Monday, August 04, 2008

Improve Your Chess V: Annotations versus Statistics?

The advice to study one's own games, defeats especially, is common - and the usual medium through which to do so is that of annotations. Kotov:
A considerable part in chess coaching in the USSR is played by writing notes to games. The best notes are published in magazines and newspapers, but this is not the main aim. A developing player has to write notes to his games in order to develop the habit of having a self-critical approach to his play. By spotting the flaws in his play, he will more easily eradicate them.
Similarly Kasparov in his Preface to his first book The Test of Time:
My chess philosophy has largely developed under the influence of Ex-World Champion Mikhail Moiseevich Botvinnik . . . Especially important, in my opinion, was the assimilation of Mikahil Moiseevich's main "axiom" regarding the necessity for constant analytical work, in particular the thorough analysis of one's own games. By strictly observing this rule, with the years I have come to realize distinctly that this provides the foundation for the continuous development of chess mastery.
The point is that annotating your games isn't meant to be an end in itself, but that it should function as a means for improvement. Annotating your games should mean that a "self-critical approach" becomes a "habit" whereby "flaws" are identified in order to "eradicate" them; thus there can be "continuous development".

Unfortunately, I myself have never been able to satisfactorily annotate my own games, something I will discuss further later on. But I have found that studying them using basic statistics via a spreadsheet extremely useful. Today's article, then, is about two things. Firstly, I describe my use of a results spreadsheet integrated with advice to those wishing to employ similar. Secondly, I discuss my problems with annotations with the hope of opening the subject up for debate in the comments. For clarity's sake there are two large-font subheadings to separate these two things out, and a number of other subheadings in the first section. Those not interested in reading about how to use a spreadsheet to study their own chess results might skip to the section on Annotations, which can be read independently.


I first discussed my use of statistics here and today I want to expand on that and say how my use of statistics has changed. However the basic set up is essentially till the same today. I have a spreadsheet,
and after each game I add the information required for each column. I then analyse these results for information that may or may not be hidden there. Before that though there are some basics to establish.

The Essential Columns
The following columns are essential: opponent's name and grade; the date (including round if appropriate), venue, opening and result of the game; your colour. Why? Your opponent's name is important because you may have cause to recall them and the game for future use. For instance, early this season I lost a game to a William Linton. Then looking at a match fixture ahead toward the end of the season I realised from previous team-sheets that I was likely to face on board 1 a certain Mr. Linton again. Thanks to the spreadsheet I was able to recall who he was and how the game went. I drew the second game but had chances to win it; an improvement. It's also useful to place your results in a fuller context later. How you feel about surprise wins against higher-graded players or bad results against lower-rated players might be modified a year later if you choose to look up their new grade the next season and find it's radically different.

Date and venue are both important because there may be patterns in your results linked to them. For instance if you lost all your games in January, it's a good bet you were suffering from post-Christmas rustiness. The season after you might then choose to play some practice games in the New Year, for instance, to try to resharpen. Or if your grade turns out much higher at home venues than away, you might try arriving at away venues earlier in future so you can get used to them and "settle in" before the game starts.

Result, colour and opening are all important for far more obvious reasons. You need to study how you do as black, as white, and in different openings, and if there are radical differences this will give a good indication of where your strengths and weaknesses are. For instance, if you do very well in the Sicilian with both colours, but poorly as black when you defend the Stonewall, it's a good bet you need to either focus on your understanding of closed positions, or change your repertoire against 1.d4 to something more dynamic.

The Openings Column
How you fill in the Openings column needs some thought, however. For instance, if you always reply to 1.e4 with the Sicilian and play it yourself as black, there is not much point merely recording those openings as just "The Sicilian", because this will cover 75% of the games you play in a season and so you won't be able to slice the data in a very meaningful way. You would be better off writing the variation, or ECO code, or even sub-variation. However, if you never open 1.e4 and only play 1...c5 against much weaker players, who you rarely get to face anyway, you might as well record each game of yours that you start with 1.e4 c5 as "The Sicilian" and nothing else, since it will only account for a handful of games so any further subdivision would be meaningless.

Which Games?
You should only in your spreadsheet record rated games played at a slow-play time-limit over-the-board (otb). This is partly because of the ethos of centring your training around otb-simulation, but there are other reasons, specific to the alternative forms of the game.

For instance, my last tournament was in the Surrey 125th Birthday Celebrations, a rapid-play event, and I have not recorded the results of this tournament in my spreadsheet. A description of each game should make it clear why. In the first game my opponent was sixty or so points rated lower than me and playing black, but he held his own in a slightly-off-beat Sveshnikov Sicilian amidst enormous complications to reach a position where I had a passed pawn that he was firmly blockading. By rights it was either a difficult win for me or a difficult draw for him; despite matching me blow-for-blow in the most complex parts of the game, he then blundered into a basic back-rank tactic. My second game against FM Steve Berry graded 16 points higher than me you can play through here and so judge for yourself whether or not I deserved the full-point. My third round victory versus a 182 was certainly not deserved, however, because I blundered a piece for a pawn with no real compensation and had it not been a rapid game I should have resigned - yet I kept it tense, he defended poorly, and eventually he fell apart in chronic time-trouble. My fourth round game, by contrast, was against an IM rated much higher than myself, and shortly out the opening we reached the position to the left. I have the white pieces, and it's my move, and your computer can tell you just how winning this should be - but if you've seen 17.fxe5 dxe5 18.Be3! then you probably don't need it to. I hadn't and despite winning a mere pawn with 18.Bxe5 and maintaining various advantages, I was still completely outplayed in the subsequent position and I think lost around move 30. My final round game was against someone rated forty-four points lower than me, and somewhat dejected and distracted we reached an approximately even middle-game. I then launched an unsound attack which he promptly refuted outright to reach a won endgame, where he immediately offered me a draw. I accepted, of course.

In short, the games were all odd, higgledy piggledy affairs, frequently with quite random results unrelated to large stretches of the game itself. Such is the nature of rapid-play chess, and there's not much point subjecting these games to statistical scrutiny, or polluting the pool of games played at slow-time limits, where there's a lot less randomness involved.

When to analyze?
You should analyze your results when you have enough of them to do so, is the simple message. There is no point doing so after just, say, two games. I have tended to analyze mine after around thirty games, a little less than I play in one season (a season being approximately nine months followed by a summer break.) Another factor to remember is time-span. It might be more meaningful to analyze your performance over six months than six years, or, indeed, six days.

What to analyze?

This is the crunch question. In my spreadsheet, I have a column called "gpg" which stands for "grade per game". So, for victories I have my opponent's grade plus fifty, draws their grade, losses their grade minus fifty (with appropriate adjustments made for the forty-point rule.) I then sort the spreadsheet according to different criteria to see how my average grade varies according to each subset of the data. For instance, I compare things like black with white, Sicilian versus non-Sicilian (irrespective of colour), results against stronger opponents versus results against weaker opponents, and so forth.

Generally it turns out my intuition about my results is not wholly accurate. For instance, I always enjoy playing either side of the Sicilian, but statistically it is my weakest opening with either colour (albeit not by much.) This surprising finding indicates which opening I should seek to improve most urgently, contrary to my belief that it's in fact my defences to 1.d4, which statistically have been holding up really rather well, it turns out. Who knows what similar secrets lurk in your results?

One final thing to remember is that when understanding your results, grade is not everything. There is also your win:draw:loss ratio. It is worth calculating your average score between 0 and 1, because sometimes this sheds a different light on your results. For instance, last season my average grade with white was 190, and with black 181. This seems reasonable given that black begins the game with a slight disadvantage, but my average score with black was 0.75 (13W, 3D, 3L) compared with 0.68 (9W, 8D, 2L) with white! In other words, I actually scored better with black, but I faced significantly stronger opposition when I had the white pieces.

So, the main technique is to "slice" your gpgs in various ways, to look for surprises in the averages that result. With enough games you should be able to find out basic things like which openings are your weakest, to things like whether you prefer evening league matches to weekend tournaments. You can then factor these findings back into your training and approach to individual games.

The Meaningful Column
My spreadsheet has gone through several generations of columns, and one column in particular has proved particularly difficult to get right. This is the "meaningful" column where I type in some kind of summary or statement about the game, and I really don't know what to advise about it. The first such column I had was called "At what stage was the game decided?" and I thought that if I had a preponderance of things like "Endgame" crop up against my losses or "Time scramble" against my victories then I would be able to work out my strengths and weaknesses accordingly. I did learn from this that most of my games weren't decided in the opening, but little else. Chess games aren't similar enough for them all to be decided in simple and linked ways.

The second iteration of the meaningful column saw it renamed as "Summary". I've had many problems keeping this convincing, because I state details I can't remember, use it to attribute blame, boast, and generally just blurt stuff. Here is a concise example: "I positionally outplayed him I thought (Crafty is less convinced but I think I am right) only to blunder a pawn into a lost position." Pretty useless. Compare it to this, a summary of a victory: "Move order nuance and transposition confusion, messy middlegame, he blunders/sacrifices a pawn to try escape the pressure, gets a lot of compensation (at least cheapos) but eventually I am two pawns up in a BOC endgame I contrive to almost lose in a mad time scramble." Also pretty useless, but in fact the two games were thematically similar - Kings Indian type games, where I set about undermining the queenside pawn structure in the usual fashion. This isn't something my spreadsheet tells me.

