Originally posted by EcstremeVenomThis is not true. (whoever said it.) It's just a myth.
so if people make the engines then the engines cant be any better than the player themselves. the engines will never be better than humans.
I myself made a program for a game once. I used alphabeta-search and the computer whipped my *ss rather quickly.
Chess programs are just using brute force calculations rather than 'intelligence'.
I would be quite surprised to find a computer program that used 'intelligence' to learn the game. You give the rules of the play and by its own mistakes makes progress to the skills of world champions.
We are far from it in chess programs.
Originally posted by Dragon FireBut programs aren't computers. Computers run programs. With the correct program a computer can do pretty much anything.
Not with the same program!
Edit: but seriously you know the point I am attempting to make. I just worded it badly.
And if I really wanted I could write a engine/word processor/orbit calculator program that did all of those things.
Originally posted by CrazyLilTingRubbish. Several chess algorithms have been developed and implemented successfully since Botvinnik's work.
Rubish. The only algorithm I see that can go through is the one authored and published by Botvinnik in his book "Computers, Chess and Long Range Planning". He (his collaborators) failed to implement it on computer's code.
Now, after 50 years or so, there is a new commercial attempt to bring it to life. It is called SmarThink 1.0. You can find some info ...[text shortened]... ure they'd get an excellent piece of code, but for the time being, it is less than nothing.
I agree that SmarThink is relatively weak, and I found it buggy, too.
Rybka appears to offer a significant new breakthrough in chess software. However, it has not proven itself yet against strong grandmasters. Rybka's strength estimates are based on its strength against other software (and I suspect that it employs a human-like anti-computer strategy). It remains to be seen whether it will do as well against GMs as Junior, Fritz, and Hiarcs, or whether it will do better.
Originally posted by Dragon FireSpeak for yourself. Most reasonably competent writers make fewer proofreading mistakes than software spell and grammar checkers. We all make mistakes that the software catches. But, the software also identifies as possible errors many things that are correct, and completely overlooks certain fundamental errors that no competent human proofreader misses.
I get bored checking the spelling and grammer in that 50000 word novel long before a computer gives up and make more mistakes
To fail to use software proofreading is foolish and lazy; to rely upon it invites disaster.
Originally posted by HFRorbisDoubt Kasparov would beat it, as hes only looking to beat the political system in Russia right now, and is retired from Pro chess.
look at this list :
http://computerchess.org.uk/ccrl/4040/
Rybka 2.1 which has the highest rating is rated 3100 elo
When I first saw this rating I couldn't believe it.
It is incredible that a chess engine overcomes 3000 elo
At last chess engines have become better than the best chess players
Could Kasparov beat it ? Don't think so
Maybe online, as he does play occasionaly, i heard
Originally posted by Wulebgr... and all brute force search algorithms, based on alpha-beta prunning, with bactracking and the most succesful using a better evaluation function at the nodes.
Rubbish. Several chess algorithms have been developed and implemented successfully since Botvinnik's work.
I agree that SmarThink is relatively weak, and I found it buggy, too.
Rybka appears to offer a significant new breakthrough in chess software. However, it has not proven itself yet against strong grandmasters. Rybka's strength estimates are based ...[text shortened]... her it will do as well against GMs as Junior, Fritz, and Hiarcs, or whether it will do better.
Botvinnik's approach is completely different.
Please read his book, try to understand his concept (not a simple matter), and then post again your thinking about it.
SmarThink is the only program (as advertised) that tried to implement such algorithm, but it is flawed. May be next versions become better.
Cheers!
- J
Originally posted by CrazyLilTingLet me get this straight: you are calling "rubbish" the statement regarding Rybka's strength from promotional materials advertising the program based on your reading of a book published in 1970?
... and all brute force search algorithms, based on alpha-beta prunning, with bactracking and the most succesful using a better evaluation function at the nodes.
Botvinnik's approach is completely different.
Please read his book, try to understand his concept (not a simple matter), and then post again your thinking about it.
SmarThink is the only pro ...[text shortened]... plement such algorithm, but it is flawed. May be next versions become better.
