Monday, August 06, 2012

E-mail from a real chess player






I got an e-mail from a “real chessplayer” from the Ukraine. His name is Alex. I don't know his rating. To me a real chess player has at least a rating of 2200. I hope Alex will enlighten me one day.


I “translated” his e-mail, so the bad english is mine. Furthermore I “clarified” the text with my own assumptions, which might be wrong. I hope Alex will correct me when this is the case.
Sometimes I feel inclined to comment on his writings, since shutting up is not an automatic habit of mine. Translation and clarification are in blue, comments in black.

Recently in the last years because of computers the methods have dramatically changed.
Specially on GGMM-level.

I don't know what GGMM is, but it sounds even more impressive than GM.

It's based in the theory of managing advantages, it said basically that the main difference between players is the amount of advantage they could handle succesfully, securely and safely to the next step of advantage.
A novice will need  7.00 pawn units or a lot more (the pawn units in this text are based on standard Houdini2.0/year 2012). Houdini2.0 might need 0.35-0.40 pawn units to almost surely win against himself and a disadvantage of -0.10 to be sure to not loose a game against other strong UCI chess engines.
Anand might need 0.90 or a bit less to manage.
An 2800 player could handle 1.00 at some stages.1.30 is the secure advantage they could handle at any time. No 2800 player can handle more than 0.60 to 0.70.
Now imagine others levels, what a 2600 GM could handle, what a 2500 etc..  (2 decimals is too much margin and its just expressed as example,it need a lot more complicated explanation but this is the basis). This is a fact.
Ukrainian (grand)masters (= my interpretation of "they", I might be wrong, though) work with posititions that are based on how much pawn units the can handle safely. Such positions are handled in a special software program developed by an ukrainian developer.
A player select his main opening lines, a set of final positions he could reasonably reach in his opening during real tournament games (his set contains the variations of a specific opening). Then the computer create a database with games generated by chess engines which play against each other. A chess engine simulates a human model and plays thousands of games vs another chess engine simulating a human model and put them in a database.
From that database is determined the overal percentage of well known standard typical :"positions and situations" wich is mainly divided in positions with queen presence and without queen (queens dramatically influence the tactic and strategies in a chessgame). Further there are positions with a well classic middle game center pawn, derivatives with other types of centers, opposite castling, etc..
All very typical position that are derivates from each other and that are related with each other create a red network of 60 knots or even more, with situations like" the attack that stopped". Another:"pawn down compensated"  ," "a draw no one could stop" etc..
The situations are like a poem to be remembered, the positions are clear like a math table. This is done to organize the knowledge the chess player is accumulating into categories ( its how our brain works, unconsciously remembering sets of thousand typical positions and situations present in daily chessgames. The opposite is chess960 fisher random or composed problems, despite the fact that these also make use of the chess rules, it wouldn't help as a training knowledge background)

The new acquired knowledge is always connected and related and basically 99% of what they (it is still not clear who "they" are, the users of the software, at least) are learning from computers is how they are managing to materialise such advantages. The time cycles are administrated carefully, watching the percentage  of such position in the database (for a computer an almost equal endgame is not viewed as a goal, instead it's a fault in materialising advantages in the middlegames, do not administrating 30% of the training time to endgames that only represent 8% of overal games and so on).
They never analise human games (only for practical purpose,openings and so on).
All of this is derivated from openings they use to play.
The goal is to go deep in understanding the complexity, not going wide triyng to achieve many positions.
They do not view tactical training as a set of well known positions that contain an almost forced sure win or a combination to win as in real games such rich positions are rare. It is disorientating for a player to "fish off" a forced combination in his real games, since it is far from reality.
Tactic and strategy are viewed at differently, since those concepts are united.
Any position that has about 3 alternatives with about 0.08 pawn units in difference is considered a strategic position, were no forced tactical blow is near and choosing between the alternatives is hard only with regard to strategic fundaments (contrary to the general view that strategic moves are always in quite waters).
A tactical position only differs 0.25 pu. or more.
So basically a normal game is a set of short tactic solutions( as humans we can only see arrays of short tactic solutions in chess) were strategic solutions are "the tactical difficult ones" to be clarified.

When such positions are set up, they play two games vs the main program (then back I remember rybka2.3  was the top) one of 25 minutes and another of 35 minutes (they never play fast games, and paradoxically never classic 2 hours games because there is too much time the mind is idle and resulting in an unproductive training session ). During their play they are trying to materialise such exact amount of advantage given. The computer is playing 6 different games in the background from the same position but with different programs lowering in level (in order to later show how the advantage could have been materialised from different perspectives in different situations vs different levels of resistance).
Every session is very emotional intense lived ( as chess was, is, and will be a challenge, not a nursing course) and emotions push forward the mind and the memory and the association and the analysis, and the knowledge saved into the brain will be long lasting.
Later such questions are always a rule:
What was the common point in all these different solutions?
Why have these almost similar positions a different score in pawn units?
Looking for the common in the different positions and the particular differences in the almost common, it is applied to everything he saw this day.
Later such questions are for others games and other days ,the computer just asks, and it is not comparing results, it is just an open task the player should do. The player is always flooded by tasks as long as possible, but keeping the motivated tone of challenge as long as it could.

