Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / The Secret of Chess
1 2 3 4 5 Previous Next  
Parent - - By Banned for Life (Gold) Date 2017-11-18 22:14
Robert,

I think the English is good enough. Probably much better than your Bulgarian and certainly much better than mine...

I would disagree with the comments regarding Nunn and Polgar unless the audience of the book is top tier chess players (I don't think it is). Those examples are of people who have spent a good part of their life working on their pattern recognition skills, and this is probably a better method than the rule based approach to evaluation provided by the author (and used by some good players like Larry Kaufmann who by his own admission has pattern recognition skills that are just terrible). But for the great majority of players of intermediate players, solid evaluation rules are probably very useful, and criticism from top players of this type of approach just isn't helpful.
Parent - - By Dr.X (Gold) Date 2017-11-18 22:55 Edited 2017-11-18 23:03
His exampled positions were " not " complex. Perhaps they weren't meant to be. I was more interested in his scoring of those positions.

Addendum:

I'm no longer good at multi-tasking: :roll:
Parent - By Banned for Life (Gold) Date 2017-11-19 00:11
In the excerpts, he focused on one feature at a time, which has advantages and disadvantages. In real life, you will certainly have to decide which features are important, and which are not...

I was never good at multitasking! :razz:
Parent - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-19 12:02

>His exampled positions were " not " complex. Perhaps they weren't meant to be. I was more interested in his scoring of those positions.


>Addendum:


>I'm no longer good at multi-tasking


The example positions are not complex, because they are meant to simply illustrate the specific
evaluation term.

I could add such positions relating to those terms, that your head will start swirling.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-19 11:59

>Robert,


>I think the English is good enough. Probably much better than your Bulgarian and certainly much better than mine...


>I would disagree with the comments regarding Nunn and Polgar unless the audience of the book is top tier chess players (I don't think it is). Those examples are of people who have spent a good part of their life working on >their pattern recognition skills, and this is probably a better method than the rule based approach to evaluation provided by the author (and used by some good players like Larry Kaufmann who by his own admission has >pattern recognition skills that are just terrible). But for the great majority of players of intermediate players, solid evaluation rules are probably very useful, and criticism from top players of this type of approach just isn't >helpful.


Hi BFL, thanks for your support.

What is the difference between rules and patterns?

My book is following the pattern approach.
Parent - - By Banned for Life (Gold) Date 2017-11-19 12:46
When you look at the output of even a relatively small neural net that can be run on a computer, it's generally hard to understand the intermediate results that contribute to the final outputs. The best chess players generally have finely honed pattern recognition skills when it comes to evaluation that work well (significantly better than those in an engine with similar search depth to a skilled human chess player). Unfortunately, there is no reason to believe that great chess players will be able to explain the intermediate steps or features and weights which make up their evaluation of a position.

In the case of the excerpts, these are patterns, yes, but simple patterns that are suitable for converting to rules that can be used by an engine or a person to improve their play. In contrast, if you gave S. Polgar ten years to write down rules for playing good chess, you would almost certainly end up with patzer level play if you actually followed them...
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-20 00:41

>Unfortunately, there is no reason to believe that great chess players will be able to explain the intermediate steps or features and weights which make up their evaluation of a position.


Why so?

Most are able to do that, though it is true that even the top rely more on routine knowledge
involving pattern recognition that they are able to define as good or bad without actually knowing
which patterns constitute the building blocks.

Most of them simply have not thought in that direction, they just need precise calculation
more than that.
Parent - By Banned for Life (Gold) Date 2017-11-20 06:17
Why so?

Neural networks are renown for not providing clues as to what is being done behind the scenes. Using a well known example, a quick glimpse of a never before seen room, even from a photograph, allows a person to instantly identify the objects and their locations in the room. This feat, which young children do without even thinking, can't currently be replicated by machine...

