>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.
>Any idea how often Fischer played the Grunfeld with black, and open Sicilians with white?
How can I use an opening book, when I don't know a single theoretical move? It is all improvisation.
Fischer did not play very often the Gruenfeld, just very rarely; his black KIDs are way more numerous.
That is the point, the Sicilian is good for black(better than 1...e5), precisely because of the same pattern: a semi-central c pawn
capturing a central white d pawn. The Sicilian is split 51-49, one of the best openings for black.
> How can I use an opening book, when I don't know a single theoretical move? It is all improvisation.
> Fischer did not play very often the Gruenfeld, just very rarely; his black KIDs are way more numerous.
> That is the point, the Sicilian is good for black(better than 1...e5), precisely because of the same pattern: a semi-central c pawn
> capturing a central white d pawn. The Sicilian is split 51-49, one of the best openings for black.
You know plenty of openings for someone who doesn't know a single theoretical move.
1. c4 Nf6 2. Nc3 d5 3. cxd5 Nxd5 4.g3 e5
As for the game, no human would last very long against an engine in this position, with either color.
>Hehe. Not sure if I should take this as a compliment.
>As for the game, no human would last very long against an engine in this position, with either color.
White is considerably better, even though top engines might still not understand it.
About playing this position against an engine, no problem holding the draw at all, the query is whether
a human can easily win this.
My experience is that, in more open positions, even if I have some advantage, sometimes bigger one, I will mostly
only draw, but that is based on fast/blitz games. I guess at longer TC, the human chances considerably improve even in open
Matter of fact is, you can not equalise this with black.
So you're saying it's not a problem to hold this position against a top engine in tournament conditions? How about a bet?
By the way, would you mind if we opened a new thread for the game? It's getting a bit cramped here at the margin.
>It seems that your secrets of chess (such as 'it's bad to trade a d-pawn for a c-pawn') have not only eluded top programs, but also virtually all top human players in the world, most of whom have played this setup with >black (Carlsen, Aronian, MVL, Karjakin, So, Giri, Topalov, etc.).
>So you're saying it's not a problem to hold this position against a top engine in tournament conditions? How about a bet?
>By the way, would you mind if we opened a new thread for the game? It's getting a bit cramped here at the margin.
Whatever you decide about the thread, we will not play very long, in any case.
I guess it is fine here, unless it is impossible to post at all.
Let's leave bets for some future point in time, I am not going to compete, because someone wanted something proven.
At some point, I guess, this will happen too.
You enumerated some names, those are mere mortals, Fischer, on the other hand, knew better.
[Site "Siegen Exhibition Ga"]
[White "Fischer, Robert J"]
[Black "Andersson, Ulf"]
[Opening "Nimzovich-Larsen Attack, Modern Variation"]
1. b3 e5 2. Bb2 Nc6 3. c4 Nf6 4. e3 Be7 5. a3 O-O 6. Qc2 Re8 7. d3 Bf8 8. Nf3 a5
9. Be2 d5 10. cxd5 Nxd5 11. Nbd2 f6 12. O-O Be6 13. Kh1 Qd7 14. Rg1 Rad8 15. Ne4
Qf7 16. g4 g6 17. Rg3 Bg7 18. Rag1 Nb6 19. Nc5 Bc8 20. Nh4 Nd7 21. Ne4 Nf8 22.
Nf5 Be6 23. Nc5 Ne7 24. Nxg7 Kxg7 25. g5 Nf5 26. Rf3 b6 27. gxf6+ Kh8 28. Nxe6
Rxe6 29. d4 exd4 30. Bc4 d3 31. Bxd3 Rxd3 32. Qxd3 Rd6 33. Qc4 Ne6 34. Be5 Rd8
35. h4 Nd6 36. Qg4 Nf8 37. h5 Ne8 38. e4 Rd2 39. Rh3 Kg8 40. hxg6 Nxg6 41. f4
Kf8 42. Qg5 Nd6 43. Bxd6+ 1-0
And this was only one of many Fischer games with this concept. There is no doubt, Fischer is the best theoretician of the world champions,
way way above the rest.
> Let's leave bets for some future point in time
Cluck, cluck, cluck.
>Cluck, cluck, cluck.
Well, I know hens do like this, but I am not certain of the figurative meaning.
>Please continue the commentary. It's much better than the game!
As soon as I mate him.
>Sorry for the somewhat arbitrary question: What is the approximate size (entropy) of an evaluation function in a chess engine?
Smaller(including less details) than what is necessary in order for human players not to be able to exploit some
of its weaknesses due to imperfect evaluation and beat it.
One method that I've proposed in the past is testing with an eval function in a manner not constrained by time, i.e. maybe counting the eval calls rather than looking at the time spent in them. In this way, it could at least be determined whether an idea was good, decoupled from the efficiency of the implementation.
As an aside, I have a suspicion that if the speed of the eval function wasn't so important, methods other than rule based systems, e.g. neural net based implementations, might be considered to be better implementations. But Bob Hyatt thought this was a terrible idea and wrote a whole bunch of messages saying how idiotic that concept was!
So how quick is the evaluation in a top engine? And how much slower would the evaluation derived from a neural net be? (I know, I'm asking these impossibly broad questions in this thread.)
It would almost certainly work extremely well if you could figure out how to train it! This is not dissimilar to the human case!
Most attempts at applying NNs to chess have simplified the training problem by putting in specific features of the chess problem, rather than trying to work with the complete state. I worked with a Polish PhD student a number of years ago, and we put in the complete state using the FANN package, and then tried to train it on a million positions that were evaluated by Rybka 2.2n over a period of 30 seconds. The flaw with this approach is that we tried to replace both the eval and the search with the NN, and while the NN could almost certainly be trained to provide a better eval than a chess engine (albeit much more slowly implemented), the search function is still essential.
If someone implemented a NN evaluation based on the full state and coupled this with a good search function, a very strong engine could possibly be implemented that played beautiful chess (if the NN could be trained), but it's still not assured that a slower evaluation function could compete head-to-head with a faster but inferior evaluation function. Unfortunately they can't be decoupled because the eval function must guide the search, so the whole thing is slowed down to a point where a deeper search with an inferior eval is stronger. Maybe another option which might be considered would be to just use the NN on the terminal nodes.
I think the difference in go is that use of the eval in intermediate nodes is much less useful in guiding the search, so it's more a case of going as deep as possible before evaluation, and maybe in this case a NN competes well with a faster rule-based evaluation function...
DeepMind has made some positive experiences with a neural net that inputs the complete state. Also, their neural net is on the level of a professional player, without doing any kind of search.
The previous AlphaGo version had two evaluation functions, the faster of which was used for the Monte Carlo rollouts. It takes two microseconds to select a move, which seems acceptable to me. (The other, slower but better, evaluation function takes 3 milliseconds.) Apparently, their latest version only has one evaluation function, but I haven't found any information on how fast it is.
I would love to see someone try to train a neural network for chess.
Thanks. That seems to have happened in the last few years. I'm pretty sure that they will not be able to reach top engine levels without implementing a guided search function. But I would not be too surprised if they can reach top human levels without a separate search function (a multilevel NN should provide the equivalent of a few plies of search). This would still be very interesting though because even though it would miss deep combinations, it would really be able to teach the fundamentals of positional evaluation much better than any engine and maybe even better than any human.
My recollection is that an engine's eval took a few microseconds several years ago when Bob measured it.
I would love to see someone try to train a neural network for chess.
I would also love to see this done as an open source project based on the complete game state. In fact, it would be really interesting if this was developed and distributed to the community to run a competition to find out who could train the same chess playing NN most effectively!
>I would not be too surprised if they can reach top human levels without a separate search function
I would be really surprised about that. I think chess is too concrete and tactics are too dependent on small details. But I'd be happy to be corrected.
Let me note that I don't think it's impossible to have an evaluation function that plays very strong chess - after all, there is a sense in which tablebases represent an evaluation function. I just think that an evaluation function that is practically feasible can't be that strong.
Tablebases are a special case of course since they really on exhaustive enumeration rather than pattern recognition...
authoritative source on the matter.
The primary problem with improving top chess engines, apart from finding new evaluation and search ideas,
is to be able to get rid of countless inherent redundancies, both in search and eval. And that is very difficult,
because, you should either know the whole engine structure by heart, making mental measurements what could be
changed and what not, or run countless tuning tests.
Stockfish even don't tune any more these days, they just test single ideas one after another, maybe Komodo does
the same, I am not certain about Houdini.
Either way, both approaches are extremely difficult to implement.
authoritative source on the matter.
Many people on this site can attest that I've had my differences with Bob over the years but Crafty was certainly a top chess engine for many years before time passed Bob by. This is hardly an indictment of Bob though, as all engine developers have lost the bubble at some point...
But regardless, nobody can dispute that Crafty was a top engine for many years, or that Bob contributed as much as anyone to computer chess...
>I guess the main problem with our communication are simply different levels of, how to say it, understanding.
I strongly disagree. The writing is pretty bad in many places.
>The version date and the real version might be different, in any case, I have not made updates on the 18th.
The date given was when I accessed the page. I have no knowledge of the versions.
>this is an issue with the ebook on Amazon, and the fault is with their file convertor, in the original doc/pdf there is a space.
Yikes. That's a pretty serious problem then as it can affect sales if the text appears all wonky. They probably have guidelines or something for how their previews are generated so you can mitigate issues.
>Some are full of spelling mistakes in the description even,
I've actually never seen this. In any book actually. Which is not to say it doesn't happen, but it must be exceedingly rare.
>not to mention double spacing.
It's not so much about double or single spacing, it's about being consistent.
>Irving Chernev's 'Move by move' features zero spacing sometimes, and still it sells incredibly well.
The Table of Contents on that book is vastly superior to yours. Much cleaner/readable layout, consistent spacing, use of bold, page numbers, and so on.
>Not mine, sorry, I guess this is a problem with all paperbacks on Amazon
I've never seen it before, but I believe you that it could be a problem with Amazon's file conversion mechanism. Sadly, if it is not amazon that will adjust their software you will have to adjust your own files to work with it.
Yes, my bad, I did mean 'foreword'. The 'in lieu of', while known, is still out of place imo. It's a bit too pretentious.
>I guess this is a bit pedantic; I always like to underscore rhyme, even in prosaic text, it might not sound well in English, but that is how I like it, that is my inner nature.
What rhymes in "Pawns on squares the colour of the own bishop"? Rhymes are words that have the same sound, like 'sound' and 'round', or 'lake' and 'cake'.
"the own bishop" is flat out grammatically incorrect. It is not stylistic or pedantic.
> the title reads 'Special rule for the imbalance knight versus bishop' and I see nothing wrong with it. As said, your suggestion lacks rhyme.
It makes no difference to my point whether you use 'and' or 'versus' here. "Special rule for the knight versus bishop imbalance" is still more correct. Any time you have a category that is noun it should go after the name of the object. It's a tennis racket, not racket tennis. If it's only an adjective then yes, before is preferred. A blue racket. Combinations are fine: a blue tennis racket.
>Don't understand you; on the chess board, sometimes double pushes are available, and sometimes not(for example, when the pawn on its 2nd rank is blocked by an own or enemy pawn/piece, or there is a pawn/piece on the relative 4th rank. So that, my phrase is technically fully correct.
I did say it was correct, but it is very strange. Availability is determined by the rules of chess here. If you're talking about it, the assumption is that it is legal, therefore available.
>One says 'trapped by own and enemy pawns', while the other just 'trapped by enemy pawns'.
It seems that one is a subset of the other, as in 'trapped by enemy pawns' is a superset of 'trapped by enemy and friendly pawns'.
You cannot say 'an own pawn', this is not grammatically correct.
>but I prefer to stress the making function
I might be able to tolerate 'passed pawn makers'. Maybe 'builders' would be better. Passer-makers.. yich.
>>If the number is less than 10, spell it out.
It isn't correct grammar otherwise, and it also looks really shoddy.
>Nope, vertical isolation is a term present in many chess books, and it is a very real term, referring to a pawn that has other friendly pawns on adjacent files, but those are not on an adjacent rank.
I've never encountered this term before, and google hasn't either (outside of your writings). So if you play 1. g4, is the g4 pawn "vertically isolated" since it has friendly pawns on adjacent files (f and h), but they are not on an adjacent rank, which I assume would be the third or fifth rank? This doesn't fit the standard criteria for an isolated pawn though, and it implies that there is such thing as a 'horizontal isolation' which is even stranger.
>Backward-maker is any pawn that makes an enemy pawn backward in some way,
It's hard to believe that this warrants its own term.
>and an unopposed pawn is a pawn that does not have an enemy pawn on the same rank. I guess you can do the maths.
I think you meant same file here? Same rank wouldn't be opposing.
>Again, rhyme and elegance of perception; I can use colour, as it writes good, but no 'centre', as there are 3 consonants one after another, and that is ugly.
Stick to colour/centre or color/center!
>Nope, there are 2 central definitions in chess: the innermost center, the 4 central squares, and the extended center, so a distinction is just obligatory, especially in a title.
If you say 'center squares', you mean d4/e4/d5/e5. If you are talking about the extended center then you must specify.
>The is a chain and an inchoative chain; and inchoative chain is a chain that has started being built, but is not yet fully built.
The word you are looking for is inchoate. 'Inchoative' for this purpose is incorrect.
>The new idea are the 150 new evaluation terms.
So the concept is that, instead of normal 'evaluation terms', such as open files, weak squares, doubled pawns, and so on, that there are new ones that a human chess player should consider, and they should consider them because they've been a big help as heuristics for computer chess engines? If so, this is not adequately conveyed in either the foreword or the description.
>Right, but more than 20 of my evaluation ideas have been incorporated into Stockfish code, so they have passed their verification.
As heuristics for a chess engine perhaps, but humans are not chess engines. Such heuristics might help humans as well, but this is not an obvious connection to be automatically made.
>I strongly disagree. The writing is pretty bad in many places.
it is too long to comment on everything, one way or another we will never agree,
so just briefly on the most important points.
What matters to me is that Mark Lefler, the author of Komodo, said about the book:
'Great book. In the first 30 or so pages I skimmed I found 30 ideas to try implementing in Komodo. Great stuff."
He also considers my English and grammar acceptable enough.
What also matters to me is that couple of highly intelligent university professors teaching in the US
assured me the text is good enough and there are no grammatical and spelling mistakes.
Some of them even referred to the text as 'excellent'.
And that Stockfish programmers appreciate my efforts.
That is sufficient for me and a way more valuable feedback than anyone else could give me.
Of course, university professors and other people don't see things in the same way.
Otherwise, thank you very much for your moral support, I strongly appreciate that.
>3 months on, I have to ascertain there is not much feedback/interest in my book. For myself, I am not entirely clear what the main cause is. What do you think?
Now you are saying this:
>Stockfish programmers appreciate my efforts
>university professors teaching in the US assured me the text is good enough and there are no grammatical and spelling mistakes.
>That is sufficient for me and a way more valuable feedback than anyone else could give me.
The timing of this abrupt change of heart does not show you in the best light.
..and if a University professor in the U.S. told you 'the own pawn' is correct grammar they must not be native!
This book was not well received even on talkchess.com where chess programmers like to go, why would you think it'd sell well to the general public? While chess programming is not a big market, it's also not bad. There's not a single decent book on chess engine programming, anybody who makes a new good book is almost certainly sell relatively well. Chess engine programmers are generally well paid in their day job, they have the money to buy a good engine programming book they want. Money is not an issue, it's the contents and Lyudmil Tsvetkov himself.
Look, Lyudmil Tsvetkov is an arrogant fool. He's blaming his own failure to the market. It's his own personality that brings to his failure. It has nothing to do with the market, because not even chess programmers buy the book. He likes bull-shit his craps, unless he can claim his theories, this book is simply a PDF document with some stupid random games. There's nothing to learn, it's an e-trash.
Lyudmil Tsvetkov is supposed to pay us for reading his dumb book, but the other way around.
>know how to code alpha-beta?
It's not about chess programming.
>talkchess.com where chess programmers like to go
No way, no one has ever heard of that forum before.
I really don't think you have anything meaningful to contribute to this conversation :|
Thank you, Labyrinth, again!
Anyone knowing where the 'Reply with quote' button on this forum is?
I don't see it immediately, and I would hate to read all the instructions.
>There is no 'reply with quote' function here. I just add a > and copy/paste. What's after a '>' the forum automatically greys if it's on the first line. You can use the preview button to see the result.
ok, I understand now.
without a verified purchase, of course.
You are even mixing up the 2 books, 'Secret of Chess' and 'Human versus Machine',
the one does contain full games and the other does not, but you picked the wrong one for the games...
I wonder why people who HAVE NOT READ the book should write reviews?
because it is a valuable book, and also a one of a kind book, a trail-blazer.
Machine Learning, I would strongly recommend that you watch your language.
If you read talkchess at the beginning of this year, we quarelled a lot about his claims he is + 30 elo
with his McBrain to Stockfish, but he still bought the book.
He bought it, because he is a an intelligent person, a person with experience, and he knows to
distinguish between good and bad things.
All the reviews I have from people who actually read the book, friends or not, are GOOD.
The 2 reviews from persons who DID NOT read the book are 1 star.
when will you stop that?
You are one and the same person here, on hiarcs forum and chess chat, with 3 different names:
- Machine Learning
That is all the support you UGLY denigration campaign has: a single person, not very well educated,
quite probably young and inexperienced, and full of venom.
Can you please tell me, what did I do to you to behave like that?
Did my book kill your selling niche?
Your vision of the world?
It is good to be negative without a reason, to a certain point, but that much...
PS. Most of the people I know come from talkchess, you are registered there and certainly have an account,
I try to remember who you might be, still unsuccessful, maybe you will reveal your real name...
>No. I only post here and talkchess. They are not me. I'm here simply telling the truth.
I know very well what truth you are telling.
You just wrote 2 1-star reviews for the 2 parts of 'Human versus Machine', and another one
for 'The Secret of Chess', under the name Serverless.
But don't you feel ashamed?
I have reported the issue for abuse to Amazon, hope they take the appropriate measures.
God is high above, rest assured you will be appropriately punished.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill