Chess theory

Ziryab, it does not translate to blitz because a minimum amount of time is required for it.  However, I did not say that it does not translate to classical chess.  I have not played OTB chess for a long while but I see no reason why it would not work in OTB.  Quite the contrary.  I suspect it would result in rapid play by the person who mastered it.

I agree. So, now, you need to work on improving the explanation of your system so that it might be translated into practice by others.

Ziryab, it would take a book to do that.  If you look at TT, I did over 22,000 puzzles.  But I did many more elsewhere.  It is best to understand the system by examples and that is many pages of text and diagrams.  I have a certain sympathy for the accusation of too many generalities here.  But I don't have the time to write it all up here and if I did I might as well write a book.  At any rate, I have given a number of clues.  When you solve a puzzle, ask yourself why you did not see the solution sooner.  I came up with many answers before I started settling on what I have got now.

FrenchDefender, I agree with that.  However, also need to improve your general level of play (and endgames).

Ziryab, TT no longer penalises for a right answer reached slowly. This translates to better training for classical OTB chess, which I think is a good thing for training thought processes.

There are no such thing as purely secret but the thing you ask for is top secret.Most of the  Strong Players know it but they do not share with people  who do not know it.

Yes, i want to help those people who are genuinly love chess and are doing research. If they want to ask any specific question ,i am ready to give them hint. If people show eagerness to know, i will open a thread to inform about this.

RobbieCoull, there is a problem with TT in that it tends to give the computer solution rather than the most instructive solution (unless things have changed in recent times).  The computer solution (from the defender's perspective) will avoid the main line if it leads to a quicker loss (play an anti-s move in my system).  In extreme cases, this leads to moves that look absolutely insane.  A human (as a defender), on the other hand, does best to choose the continuation which is the most difficult for a human to see, even if it does lead to a quicker loss.  Computers do not have that concept built into them and to that extent they are less instructional.  On the other hand, you can learn a lot from computers in how they generate moves (ideas).  The moves can look extraordinary at first but after a while you begin to see that they are a response to the position without any prejudices.

Yes, I just completed a defensive problem at the 1700 ish level that was one of those.  Really, calculating the best move to avoid perpetual was insanely difficult, and all but the most savantish humans would have to use principals (such as preferring to stay away from the edges of the board) to decide between what looked like two equal moves.  

(ChessTempo, which I have also used, has the advantage of letting you off with 'equally good' moves which avoids some of the frustration in this.)

RobbieCoull, one day I will reveal spike theory which might help you in such puzzles.

I play a lot of set positions (mostly endgame and middle game) against the computer. Sometimes the problem you describe here becomes an issue. The past few days, I been playing pawn endings from the sets at Dejascacchi. The computer runs its king to the center instead of requiring me to execute the final stage of the winning idea. OTOH, another problem where the winning idea was simple produced a Q+P vs Q that tablebases give as a win in 67 moves. I had a terrible time trying to win it against my iPad. 15-20 queens moves, and then my pawn advances one square.

Ziryab, my advice is to do what great players have done before you.  Learn the technique in technical positions.  That is what Fischer did.  That is what Botvinnik did.  If you want to reinvent the wheel, you can but it is a slower process than using existing technology to assist you.  You begin to see patterns in what first appear to be random moves.  Non-technical positions are a different story.

That's what I do eastyz.

In the position that I describe (QP vs Q) as well as Q vs R and defending R vs RB, the computer's moves will present more practical problems than most humans, certainly more than humans my level. But, in some pawn endings, the computer's play loses instructive value.

OTOH, I remember failing a problem in Chessmaster's training course because I found a faster checkmate than the programmed thematic solution. Creating instructive problems to be solved on a screen presents problems for developers. Does a solution that is two moves longer than optimal fail?

Ziryab, I know exactly what you mean.  I have seen it a number of times.  There are also middlegame positions where the computer solution is completely wrong.  They are limited in number but they come as a surprise or maybe not because the dumb beast only analyses as far as it is allowed it.  But I had something stranger still a few days ago.  It was a middlegame position from an old game from well known players.  The computer solution was wrong.  I then came across the same position some days later.  This time there was an insignificant pawn on h4.  The computer gave the same solution (this time it worked).  I am not a computer person but there are clearly progamming issues.  Then again we should not complain because there was a time when such resources were not available.  If you want to become a strong player these days, there is really no excuse.  I suspect that somebody like Fischer would have become a monster with today's resources.  I know many would say that Magnus would eat him alive.  I am not convinced.  The other day I did a list of the people helping Spassky or giving him advice for the world championship match: Botvinnik, Tal, Petrosian, Karpov, Geller, Korchnoi, Smyslov,...Still Fischer did what he complained he had to do to win: beat the soviet team.

Earth64, would your system give the correct solution after black's move 31 in Janowski v Berger, Carlsbad 1907?

Earth64,

I finally got through your three pages of explanation of your system. It didn't start to make much sense until I got to page 3.

On the positive side, I see your thinking has somewhat paralleled mine in that you felt the need to come up with terminology that describes basic unit situations and their relationships. I even borrowed your term "force."

Now, on the negative side... I was disappointed in several ways. For one, I think you badly need some examples of how to apply your system to specific positions or puzzles. Also, it looks like your algorithm is forward chaining, which I believe is the opposite direction that is best: I believe one reason humans play chess better than computers *relative to the number of moves examined* is that humans can visualize goals and positions that don't exist on the board, and can make estimations of degree of similarity, which means that humans are using essentially back chaining instead of forward chaining (http://l3d.cs.colorado.edu/courses/CSCI5582-97/lectures/winston-ch7.pdf). Also, I believe the criticisms that people posted about the system being slow due to all the necessary checks are correct. That's not a critical problem since one can convert between rule representation and pattern mapping approaches (if you want, I can give some examples of what I mean), but as presented it does make the method sound inefficient. Related to all this is that I didn't get a sense that the different tactical motifs clearly map to, or can be clearly mapped from, the concepts in your system. It would really help and would be more convincing if you showed how existing tactical concepts related to your concepts. Also, your system differs from eastyz's system in that it doesn't rely on a "database" of useful patterns that would speed up the algorithm.