This season I've divided the meaningful column into two: "Strengths" and "Weaknesses". My hope is that if I frequently see things like "calculation" or "composure" cropping up in one column and not the other, I'll get a clearer sense of what I should work on, what not. I'm not crossing my fingers, probably because the "The Meaningful Column" is really just a poor man's version of . . .


I started this long article with advice from Kotov, Kasparov and via him Botvinnik about annotating your own games. Once I'd gotten over my own resistance to advice, I then decided annotating my own games was a good idea. I did this via a diary initially, but when that became little more than an exercise in gloating or self-laceration, as per the result of the game, I gave it up. Next I tried maintaining a file called My Chess Games, and got as far as game 1. And looking over the annotations again, I can openly admit now what I secretly felt then: I didn't do a good job, and was in some way deceiving myself.

For instance, here is the position after 11... Nc6 12. Nb3:

I annotated it as follows:
The point of 11...Nc6 is that 12.g5 can be met with 12...Nxd4 13.Qxd4 hxg5 (the order of these captures is important, it took me an age to realise, because if they are inverted with 12...hxg5 13.fxg5 Nxd4, 14.gxf6 Bxf6 15.Rxf6! wins for white) 14.fxg5 d5 and the threats of ..Qxh2+ and ..Bc5 oblige white to search for ways to bail out.

I had not realised, however, that my opponent would play the retreat 12.Nb3 automatically. Indeed it is consistent to avoid the knight exchange, thus leavning the bishop rather roleless on d7, and making sense of a4 by supporting a5. However, such subtle queenside concerns are not consistent with the shape of white's kingside pawns, as 12.Nb3 is hardly an all-out attacking move. White's inconsistency soon starts to tell, especially as he eschews g5 now entirely on move 13 and 14, allowing me time to organise my own break.
Here, I fall into two psychological traps. Firstly, there's implicit boasting about calculating a somewhat-complex sequence, and the implication I should have done it quicker ("it took me an age..."). But in fact this sort of complexity does take me time, and is probably close to the threshold of complexity that I can be confident dealing with. Secondly, the second paragraph is long-winded and over-elaborated. All that happened in this game is that white punted g3-g4 in the opening without having waited for me to castle kingside. As a result, I got to castle queenside and prize his own king open with a well-time ..g7-g5 break. However, I annotated g3-g4 earlier with a !? rather than the ? it deserved, and what I'm trying to do in the second paragraph is imply I out-thought my opponent over a number of moves based on a deeper understanding of the opening, rather than just admit the fact he muddled his move-order and I exploited it.

One final problem I have with annotations is using the computer effectively. I don't always feel confident enough to argue with it, and don't always understand what it's telling me, especially when its evaluations don't tend to settle but bob around uncertainly. Also, when staring into a position with a computer one is usually reminded of how vast and unknowable chess really is, how little we grasp of it. That the computer suffers similar problems unaware, especially to do with positional understanding, is scant consolation when you're trying to improve your chess. Nowadays, I just shrug my shoulders, make light notes in a Chessbase file, and put it on blunder-check in Fritz.

So, I know what Kotov, Kasparov and Botvinnik think about annotations. I just don't know how to do it. Any advice?


ejh said...

Yes, I suspect it's really advice for aspirant masters (and promising juniors) rather than the likes of us.

I am cautious about any advice which leads to the conclusion that one should change one's opening repetoire.

ejh said...

Meanwhile I note that my favourite, Bogdan, is playing against my favourite anti-Catalan system (5...Nc6). Who to support, the player or the line?

Tom Chivers said...

The winner!

I see Felix Jose Ynojosa got knocked over in no-time today, and there are some games getting exciting elsewhere - Mike Surtees has just sac'd the exchange against Jovanka Houska whose King looks obviously vulnerable.

ejh said...

She might be dead already, I think.

Tom Chivers said...

Interesting putting it in Crafty - Crafty initially sees no ships, then sees a draw, but then when you play the draw line, sees a win for white.

ejh said...

And so it came to pass.

Let me ask a question I already asked on John Saunders' BCM blog: if Bogdan wins it, do we think he will he have played the fewest moves of any winner?

ed g. said...

On my spreadsheet I keep track of how much time I use on my first 10 moves and first 15 moves (I use scorebooks with a column for time). I've discovered that in games where I get through the first 10 moves in under 15 minutes, my performance is more than 100 Elo points better than when I'm slower. So I've become a believer in
preparing my openings better.

As for analysis, what's worked for me in the past is: right after the game, I write down everything I can remember thinking--what I had calculated, what I did by intuition, what moves suprised me, and so on. At home, I work on the game from the end backward, looking for places where the result could have changed. After that, I check that analysis and my in-game notes with the computer. That gives me some idea of what kind of opportunities I miss and what kind of misconceptions I have.

It's a lot of work, so I've only been able to do it for a few months at a stretch, many years apart, but both times my rating went up sharply in the year afterward.

Anonymous said...

great post , thank you!