Such sessions are later expanded in a general knowledge background, and everything that passed on the screen of the pc is saved, all his games, and all the moves that overpassed the others moves in 0.30 are also saved (later such key moves are showed, time and again, confronted with new knowledge and compared even months and years later).

The training sessions are in cycles, and there is a week were the player is analyzing nothing, or playing a single game, he just sits back and the entire week he is remembering the solutions he used to play (automatically showed by the computer), the responses, the games, the days, it happened the comparision with now days, it creates a sensation of special pleasure while remembering, and apparently the mind is idle, but it is really in such momemts that the training is getting high effects, and then all back to the same.

They never play a game vs computers from the start, since they couldn't handle an advantage needed to training and barely they could reach a 0.40 or 0.50 and hold it for some time.
Maybe humans could reach a level of 0.80 within some generations, perhaps in 25 years, who knows, but 0.60 is far out of reach of humans and equals to rybka or houdini: mission impossible.
Kasparov used that system and hired some qualified players for the task of carefully and profesionally classifying the positions ,and its looks like it's based on Davidov theory books of pedagogical science and psychology of soviet times, the developers are all ukrainians.
The software is entirely built in russian, it has many other features, like a transposition option to turn black pieces into whites  instantly to avoid the "Semeon Furman effect: world champion with white". It turns the board upside down entirely with all comments and turns the same pieces but on different squares. (I'm sure you get the idea :)
Soviet masters improved very well very fast because they used a trick the burocratic system allowed them to do. Every master was allowed to have a fixed income when he was assigned to schools, institutes, universities etc.. Everywhere is was possible to get a teacher job. They dreamed about national championship qualifications. This allowed them foreign trips to West Europe countries and a big appartment. So they decided to train themselves instead of training someone else, but the process was mutual benefical, their training was like this: day and night constantly they organized a handicap game vs their students. A master sitting in a corner playing a game as serious as a tournament game, students in another corner consulting and moving the pieces trying to defeat him, the result: playing vs rybka4 since 1950!
Just as simple as that, Petrosian kept his power, Beliavsky and many other that "copied that way of life" that the Soviet system allowed. Look at Capablanca in 1938 his 4 games vs 4 chess club players in consult with chess players from Havana, and two from Santiago de Cuba provinces, against 11, 12,  or 13 players. Those players played like Deep Junior. Capablanca could only survive hardly just in one game with draw, loosing the others.


6 comments:

  1. This is all very interesting and important. I hope that there will come one day that I don't mess up move orders during calculating tactics. Then I will be ripe to follow a training regimen like this.

    ReplyDelete
  2. This is a very interesting technique for improving...well, the best word for it is technique. It is a much more systematic and thorough version of the more common practice of taking a theoretically winning position and playing it out against a computer.

    That said, it does not seem to be comprehensive, since it does not address the necessary part that comes before the winning advantage, which is how the player achieves this advantage. Coming out of whatever opening is normally played at the professional level, no one should have more than (say) a 0.30 advantage in terms of computer evaluation; the most common evaluations I've seen are 0.25 and below.

    In other words, this seems like an outstanding training mechanism for the later part of the middlegame, when a significant advantage is already achieved by one side, but not the early middlegame, when the players are just emerging from theory.

    ReplyDelete
  3. @ChessAdmin,
    If chess is a draw, you can only make progress by maintaining what you have and make use of the mistakes of your opponent. You can't force those mistakes, allthough you can encourage them.

    ReplyDelete
  4. It reminds me what Magnus said in an interview when he was asked how deep (how many moves in advance) he can calculate.
    His answer was approximately like this: "Surely many moves if neccessary. But it is not so important how deep you can calculate. The real 'trick' lies in the evaluation of the resulting position."

    The evaluation of a position is pretty difficult for me.
    For instance: I have the pair of bishops.
    Is this +0.3 or +0.5?

    Or I have a double pawn. Am I down -0.2 or -0.6?!

    Obviously it is dependend on the situation. But even if it was not - I would not even be able to get to a good evaluation if a double pawn had a fixed negative value.
    If I weigh up all details of a positon I would hardly get to a good score.
    I am glad that I am not the only guy suffering from this, but the real chess player here says, that most people would not be able to take advantage from +0.6 pawn units.

    I experimented a little bit with evaluations. I remember one extreme where I judged an outcome of the opening in the grunfeld defense as rather bad as black, probably lost. Reason: I had little counterplay and was an exchange down.
    Rybka 4.0 however told me different. Rybka judged it +1.1 for black DESPITE being material down.
    I took then the white pieces and soon had to agree, that the white position was indeed probably lost. No matter how often I took back a move, I could not dig white out of the misery.

    (This was by the way a good training. I understood this variatiin afterwards much better than before).

    ReplyDelete
  5. In the capablanca 4 games vs advanced players,i said clearly eco codes a11,a12,a13,..no 11.12,13 players :)

    ReplyDelete
  6. GGMM are focusing a lot on openings because their level is very similar,so openings is the only place were they can "make the difference" ,Dedicattion in openings pay the price very well,and they know it,but "training openings " is complicated,since the openings is the hardest things,as no one can hope to gain morethan 0.20 --0.25 ,"understaining openings" is not a training per se,

    ReplyDelete