Chess uses much simpler patterns, but the concept is the same...
Parent - - By Dr.X (Gold) Date 2017-11-25 06:43
It is interesting that you mention here ' pattern recognition ' which combined with rule based approach is not usually undertaken in most chess software training programs! With the exception of one that I know of -Stefan Meyer-Kahlen Chess Tutor program. He places pattern recognition as essential to master an understanding of the board. You don't progress to the next step if you get one position wrong -may as well just get the concept under your belt and move on. In the beginning- it quite frankly drove me nuts, because I'm a perfectionist. Getting one wrong and I ended up starting all over again-and that is when you really find out if you just memorized the position or learned the concepts.

This was a few year ago. There was someone else on the forum who comment on Chess Tutor -I forget who it was. I do wish Stefan had developed his program further.
Parent - By Scott (****) Date 2017-11-26 00:36

> It quite frankly drove me nuts.


Always wondered what it was...thanks for clearing that up! :smile:
Parent - - By Banned for Life (Gold) Date 2017-11-29 06:01
Most great players rely on very highly tuned pattern recognition skills. But these can't easily be translated for use by a chess program. Larry Kaufmann has, by his own admission, very poor pattern recognition skills. The example he gave once was that when he got up from the chessboard to use the restroom, he wouldn't recognize his opponent when he returned...

Larry's poor pattern recognition skills are an advantage for his work on chess engines because they force him to rely on a rule based approach to evaluation which is much simpler to code...
Parent - - By Dr.X (Gold) Date 2017-11-29 07:15 Edited 2017-11-29 07:19

> The example he gave once was that when he got up from the chessboard to use the restroom, he wouldn't recognize his opponent when he returned...


That! Is a very strange example. However, not so strange for someone who is more attentive to what is going on problematically on the board than the person moving the pieces creating the problem for him.

I'd rather not call it " pattern recognition " rather I'd reference it as understanding positional board imbalances. Which, I might add infers a rules based understanding!

addendum:

Although, I think you are coming to this from a totally different perspective.
Parent - - By Banned for Life (Gold) Date 2017-11-29 07:43
Facial recognition is actually something that most people are really good at, and it's been demonstrated that people can recognize famous faces, e.g. George Washington, with only 16 gray scale pixels!

Larry is deficient in this area, and this has forced him to rely on evaluation based on more rule based formulations, which can be coded. It isn't obvious to many people, but things that seem like very simple concepts can frequently be incredibly difficult to code...
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-29 18:29

>Facial recognition is actually something that most people are really good at


I can never recognise 2 black people from one another, unless they are of significantly different size and
body structure.
Parent - - By Banned for Life (Gold) Date 2017-11-29 18:53
Well, I'm guessing that means that you aren't interfacing with very many black people in Bulgaria... Try living in East Cleveland for a while, and your facial pattern recognition skills will likely improve significantly in this area! :lol:
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-30 17:47

>Well, I'm guessing that means that you aren't interfacing with very many black people in Bulgaria... Try living in East Cleveland for a while, and your facial pattern recognition skills will likely improve significantly in this area!


Give me all those black men. :smile:
Parent - By Dr.X (Gold) Date 2017-11-30 19:44
Well, it didn't take much for this  thread  to spiral down and get delimited to ethnocentric terms of class-consciousness. :twisted:
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-29 18:26

>I'd rather not call it " pattern recognition " rather I'd reference it as understanding positional board imbalances.


What is the difference between 'pattern recognition', 'rules based approach' and 'imbalances'?
I guess we are talking about one and the same thing.

The problem is not HOW you call it, but HOW MANY of those patterns/imbalances/rules a human/an engine considers,
and how accurate they are.
For example, all modern top chess engines have significantly more evaluation than top engines from, say, 10 years ago.
So, in order to get better, you should consider more, including patterns no one has thought of.
Parent - - By Banned for Life (Gold) Date 2017-11-29 18:51
There's definitely a difference between rule based processing and pattern recognition processing. In fact, they are performed in different parts of the human brain.

But go back to the example I gave. Most Americans can identify a picture of George Washington based on 16 grey scale boxes. Try to write a rule for doing that...
Parent - By Dr.X (Gold) Date 2017-11-29 23:56

> But go back to the example I gave. Most Americans can identify a picture of George Washington based on 16 grey scale boxes. Try to write a rule for doing that...


