11 Sep 05
Originally posted by XanthosNZGood answer; done any ACTUAL analysis of GM games or just running suck as usual?
No1 may shout the loudest but that doesn't mean he actually has all the answers. Raw engine matchup only tells part of the story. The real test lies in distinctive engine moves and matchup in non-tactical, many possible move positions.
11 Sep 05
Originally posted by dottewellI ran some representative games from Capablanca, Alekhine, Fischer and Kasparov OTB games and the average matchups to Fritz8's first move was 65-75%. These are widely considered among the four greatest players of all time. If someone can locate some GM correspondence games, I'll be happy to run them. My own experience suggests that matchups raise 5-10% from timed games to RHP, but I'm not a GM and GM time controls are generally far more generous than those you can get on sites like FICS, Playchess, etc.
I don't want to bring this topic up again, really, but it might be useful for those of us who don't understand the issue.
What kind of percentage match with major commercial engines do you find in a typical GM game?
12 Sep 05
Originally posted by no1marauderThis just in, Kasparov didn't use an engine.
Good answer; done any ACTUAL analysis of GM games or just running suck as usual?
My method requires maybe a hundred games rather than one. Give me a year and a well-known GM (recent is better) and I'll give you a rundown of my analysis. I wouldn't want to go choosing my own sample now would I?
Posting the whole analysis would be impossible. The raw output would be even more formidable than yours.
12 Sep 05
Originally posted by XanthosNZOf what possible use is a "method" that you admit can't be done in any reasonable time? Some people would prefer that those usernames who are matching up to an engine 85-90%+ on a regular basis be tossed sometime before the next milennium, which is when your research project will be apparently completed.
This just in, Kasparov didn't use an engine.
My method requires maybe a hundred games rather than one. Give me a year and a well-known GM (recent is better) and I'll give you a rundown of my analysis. I wouldn't want to go choosing my own sample now would I?
Posting the whole analysis would be impossible. The raw output would be even more formidable than yours.
12 Sep 05
Originally posted by no1marauderWhere did I admit that it can't be done in a reasonable time?
Of what possible use is a "method" that you admit can't be done in any reasonable time? Some people would prefer that those usernames who are matching up to an engine 85-90%+ on a regular basis be tossed sometime before the next milennium, which is when your research project will be apparently completed.
I guess I wasn't clear with my 'give me a year comment', that was meant as give me a year from which to take games of a player from (say 1995).
I could complete the whole lot in maybe a few hours work plus a day of computer usage.
Raw matchup can still be used to determine targets for my method, which I agree is more time and labour intensive than just plugging a game into Fritz and then going off to play hide the pickle.
12 Sep 05
Originally posted by XanthosNZI misunderstood your "give me a year" statement. I like Shirov, so how about Shirov 1997? I'm still curious to see how these results would lead to some specific guidelines for determining engine use; even a 100 games from a GM is going to leave you with a limited database for positions to analyze. Perhaps I'm not understanding what exactly you are claiming can be accomplished vis-a-vis this method.
Where did I admit that it can't be done in a reasonable time?
I guess I wasn't clear with my 'give me a year comment', that was meant as give me a year from which to take games of a player from (say 1995).
I could complete the whole lot in maybe a few hours work plus a day of computer usage.
Raw matchup can still be used to determine targets ...[text shortened]... abour intensive than just plugging a game into Fritz and then going off to play hide the pickle.
12 Sep 05
Originally posted by no1marauderI'm not sure what can be gained from this study, but I think that the year of choice would have to be a very recent one, given the fast advancement of chess engines and the fact that GMs use chess engines a lot during training.
I like Shirov, so how about Shirov 1997?
I think, if any year is going to be of use, it would have to be 2004.
D
12 Sep 05
1 edit
Shirov 1997 sounds good to me, if possible. At the very least if the match up is far lower than some of the players here that should suggest either (a) Shirov 1997 would beat Fritz (or whatever) by finding moves a top commercial engine could not (unlikely), or (b) players here who match more than "Shirov 97" of the time would have beaten Shirov in 1997. Which would be pretty staggering, even at correspondence chess.
12 Sep 05
Originally posted by RagnorakI completely fail to see the relevance of either of your "points". What does the "fast advancement" of chess engines have to do with it; Xanthos will be using a commercially available engine from this year. And as to GM's using chess engines during training; what of it? As I understand it, he is trying to finding specific patterns where engines always make a certain move but humans don't. You're not seriously suggesting that GM's try to make their games just like chess engines are you??
I'm not sure what can be gained from this study, but I think that the year of choice would have to be a very recent one, given the fast advancement of chess engines and the fact that GMs use chess engines a lot during training.
I think, if any year is going to be of use, it would have to be 2004.
D
Shirov is a very aggressive, tactical player so I would think he would match up higher than most GMs. That, of course, is not terribly relevant to what Xanthos is trying to show. He gave an example of 1995 and I picked a more recent year than that. I'll be curious to see what his results are, assuming he has time to do the project before I leave.
12 Sep 05
Originally posted by no1marauderDo you have to try to be this thick, or does it come naturally?
I completely fail to see the relevance of either of your "points". What does the "fast advancement" of chess engines have to do with it; Xanthos will be using a commercially available engine from this year. And as to GM's using chess engines during training; what of it? As I understand it, he is trying to finding specific patterns where engines alwa ...[text shortened]... l be curious to see what his results are, assuming he has time to do the project before I leave.
I can't see how you honestly don't see the relevancy of my points. Read the Northern Lad's post about one of his games with IM from here...
http://www.timeforchess.com/board/showthread.php?threadid=28690&page=2
Northern Lad: 'Engines will every now and then come up with moves a human being would hardly consider. Some times they are very strong. An example of this is my game against Ironman (810608 - unfinshed though I'm objectively losing), in which he refutes a line of the Schliemann I had played numerous times OTB (including against strong players who had prepared against me) with the ugly (but very strong) computer move 11.Qe3 with the unpleasant point that the natural 11...Rf7 runs into 12.Qf3 threatening both Qxf5 and Bc4. Typical computer move!'
Are you telling me that Northern Lad won't now add this game to his database, and possibly use this ugly, very strong, typical computer move??? I reckon he will if he runs into this line again. Now, he's just after matching up with a very computer/ very non-human move. Should this be counted as a match?
Since GMs play the game professionally, spending hours a day practicing (possibly against engines), are you telling me that people who have trained with the more recent engines won't have found more of these little nuggets for their repitoire: ugly, strong, very computer moves.
Any reasonable study on GM matchup with engines would have to take this into account.
D