It might be wise to accept eastyz's last challenge and describe how your system would solve that particular puzzle.

Sqod, I understand more or less what you say about Earth64's system except that I was not too sure what you meant by the following: "convert between rule representation and pattern mapping approaches". Can you please elaborate?  One of the hardest things I found in developing the system was looking at the board position, seeing something, and then converting it to a rule that fitted the "basic unit situations", to borrow your expression.  Every time I seemed to get stuck, I began to question the validity system and it led to modification of the system as required.  At the very basic level, chess is a simple game as the pieces are pieces, they move in given ways, and they move over a limited number of squares.  Sure, mathematically there are many moves and there are idiosyncratic moves such as castles and idiosyncratic ideas like zugzwang.  However, once one penetrates into a position that is "concrete", one sees that the likely candidates are relatively limited in number.  It then becomes a question of calculation to determine which move is preferable to the other (and a feature of the system is that it is intended to help the calculation process itself or it is a failure).  If the position is not concrete, then it is a completely different story.

Here's a somewhat abstracted example...

Suppose there exist three positional attributes of interest: A1, A2, and A3. Let's suppose that if A1 and A3 exist at the same time, then strategy S13 is called for, which is different from the other strategies, otherwise each attribute alone calls for its own unique strategy. In other words, the attributes map to strategies as follows:

A1 => S1

A2 => S2

A3 =>  S3

A1 & A3 => S13

In a rule-based system, you'd have 4 rules that almost exactly match the above notation:

if A1 then apply S1

if A2 then apply S2

if A3 then apply S3

if A1 & A3 then apply S13

You would make 4 tests, total, per each board position. In a pattern mapping system you could do that mapping in parallel so that only one operation were done, which would save tremendous calculation time when large numbers of patterns were involved. Say create a vector of 3 elements for input, and a vector of 4 elements as output:

[<A1 present> <A2 present> <A3 present>]

[<use strategy S1> <use strategy S2> <use strategy S3> <use strategy S13>

Then train an artificial neural network, parallel processor, or the equivalent to perform the following mappings:

[1 0 0] => [1 0 0 0]

[0 1 0] => [0 1 0 0]

[0 0 1] => [0 0 1 0]

[1 0 1] => [0 0 0 1]

Those arrows "=>" are all done in parallel, which takes only a single step instead of four steps. Per earlier thread conversations on mental checklists (whose links I can probably provide if you want), human brains seem to use the pattern mapping approach, which partly explains how we can recognize patterns so fast.

 [input vector] ===parallel activations===> [output vector]

Thank you Sqod. I am not an IT person.  I do agree that the human brain tends to use pattern mapping although I would not say that there is parallel processing or its equivalent (incidentally, while there is pattern mapping in my system there is also an analytical step to complement it in case the pattern mapping fails or to reinforce the conclusion reached from pattern mapping).  The human brain tends to do what the ancient Greeks (Plato, Aristotle) discovered and that is to recognise common traits and categorise responses, whether consciously or not.  After all, your input vectors in chess (and even in life) are rarely ever going to be identical.  What the brain can do is recognise that certain differences in the input vectors are likely (or in fact) immaterial.  As a simple example, you teach a beginner that where white has a rook on white's back rank and black has an unprotected back rank, all it takes for white is to move the rook down to give checkmate.  The beginner might realise that it is not necessary for white's rook to be on white's back rank - it can be on the second or third or fourth rank, etc.  Not all beginners will go through that process which will be one of the things that separates beginners with an "aptitude" for chess and those without (I used the inverted commas for a reason, but that is for another day).  The beginner who comes to the realisation that the rook can be on a number of ranks and still give checkmate could come to that realisation through a number of processess:  pure assumption without any evidence or logical thought, trial and error, luck, rationalisation, etc.  Anyway, I digress although it was these sorts of insights which led me to believe that a system could be developed.

I guess my only real point is that one shouldn't worry too much about specific algorithms, that the important thing when designing a system is to define which patterns are important. For example, I suspect in Earth64's example where the queen gets pinned against the king, the brains of chess players have learned through experience to watch for certain alignments like that, whether it's a bishop or a rook that does the pinning of a queen against a king. Such patterns can be generalized, categorized, prioritized, and put into algorithms or mappings, but the key component is the pattern itself. Unless I missed it, that is one thing that I think is missing from Earth64's approach that is prevalent in your approach (which began from exactly such pattern recognition). (Earth64  can correct me if I'm wrong about that.)

What I like about Earth64's approach is his concept of "blocks," which is roughly my concept of "sou's" (squares or units): he reconsiders the essence of the board and units from scratch. As I mentioned in one thread a few months ago, I've noted a similarity of the 15 Puzzle with chess in that if one removes the variety of forces from the units, what is left is still a complex system with tiles ("blocks"!) moving around and physically interacting with each other. I believe that relates to tactical motifs like interference and obstruction since in those motifs the forces aren't nearly as important, but more important are the physical interactions of moving things.

Isn't there a contradiction in what you say in a sense (although it probably doesn't matter).  How can you have a notion of "forces" (pieces) as against the interaction of moving things?  The latter is the enaction of the pieces.  The pieces have value because of their "force" (mobility) - aside from the king which is a special case.

Can you elaborate on sou's or blocks?  I reconsidered the essence of the board and units but it seems from a different angle to you and Earth64.