Cheers!
- J
I also think that you are using the term "brute force" a bit more broadly than standard usage in the field of computer chess.
1 edit
Originally posted by WulebgrWrong. I called "rubbish" the asertion that the author of Rybka (sp?) has incorpored new algorithms to get positional evaluation. He at most, has implemented new evaluation functions, most probably only have modified some coeficients in the existing code.
Let me get this straight: you are calling "rubbish" the statement regarding Rybka's strength from promotional materials advertising the program based on your reading of a book published in 1970?
I also think that you are using the term "brute force" a bit more broadly than standard usage in the field of computer chess.
That do NOT qualifies as an algorithm modification.
BTW, the book by Botvinnik was published in the '60.
And I'm not criticizing Rybka based on that!
Please read carefully.
If you are in doubt about what I meant, I claim that ALL current commercial chess engines algorithms are based on a brute force search. They all prune the tree search using the alpha-beta algorithm, and the performance difference are only based into 2 items:
- How successfully the a-b algorithm has being implemented
- How precise (sp?) the evaluation functions are.
Some algorithms don't suffer of blindness caused by the frontier effect.
I hope I can post an example position (I don't know FEN notation), that NO ONE existing commercial engines could solve, in spite that the position is a forced win (mate) in 15 moves.
Cheers!
- J
Edit: Here is the game:
http://www.chessgames.com/perl/chessgame?gid=1031957
Up to now, I coulnd't find an engine that see 30. Ba3!!
It should not difficult for actual engines to find it. But it is curious that only applying the Botvinnik algorithm (by hand) it is easy to find the forced winning variation.
Try it in your engines and pls tell me what the results are. I'm very curious. No engine that I've tried has found 30. Ba3...
Originally posted by CrazyLilTingCrafty 20.14 finds 30. Ba3 in 266.73 seconds (696 knodes/sec), at a search depth of 15. It scores +1.34 (ie not a forced win), increased to +3.74 in 420 seconds. I'll leave it running overnight to see if anything changes. Although Crafty is not a commercial engine...
Wrong. I called "rubbish" the asertion that the author of Rybka (sp?) has incorpored new algorithms to get positional evaluation. He at most, has implemented new evaluation functions, most probably only have modified some coeficients in the existing code.
That do NOT qualifies as an algorithm modification.
BTW, the book by Botvinnik was published in ...[text shortened]... what the results are. I'm very curious. No engine that I've tried has found 30. Ba3...
Speaking as a professional programmer if you mess with the code you've modified the algorithm. The evaluation function is fairly crucial to the whole thing so changes there can alter the behaviour quite drastically, more so than getting an extra ply or two from a finesse in the search algorithm.
Originally posted by CrazyLilTingThe assertion that you called rubbish was quoted from http://store.convekta.com/shop_model.asp?gid=122&sView=Catalog or http://www.chesscentral.com/software/rybka-chess.htm or some similar source, and presumably originates with the manufacturer.
Wrong. I called "rubbish" the asertion that the author of Rybka (sp?) has incorpored new algorithms to get positional evaluation. He at most, has implemented new evaluation functions, most probably only have modified some coeficients in the existing code.
That do NOT qualifies as an algorithm modification.
BTW, the book by Botvinnik was published in ...[text shortened]... what the results are. I'm very curious. No engine that I've tried has found 30. Ba3...
