that what does it mean if they play a 3 rank stronger player? Maybe they will get out-read and screw up a fight and lose by 100 points.

* Let's say there is are two moves worth 5 points and another worth 15 and another worth 13. Player A plays the 5, you classify as a 10 pts mistake. B plays the other 5, another 10 pts mistake. Then they correctly get the 15 and 13. Now we suppose A reads your book and improves by 10 points of endgame, whatever that means, (B's skill remains the same) and plays the 15 first. B, who thought the 15 and 13 were miai but undervalued them, gets the 13, and then they get the 5 and 5. So A has improved by 10 points and B didn't improve but the score didn't change.

RobertJasiek · **#10**

For assessing Black's lost points at a move, White's playing strength is immaterial because it is independent. There is only the difference between Black's mistake versus correct move.

It is very much possible that implicit double counting is involved because locales overlap or dynamic changes do not reset a locale entirely.

So maybe you are right that the numbers are still somewhat too high.

Another possible error source might be my guesses for the value of attack/defense or smaller gap mistakes; I do not really understand the exact sizes of their losses yet.

RobertJasiek · **#11**

Uberdude, 42 points better endgame are worth 3 ranks in a single game, even if played against a 3 ranks stronger opponent. I agree that those 42 points might not appear in a fighting game. However, recall that improved reading is an important part of improved endgame. If a player improves his endgame reading skill, he will also have better reading in fighting; not worth 3 ranks but also worth clearly more than 0 ranks. So an actual improvement must be somewhere in between.

Your 5 - 13 - 15 points example is too contrieved because usually there are not such big value gaps from small to intermediate values. More likely, it is similar to 5 - 7 - 9 - 11 - 13 - 15.

Although everybody tries to talk small my estimates, something should be very clear: several ranks are possible by endgame study. And related study is comparatively easy because by far most kyu player's endgame-like mistakes can be identified without value calculations, using rough comparisons , if necessary, reading or other techniques.

Uberdude · **#12**

Here is some evidence your numbers are too high. I was a 2d and played the endgame of a game against a 4k. We swapped colours and did it again. The difference was about 15 points. Now of course there are problems from this like not playing optimally if you are behind or the 4k learning from my first game, but this suggests the point difference in endgame skill is about 3 points per rank, which is a lot less than 14. I would suggest more experiments of this nature to get a better idea of how the difference in endgame skill of different ranks relates to points.

quantumf · **#13**

Never mind double counting, I would suggest you're over-estimating by a factor of somewhere between 10 and 20. Your method only makes sense if every move is not only locally independent, which they obviously are not, but also globally independent, which will not be true for every move.

RobertJasiek · **#14**

Uberdude, post that game and we can assess its value of lost endgame points!

EDIT: There are problems with the description of your game: Stronger kyus and dans make of course smaller amounts of endgame losses; I have not doubted this by showing a weak kyu game example. By doing only "the endgame", you have studied only part of the game and, particular, overlooked endgame(-like) moves during opening and middle game. It is a common misconception that all the endgame moves would occur after the middle game; opening, middle game and endgame aspects occur simultaneously.

EDIT 2: It is unclear whether either of you were particularly weak or strong at the endgame for your levels at that time.

quantumf, factor 10 or 20 bears no justification. We need to study interdendencies to assess the extent of double counting.

Sennahoj · **#15**

quantumf, I think 10-20 seems like a too large factor. Simply taking Roberts numbers at face value, his predicted score difference is 237-140=97, which is only a factor 2 or so larger than the actual game.

Your point about independence it technically true I guess, but it is equally valid for miai-counting in general --- for such numbers to be interpreted as actual score gains, one must assume that there exists a very large number of independent moves of different sizes.

I'm personally not convinced of the score -> rank transformations, but I also don't find that so interesting.

quantumf · **#16**

RobertJasiek wrote:

quantumf, factor 10 or 20 bears no justification. We need to study interdendencies to assess the extent of double counting.

My estimate of the factor (10 to 20) is as justified as your estimate (1). Yes, we need to study the interdependencies, but that is impossible to do well - every move would have to be compared to every other actual move and every other better move, and also compared to all alternative follow-ups to every move.

Even to do it very badly will be extremely difficult, but we can try, I suppose. Let's pick two moves more or less at random (208 and 218). How would you assess the value of 218 if 208 had been played in the correct place? Would actual 218 even have been an option if 208 had been played correctly?

Don't get me wrong, I'm intrigued by the idea you have proposed, but I'm sceptical of how meaningful it will prove to be. My take on it is that, yes, basically every move has an endgame aspect, but players need to be ready before they play the correct (endgame) move. They need to understand how their choice impacts on the opening, middle game and endgame, and select the right move based on their overall ability - selecting the "correct" move may be catastrophic for them in the middle game if they are unable to deal correctly with the other implications of the move.

quantumf · **#17**

Sennahoj wrote:

quantumf, I think 10-20 seems like a too large factor. Simply taking Roberts numbers at face value, his predicted score difference is 237-140=97, which is only a factor 2 or so larger than the actual game.

My alternative estimate was based on the gross value of 237, not the differential. I don't believe any player would have gotten 237 extra points by playing correct endgame. I'm suggesting that perhaps 15 - 30 is more likely, but it's incredibly difficult to make this statement (or any similar statement) in the context of an actual game with lots of other non-endgame stuff also happening.

S2W · **#18**

I think Logan's joke post still has a point. It's one thing to declare a loss of x points and another to show it - let's see the variations!

Uberdude · **#19**

RobertJasiek wrote:

By doing only "the endgame", you have studied only part of the game and, particular, overlooked endgame(-like) moves during opening and middle game.

How is this an "endgame-like" move?

Click Here To Show Diagram Code: [go]$$W $$ +---------------------------------------+ $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . O . . . . . . . . . . . . . . . | $$ | . . . , . . . . . , . . . . . X . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . , . . . . . , . . . . . X . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . 1 . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . O , . . . . . , . . . . . X . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ | . . . . . . . . . . . . . . . . . . . | $$ +---------------------------------------+[/go]

Also 7 as the points value of this mistake strikes me as maybe too much. There was a series in Go World a while back about "How many points do these mistakes lose", by Ishida 'caluclator' Yoshio iirc, and I remember being struck how surprisingly small they were given there were pretty obvious dumb kyu mistakes. You probably wouldn't like its lack of rigour, but even Ishida's intuition shouldn't be too wrong.

Also you seem to give 7 points as the value of not attacking/defending a weak group. But if you play a perfectly good opening point you don't really lose points, it's more like a missed opportunity to gain more than a fair result from your opponent's mistake. And if your opponent does defend it that won't gain him many points and you get another big opening point and race on ahead. You don't have to chase a weak group to profit from it, merely getting many big moves whilst your opponent plays small moves is another way. Or what if you do attack it the next move after your opponent failed to defend? As far as the end result of the game is concerned the points are the same, but according to you I lost points. This is similar to my 5,5,15,13 example: changing the order of moves which has no net effect on the game score does have an effect on your "endgame points mistake total" score. If you were doing double entry accounting, your books wouldn't balance.

P.S. about my game, it was a long time ago I can't remember it.

Sennahoj · **#20**

quantumf: well but it's linear right, any factor you apply to the gross needs to apply to the net....

I also think that the number is too high, but disregarding double counting etc, I think the main reason that it seems so high is that Robert includes lots of moves that people wouldn't usually think of as endgame. It seems likely that a very strong player could win consistently against this White with 100+ points, that could come from attacking at moments early in the game (where Robert credited points for mistakes).

Endgame: Lost Points

Who is online