You're coming from a different perspective.
Parent - By Dr.X (Gold) Date 2017-11-29 19:03

> The problem is not HOW you call it, but HOW MANY of those patterns/imbalances/rules a human/an engine considers,
> and how accurate they are.


No doubt about that! I might be able to compete to "some" extent with a chess engines in the opening in that it will  follow some level of statically acceptable book moves. Once it leaves book, I'm on my own and calculating against a chess engine isn't fun.

Addendum:

Unless it is Crafty! :grin:
Parent - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-19 11:54

>Hi, Lyudmil, let me first state that I like the overall idea you tackled and the apparent energy you put into putting this into presentation. One of the major problems isn't the "idea" it is your attempt to translate that idea >into English. I haven't read the book just a smattering of the excerpt . Also I would have like to see more of your ideas on piece values as related to the transitioning position presented -that is,  why those values changed. >That is just my humble opinion.
>Of course, when one gives their  inferences of understand the strategic and tactical elements of a given position you will be suggesting to the reader your own level of understanding of the game of chess!
>There is where the sand separates from sea, when an International Master like John Nunn or , Susan Polgar, who in one of their books is giving instruction and someone like myself can only  attempt to report back from a >limited understanding what I've gleaned.


I apologise in advance for the following comment, but it is true: Nunn and Polgar(Judith), more so Nunn,
because he is a weaker player, will have very hard time understanding some of my concepts, simply
because they are higher level and more precise.

I apologise again, but that is true.

At some point in time, this truth will become aparent to everyone.

That is what concerns the chess knowledge base.

I agree with you, though, the style of the book is not perfect and the concepts could have been presented in a much
better way for learners.
Parent - - By Banned for Life (Gold) Date 2017-11-18 22:07
"Right, but more than 20 of my evaluation ideas have been incorporated into Stockfish code, so they have passed their verification.(please note, that it is
extremely difficult nowadays for an evaluation idea to succeed in Stockfish, the likelihood is something like 1%)"


Note that passing verification in Stockfish means not only that the evaluation idea is good, but also that it can be applied fast enough to justify its benefits, i.e. if you have a great evaluation idea but it slows the number of nodes per second that can be evaluated down by X% it may still not be acceptable for a chess engine. On the other hand, if the evaluation idea includes some principal that can be easily applied by a person, it may still be extremely valuable...
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-19 12:08

>Note that passing verification in Stockfish means not only that the evaluation idea is good, but also that it can be applied fast enough to justify its benefits, i.e. if you have a great evaluation idea but it slows the number of >nodes per second that can be evaluated down by X% it may still not be acceptable for a chess engine. On the other hand, if the evaluation idea includes some principal that can be easily applied by a person, it may still be >extremely valuable...


That is right, but only good evaluation ideas succeed, for the most part.
Most of the newly introduced evaluation terms in Stockfish have to deal with new advanced chess knowledge concepts,
no chess authors have expounded in the past.

That only points to the fact that chess is much more complex and hides a much bigger amount of knowledge
than what we are accustomed to accept.

All the sophisticated chess knowledge and all the greatest chess games are ahead of us!
I have ascertained this during my countless analysis sessions.
Parent - By Banned for Life (Gold) Date 2017-11-19 12:50
only good evaluation ideas succeed, for the most part.

That is true. Being a good evaluation idea is a necessary, but not sufficient, condition for succeeding in a chess engine. To be sufficient, the idea must also not slow down the engine to the point where the improved eval is more than compensated for by a crippled search.

All the sophisticated chess knowledge and all the greatest chess games are ahead of us!
I have ascertained this during my countless analysis sessions.