Originally posted by HFRorbis
"The main developer of the program - international master Vasik Rajlich (Hungary) has inserted the algorithms of positional estimation into it which are as close to the chess-player’s style of thinking as possible. Excellent evaluation of the dynamic factors combined with subtle definition of the positional nuances (positional victim of the quality is the house method of the program!) allows it to achieve the outstanding results in playing and the quality of analysis! "
It may be that Botvinnik's book was published before 1970, but the 1970 edition was the one that came up in my search:
Computers, Chess And Long-Range Planning - (ISBN: 0387900128)
M.M. Botvinnik -
Bookseller: Mike Long
(oakland, CA, U.S.A.) Price: US$ 72.49
[Convert Currency] Shipping within U.S.A.:
US$ 5.00
[Rates & Speeds] Add Book to Shopping Basket
Book Description: Springer-Verlag Publishing -. Hardcover/pub.1970/Gd. condition (ex-lib)/89 pages - The quality is revealed in the study of the basic thought processes of master chess players. [AN830318]. Bookseller Inventory # 30318
[Bookseller & Payment Information] [More Books from this Seller] [Ask Bookseller a Question]
Fritz 9 found Ba3 after 2 seconds, changed its "mind" one after one minute, and at 1.28 was back to Ba3, now with an evaluation of +3.48
Rybka 1.0 Beta (the free version) found Ba3 in less than 30 seconds with an evaluation of +3.98
Originally posted by WulebgrSpike 1.0 also found Ba3 with +3.-- in about 30 seconds
Fritz 9 found Ba3 after 2 seconds, changed its "mind" one after one minute, and at 1.28 was back to Ba3, now with an evaluation of +3.48
Rybka 1.0 Beta (the free version) found Ba3 in less than 30 seconds with an evaluation of +3.98
Naum 1.6 required 2 minutes and 10 seconds to find Ba3 and an evaluation greater than +3.00
1 edit
Originally posted by WulebgrThanks for the info, Wulebgr.
The assertion that you called rubbish was quoted from http://store.convekta.com/shop_model.asp?gid=122&sView=Catalog [b]or http://www.chesscentral.com/software/rybka-chess.htm or some similar source, and presumably originates with the manufacturer.
Originally posted by HFRorbis
"The main developer of the program - international maste ...[text shortened]... 1.0 Beta (the free version) found Ba3 in less than 30 seconds with an evaluation of +3.98
[/b]
It is not important, but I wasn't referring to the Springer Verlag Edition that was a translation. I meant the native russian edition. But it is all unimportant, it doesn't make difference.
About the programs evaluation. I found that the position is clearly better for white (after Ba3). But I want to see a program telling me "this is a forced mate in ..."
I can't be comformed with only a puntuaction that indicates "decisive advantage", while there are a forced mate.
Keep it going Wulebgr!
Nice work. And again. thanls you very much.
- J
(My pgm continues working, lol! )
Originally posted by CrazyLilTingI left Crafty on all night and it says after 37,900 seconds that it evaluate the position at +5.57 at a ply depth of 20, now what this means is that there is no way of forcing checkmate after 20 moves. There is some "resource" you are missing which even if it doesn't save the game for black means that the checkmate isn't forced; the engines will be finding the line of play you have in mind and it'll generate a beta fail, and choosing a line of play that puts off defeat past whatever ply depth it's calculating to, since from it's point of view a score of +5.57 is better than +327 (or whatever value it gives to checkmate).
Thanks for the info, Wulebgr.
It is not important, but I wasn't referring to the Springer Verlag Edition that was a translation. I meant the native russian edition. But it is all unimportant, it doesn't make difference.
About the programs evaluation. I found that the position is clearly better for white (after Ba3). But I want to see a program telli ...[text shortened]...
Nice work. And again. thanls you very much.
- J
(My pgm continues working, lol! )
Although your general point that there are positions that the algorithms don't handle well is correct.
Originally posted by CrazyLilTingOkay. I think I'll order Botvinnik's book, as it seems interesting. Alas, as I cannot read Russian, so it will be the Springer Verlag edition.
Thanks for the info, Wulebgr.
It is not important, but I wasn't referring to the Springer Verlag Edition that was a translation. I meant the native russian edition. But it is all unimportant, it doesn't make difference.
About the programs evaluation. I found that the position is clearly better for white (after Ba3). But I want to see a program telli ...[text shortened]...
Nice work. And again. thanls you very much.
- J
(My pgm continues working, lol! )
Botvinnik's position is a good test position, as most engines initially see equality. Postmodernist 10.10 was still at +0.34 after three minutes, for example.
Forced mate in 17 (or there about) is beyond the horizon of most software, although several pruning algorithms in common use today should bring them fairly close.