I believe you are correct that we are much closer to the beginning than the end when it comes to chess knowledge. Chess engines compensate for this by searching deep into the game to see the results of many different options. People don't have that luxury, so a better understanding of chess knowledge based on better position evaluation with minimal calculation is much more important.
Parent - - By Kappatoo (*****) [de] Date 2017-11-26 10:20
Sorry for the somewhat arbitrary question: What is the approximate size (entropy) of an evaluation function in a chess engine?
Parent - - By Labyrinth (*****) [us] Date 2017-11-26 22:08
What do you mean? Lines of code? Binary bits? Size of the equivalent logic in a human? Any chess engine or a strong one?
Parent - - By Kappatoo (*****) [de] Date 2017-11-27 21:29
I was thinking of bits, as coded. But any measure would be helpful. Also, I was thinking of strong engines.
Parent - - By Labyrinth (*****) [us] Date 2017-11-28 07:04
Stockfish would be the obvious choice then, being open source and one of the strongest if not the strongest engine currently available. I think someone could give you the size of the source files that pertain to the eval function, but a lot of those bits would be unnecessary, pertaining to spaces and variable names and syntax decisions made for readability as opposed to terseness. Then there's the issue that when compiled it is initially going to be made much more efficient by the compiler using the minimum possible amount of information/instructions, and then made binary, which will serve to lengthen the needed bits in most instructions.

Would probably be a nightmare to say just look at pawns.cpp/evaluate.cpp/endgame.cpp, ehm, material.cpp, movepick.cpp, psqt.cpp and perhaps bits and pieces from other files, try and get them to compile into something so that all the syntax/comments and such would be stripped out, and then look at the file size of the resulting binary. Another strategy would be to gut its search function, thread stuff, tablebase support, board representation, move generator and see if what's left will compile. Shrug. Someone actively working on SF would be in the best position to do such a thing.

The SRC dir itself is ~430k, about ~140k if I include just those files mentioned above. Unfortunately there is no reliable way to translate that into what the total contributions to the binary file size would be, as a smaller source file could contain greater and lengthier instructions than a larger one when compiled.

The exe file size could represent an upper bound, stockfish_17111816_x64_bmi2.exe for example is 1,210,880 bytes (9687040 bits, 1.2 MB).
Parent - - By Kappatoo (*****) [de] Date 2017-11-28 13:01
Thanks! That at least gives me a rough idea.
What would be a lower bound on how much inefficiency is in a source code (written by competent humans) before compiling? I assume this also depends on many things, but any guess would be helpful.
Parent - By Labyrinth (*****) [us] Date 2017-11-28 18:11

>What would be a lower bound on how much inefficiency is in a source code (written by competent humans) before compiling? I assume this also depends on many things, but any guess would be helpful.


Well in the case of SF there isn't much inefficiency :-)

Really though I wouldn't know where to begin on answering this question. Just depends on way too many variables. For one there's the trivial things: comments are removed, variable names can cease to exist entirely (functions may only need relative positions of data), there's absolutely no formatting, like literally it's just an endless string of 0's and 1's that gets picked up in 64 or 32-bit pieces.

Simple lines in C or C++ often require many instructions, for example here is a C code on the left and the corresponding assembly on the right. When converted to binary each of those instructions and the data they act on will be in 64 or 32-bit binary pieces.

It seems that, based on very anecdotal evidence, that going from c++ -> assembly -> binary represents an increase in the size of the output file despite a lot of unnecessary things being removed.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-28 08:23

>I was thinking of bits, as coded. But any measure would be helpful. Also, I was thinking of strong engines.


We have been checking the relative size of Stockfish and Komodo eval some time ago and, if I have not forgotten
the precise numbers, Stockfish eval consisted of around 3 or 4 thousand lines of code(white space, etc. included).
Komodo eval, according to stats presented by Mark Lefler, proved to be somewhat bigger, but not
exponentially, maybe 1.5 times bigger in terms of lines of code.
Of course, size not always means quality, there are other factors to consider, this is just as an indication.

What is even more certain, though, is that evaluation and search/move ordering are fully inseparable,
so this makes any comparison even more uncertain.

I was just browsing the games of the TCEC superfinal these days, and it is really stunning how much more
the top engines have to learn in terms of evaluation(just to mention the closed game Komodo lost after the f5 break, and the other game with
d5 and e6 connected passers, which both engines evaluate fully wrong for astoundingly long periods of time, until closer
search suddenly reveals the truth is otherwise).
Parent - - By Kappatoo (*****) [de] Date 2017-11-28 13:04
I agree that engines can still play awful chess in closed positions. (By the way, I've never seen a real explanation for this. Any ideas?)
But I find it hard to believe that it's possible to systematically exploit this in regular tournament conditions.
Parent - - By Labyrinth (*****) [us] Date 2017-11-28 17:44

>I agree that engines can still play awful chess in closed positions. (By the way, I've never seen a real explanation for this. Any ideas?)


The traditional explanation is the horizon effect. In closed positions humans can think strategically as opposed to concretely whereas the computer is hopelessly bound to its concrete calculation. So in closed positions, the computer kind of spins its wheels on boatloads of dumb lines.

To give a really poor example to illustrate the concept, if you have say a black king on h1, a white king on f1, and a white pawn on a2, it's immediately obvious that white has a won position to a human with zero calculation. The computer however has to either contain programmed in heuristics that account for this case, or calculate it out. Of course calculating it out takes mere microseconds for the machine because there are so few legal moves, but the point is that a human wouldn't need to do such a thing at all.
Parent - - By Kappatoo (*****) [de] Date 2017-11-28 21:44
I still find it surprising. For a while, I thought that the issue might be a search tree explosion: in a closed position, there are too many equally good-seeming moves in a position that all need to be calculated. But the thing is that engines still reach very significant depths in closed positions. In my opinion, human strategizing isn't usually that long-term - I don't think my plans in closed positions are deeper than 35 plies.
I recently saw a game between Fire and Houdini in TCEC in a closed king's indian position that was really unbelievably bad. Somehow, their search often seems to miss the relevant pawn levers and breakthroughs.
Parent - - By Labyrinth (*****) [us] Date 2017-11-28 23:57

>For a while, I thought that the issue might be a search tree explosion: in a closed position, there are too many equally good-seeming moves in a position that all need to be calculated.


Also a contributing factor.

>But the thing is that engines still reach very significant depths in closed positions.


Sure but the depths don't mean as much, especially if good moves are getting pruned early.

>In my opinion, human strategy isn't usually that long-term - I don't think my plans in closed positions are deeper than 35 plies.


It's not n plies at all, it's 'I'll put the rook here to oppose the opponent's rook', or 'rook goes here because this file is most likely to open', or 'I'll transfer the knight here because it'll have the best chance of a future outpost', or 'I'll put the king here where the enemy bishop has less of a chance of reaching it'. Such moves from a concrete standpoint might have very far reaching effects but we don't think of them that way.

If you just thumb through some GM games and look at certain pieces that don't move much, you come across things like a player plays say Rc1 some time in the opening, and then 55 moves later the rook plays some important role on the c file and is just in time for some particular action. Did the GM know that was going to happen when they played it? Of course not. Strategy and positional play are sort of ply-less like that. Strategy can be really general, like preparing for a pawn break or securing certain squares.

>I recently saw a game between Fire and Houdini in TCEC in a closed king's indian position that was really unbelievably bad.


Lessee it
Parent - - By Kappatoo (*****) [de] Date 2017-11-29 11:19
My apologies to Fire, it was Chiron.

[Chiron 251017 (3013) vs. Houdini 6.02 (3184), TCEC Season 10 - Stage 2, Round 4, 2017.11.08]
1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 5. Nf3 O-O 6. Be2 e5 7. O-O Nc6 8. d5 Ne7 9. Ne1 Nd7 10. f3 f5 11. Be3 h6 12. Nd3 c5 13. a4 f4 14. Bd2 a5 15. Qc2 Nb8 16. Nb5 Na6 17. Qc3 Nb4 18. Nf2 g5 19. Qb3 Ng6 20. Rad1 h5 21. Rc1 Rf7 22. Bc3 Nh4 23. Ra1 Kh8 24. Qd1 Qf6 25. Rb1 Bd7 26. h3 Bh6 27. Bxb4 axb4 28. Qd3 Raf8 29. a5 g4 30. hxg4 Qg6 31. Ra1 Bg5 32. Nc7 hxg4 33. Ne6 g3 34. Nh3 Bxe6 35. dxe6 Rh7 36. a6 Be7 37. Qd1 Qh5 38. Re1 Nxg2 39. Bf1 Nh4 40. Be2 0-1

Black's play was fine, although I have my doubt if that Nb8-a6-b4 maneuver was necessary. But what white did was just ridiculous.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-29 19:08

>Black's play was fine, although I have my doubt if that Nb8-a6-b4 maneuver was necessary. But what white did was just ridiculous.


White could have drawn with 17.g4(so that the g2 shelter pawn is no more backward, which loses the game).
fg3 hg3 should be bad for black, improving white's pawn structure, while after h6-h5, white has simply h2-h3,
supporting g4 and should hold, no breaking through.

Houdini also did not find the best winning moves, but then all black pieces had good places in this game, so converting
was not a problem.
Give Houdini a more complex KID position and stronger opponent, and it starts blundering.

I don't know which of the 3 top are worst in closed postions, all seem to almost share the same bad place.
Parent - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-30 18:15 Edited 2017-11-30 18:17

>White could have drawn with 17.g4


I wanted to make an observation about the g4 move from the Chiron-Houdini game, and I wanted to post the fen just before white's 17th move, but am seeing only a pgn option.
Anyone knowing how one could post fens here?(in any case, I would be glad if someone could post above position, or you can just refer to the game.

I was thinking about this position and how concepts in 'The Secret of Chess' work.
'The Secret of Chess' has 2 different terms that apply to above position:
- pointed chain, one type of pointed chain is defined, from black's perspective as d6,e5,f4 black pawn, d5,e4,f3,g2 white pawns, white king on
the king side; so just what we are seeing on the diagram; this pointed chain is considered a very good positional factor and gets around 50cps bonus
- unbackwarded pawn(Labyrinth laughed a lot at me here, but I don't think it is so funny :smile:), defined as a pawn on the relative 4th rank with an
opponent pawn on its relative 5th rank on adjacent file; this feature is also considered good and gets some small bonus, around 15cps

So, why is 17.g4 the best move without a deep search and a lot of calculations, just using pattern recognition?
Well, because after g2-g4, the black pointed chain is already not there(the g2 pawn is missing), and instead on g4 an unbackwarded pawn
has appeared(g4 is on the 4th rank, with opponent f4 pawn next to it on adjacent file).
So, in a single move, white has gained some 60cps or so, by liquidating the enemy pointed chain and creating a friendly unbackwarded pawn.
Any engine that knew such features would immediately play this and draw. Instead, Chiron considers that after g4, the pawn cover close to the white king is
insufficient and that is why it avoids that move, wrongly.
Similarly with humans, it is not necessary to calculate anything, but just to recognise a lot of evaluation terms.

That is why I think 'The Secret of Chess' has its added value, including as a tutorial.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-29 18:37

>I agree that engines can still play awful chess in closed positions. (By the way, I've never seen a real explanation for this. Any ideas?)
>But I find it hard to believe that it's possible to systematically exploit this in regular tournament conditions.


It is all very simple: in closed positions there are no direct attacks of pieces upon each other that could relieve the tension and
transform the evaluation nodes, by which the engine judges how good its position is, instead, in order for an evaluation shift
to occur, both sides should do a lot of regrouping, involving sometimes more than 20-30 moves/60 plies, and that is simply too long
for engines.
Engines could have resolved such positions successfully, if they had more intricate evaluation, but that is unfortunately not the case
with modern top engines. So, they are bad in closed positions because:
- these require too much depth for an ordinary evaluation which engines have
- and engines lack the sophisticated evaluation with which they could have understood the position better even without very deep search

So, the answer is simple: depth and lack of sophisticated evaluation.
For example, top engines still completely misunderstand/lack evaluation terms for pointed chains.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-29 18:47
Concerning whether humans could exploit that under regular tournament conditions, of course they could,
they simply have not tried, apart from Kasparov, and even he blundered severely positionally in his 2003 game
against Fritz in a KID, when he should have played f5-f4, but instead allowed Fritz to capture on f5.

Not to mention the way Nakamura plays against top engines: not a trace of a positional approach, and that is
why he lost in 25 moves to Komodo 2 months ago with a knight for pawn handicap.

Humans/top humans simply have never investigated deeper/in a scientific way closed positions, that is the simple
truth.

Btw., if you have noticed, closed positions start to occur more and more often at TCEC(including the current Superfinal),
so one can say closed positions are the future of chess and the deeper an entity's understanding, the more it will prefer closed
positions.

Chess is all about depth, and the greatest depths are found in closed positions, so they are the most complex.
The most complex things are understood only last, by both humans and engines alike.
Parent - - By Kappatoo (*****) [de] Date 2017-11-29 21:41
In most cases, it's sufficient to exchange one pair of pawns to avoid getting into a closed position. How are you going to avoid that? Isn't a pretty simple opening book going to take care of this?
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-30 17:52

>In most cases, it's sufficient to exchange one pair of pawns to avoid getting into a closed position. How are you going to avoid that? Isn't a pretty simple opening book going to take care of this?


No, if changing a pair of pawns cedes a significant portion of the advantage.
It is meaningless to trade 2 pawns only for your score to drop further 20-30cps. And in many cases, it is either the trade that
will make the engine lose 30cps, or going into a closed position.
So, it is not that easy to open the position at all.
Parent - - By Kappatoo (*****) [de] Date 2017-11-30 21:27
How about a test? Let's play 15 or so moves here on the forum. Your goal is to get a position that is either clearly better for you or so closed that the engine will play it poorly. (An alternative, possibly even more telling test would be that you play out the resulting position against an engine.) What do you think?
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-12-01 20:21

>How about a test? Let's play 15 or so moves here on the forum. Your goal is to get a position that is either clearly better for you or so closed that the engine will play it poorly. (An alternative, possibly even more telling test >would be that you play out the resulting position against an engine.) What do you think?


1.c4, won't play a whole game, just till the position gets closed.
(and I don't know how often I will drop by, but when I do, I will input a move)

Anyway, I am not able to always close the position, just statistically very frequently, so this might be a game where I am unable to do so.
Parent - - By Kappatoo (*****) [de] Date 2017-12-01 20:24
So, should we say 15 moves?
1...Nf6
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-12-02 08:55
Man, you are cheating. :wink:

I don't have latest Houdini, but Komodo and Stockfish never answer 1.c4 with Nf6...
So I really wonder whom am I playing.
The contest was not whether a human can avoid closing the game against me, but if an engine can do so.

Never mind, this gets funny, let's continue a bit, 2. Nc3
Parent - - By Kappatoo (*****) [de] Date 2017-12-02 09:28
So you're playing without opening book? My question is whether it wouldn't be easy to steer away from closed opening positions with an appropriate book.

2...d5
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-12-02 16:32

>So you're playing without opening book? My question is whether it wouldn't be easy to steer away from closed opening positions with an appropriate book.


>2...d5


Of course, I am playing without opening books, why do a need a book, only to repeat moves someone has already played. Besides, I am not playing against a book, but against an engine.

No, it would not be easy, because you just ceded 20cps advantage with d5.

My answer is 3.cd5, after which white has traded a weaker semi-central c pawn for a central d black pawn.
Most probably, in centipawns, a d pawn is stronger than a c pawn by about 1/5 of a full pawn.
Fischer has won countless like those with white.

Btw., neither Stockfish or Komodo ever consider 2...d5 on my system, preferring e5 instead.
Parent - - By Kappatoo (*****) [de] Date 2017-12-02 16:41 Edited 2017-12-02 16:44
How do you avoid learning any opening theory yourself? And if you use an opening book while the engine doesn't, that automatically gives you an advantage.
Also, opening books will lead to more variation in what the engine plays.

> My answer is 3.cd5, after which white has traded a weaker semi-central c pawn for a central d black pawn.
> Most probably, in centipawns, a d pawn is stronger than a c pawn by about 1/5 of a full pawn.
> Fischer has won countless like those with white.


Any idea how often Fischer played the Grunfeld with black, and open Sicilians with white?

3...Nxd5

Edit: Let me add a board.

1.c4 Nf6 2.Nc3 d5 3.cxd5 Nxd5
Up Topic The Rybka Lounge / Computer Chess / The Secret of Chess
1 2 3 4 5 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill