Originally posted by RagnorakIF the moves are in a database they will be omitted from Xanthos' analysis. So an "innovation" that makes it into a database won't mean jack.
Do you have to try to be this thick, or does it come naturally?
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 huma ...[text shortened]... s.
Any reasonable study on GM matchup with engines would have to take this into account.
D
We all know what your game is: you are attempting to totally discredit engine analysis as a means of detecting cheaters. Since there is virtually no other way absent spyware of some sort, in effect you want to make it so cheaters can use engines. I wish you would just come out and say so, rather than engaging in a disinformation campaign.
To Xanthos: I've given my player and year and look forward to your results, even if Ragnorak has pre-decided they will be worthless.
I was running through one game (Fischer-Stein 1967 Swiss Interzonal) and since i was reading about all theste stats, decided to track correspondence with Fritz 8's first selected moves. Fischer achieved a 55%. He was the winner. Stein, the loser, achieved a 60% score compared to Fritz. Someone would have to check hundreds of games and compare to Fritz to get any reliable conclusion. But, just based on this one game, i'd say anybody close to a 100% correlation to Fritz is suspicious. The more games this happens, the more suspicious. You can never be, statistically, 100% certain anybody is using an engine, but probablility is a powerful force. Second, even tho Stein was more "accurate" (according to Fritz) he lost the game. What does this mean? That Fischer could beat Fritz? That Fritz (playing with 100😵 would beat Fischer in a match? That certain "turning point moves" need to be more accurate (weighted) than others in order to win a game. Someone well versed in statistical theory needs to tackle this, as it's way beyond me. As for evidence of engine cheats. I liken it to evidence of plagiarism. Sure, somebody, could come up with a couple of lines of, say, Virginia Woolf, on their own, but if it's a couple of pages, I'd say plagiarism is pretty well assured.
I would love to find a collection of correspondence games from 1900-2000 rated players like IM was in 1993 in PGN format and run them through Fritz. Anybody know where I can find such a collection? It would have to be from a correspondence federation that does not allow engine use, but that's most of them except the ICCF.
Originally posted by no1marauder2 to go...
I would love to find a collection of correspondence games from 1900-2000 rated players like IM was in 1993 in PGN format and run them through Fritz. Anybody know where I can find such a collection? It would have to be from a correspondence federation that does not allow engine use, but that's most of them except the ICCF.
Originally posted by no1marauderAnd you will know 11. Qe3 is in, for example, Northern Lad's database how exactly?
IF the moves are in a database they will be omitted from Xanthos' analysis. So an "innovation" that makes it into a database won't mean jack.
We all know what your game is: you are attempting to totally discredit engine analysis as a means of detecting cheaters. Since there is virtually no other way absent spyware of some sort, in effect y ...[text shortened]... year and look forward to your results, even if Ragnorak has pre-decided they will be worthless.
As for my game here... I was interested in Xanthos' study, and felt like contributing to try to make it more worthwhile. Be small minded and aggressive the rest of your life, I couldn't give two hoots.
Xanthos: to make the study better, I'd recommend doing the analysis on 2Bit's year, if you feel like taking orders from the a__, AND also do your analysis on the same player for 2004. I would guesstimate an increase in matches of about 10%+
D
Originally posted by RagnorakIs your guess based on any empirical evidence?
And you will know 11. Qe3 is in, for example, Northern Lad's database how exactly?
As for my game here... I was interested in Xanthos' study, and felt like contributing to try to make it more worthwhile. Be small minded and aggressive the rest of your life, I couldn't give two hoots.
Xanthos: to make the study better, I'd recommend doing the an ...[text shortened]... ysis on the same player for 2004. I would guesstimate an increase in matches of about 10%+
D
Here are some of the games I analysed. All from Linares 14th:
Nikolic – Shirov:
Moves tested:
25. ... b3
30. ... Rh3+
Fritz suggests:
25. ... Qh2+ (b3 is the second choice)
30. Nb6 (puts Rh3+ at least 0.3 pawns worse and not in the top 5)
Shirov – Kramnik:
This game finishes with a lot of only moves which are no use.
Move tested:
24. g3
Fritz suggests:
24.Qc6 (g3 is the second choice 0.04 behind, Bd3 is given as an improvement by Shirov in his analysis, this comes further down Fritz's choices)
Anand – Shirov:
This game illustrates one difficulty of analyzing older games. Shirov's move 8. Be6 was a novelty when he played it. However my GM game database has it played 21 times as it contains many games more recent than this one.
Moves tested:
11. ... 0-0-0
20. ... Bxf5
Fritz suggests:
11. ... 0-0-0 or Bd4 both scoring the same.
20.... Bxf5 and Be3+ both scoring the same.
Shirov – Kasparov:
Again we have database issues. 14. ... e6! was a novelty when it was played, however now it is the most popular move. Also in this analysis I was looking at annotated games. I tried to avoid moves where there was a clearly better move in the annotation, we want how close Shirov plays to a computer and analyzing his mistakes doesn't help that.
Moves tested:
24. Rfe1
30. Be3
Fritz suggests:
24. Nde2 (Rfe1 doesn't appear in its top 5)
30.Re3 (This one is a little confusing, Fritz says this is a clear pawn better than Be3 its second choice, yet no annotation is present in a well annotated game. Perhaps Fritz is missing a continuation. I'll look deeper later)
Gelfand – Shirov:
This game has an excellent queen sac for perpetual that Fritz can't see.
Moves tested:
16. ... Bb7
17. ... Rg8
37. ... Bc5
Fritz suggests:
16. ... e5 (Bb7 doesn't appear in the top 5)
17. ... Bc6 (Rg8 appears 3rd 0.2 behind)
37. ... h5 (Bc5 appears as 5th choice)
Producing this output takes a lot of time so I only did it for 5 games.
Overall I looked at 50 games and an average of two moves per game. I picked games by filtering Shirov playing in 1997 and used the first 50.
I get a first suggested move match of 18%. If second choice matchs count as well I get 41%.
Any questions?
Originally posted by XanthosNZYes. Of what possible use is it? As I understand it, hardly ever does even one of the top GM's in the world match to Fritz in your designated, critical positions (how exactly did you decide on those moves?). So if some clown on a correspondence site is matching up 60% in these situations, is he a cheat?
Here are some of the games I analysed. All from Linares 14th:
[b]Nikolic – Shirov:
Moves tested:
25. ... b3
30. ... Rh3+
Fritz suggests:
25. ... Qh2+ (b3 is the second choice)
30. Nb6 (puts Rh3+ at least 0.3 pawns worse and not in the top 5)
Shirov – Kramnik:
This game finishes with a lot of only moves which are no use.
...[text shortened]... suggested move match of 18%. If second choice matchs count as well I get 41%.
Any questions?[/b]
Originally posted by no1marauderI decided on the moves by looking for positions where there were a large number of possible moves with no real outright number 1 move. This is what takes the time.
Yes. Of what possible use is it? As I understand it, hardly ever does even one of the top GM's in the world match to Fritz in your designated, critical positions (how exactly did you decide on those moves?). So if some clown on a correspondence site is matching up 60% in these situations, is he a cheat?
The fact that top GMs hardly matchup at all under my system is good. It means that strong players need not be afraid that a high matchup will do them in. It also means that you don't need a 90% matchup for it to mean anything.
Unlike your straight move matchup there is no flat cutoff. I'm not going to say that above 68% is a cheater. The match level must be taken together with whether the player finds non engine plans (see the last game I selected) and whether he plays distinctive engine moves (moves that no human with a brain would play but an engine would). Obviously I don't have an example from Shirov's games but I'll dig through my engine games (from playchess) and find some.
Originally posted by XanthosNZIts obvious that the "points" I raised which 2Bit dismissed earlier on are having an adverse effect on your study.
Here are some of the games I analysed. All from Linares 14th:
[b]Nikolic – Shirov:
Moves tested:
25. ... b3
30. ... Rh3+
Fritz suggests:
25. ... Qh2+ (b3 is the second choice)
30. Nb6 (puts Rh3+ at least 0.3 pawns worse and not in the top 5)
Shirov – Kramnik:
This game finishes with a lot of only moves which are no use.
...[text shortened]... suggested move match of 18%. If second choice matchs count as well I get 41%.
Any questions?[/b]
Is that guy still playing?
Also, I think it'd be very interesting to see how GM's match up when given the extra time and tools which are part and parcel of the game that we play on this site; correspondence chess.
You should be able to find plenty of pgns here, if you're interested...
http://gamesarchive.iccf.com/archive_worldtournaments.shtml
D
Originally posted by RagnorakYes Shirov is still playing, the latest I have is the 14th of August.
Its obvious that the "points" I raised which 2Bit dismissed earlier on are having an adverse effect on your study.
Is that guy still playing?
Also, I think it'd be very interesting to see how GM's match up when given the extra time and tools which are part and parcel of the game that we play on this site; correspondence chess.
You should be a ...[text shortened]... here, if you're interested...
http://gamesarchive.iccf.com/archive_worldtournaments.shtml
D
I didn't miss your suggestion about recent games and you'll notice I point out some extra difficulties in analysing games that are 8 years old. I did my analysis on the year and player No1 provided as this way it can't be claimed that my sample was biased.
I will be doing some up to the minute Shirov games to see if your +10% theory holds true. My suspicion is that it won't as I tried not to use moves from home analysis which is where engine analysis is centered.
I will also use the site you linked to as a source of some correspondance games to analyse. I had some I was planning on using but not a huge number. Thanks for that.
Cheers, Xanthos. Whether it "proves" anything is in some senses irrelevant; I think it will at least provide some food for thought for lesser players who assume a very strong player will invariably match many/most/virtually all the moves of a top engine.
I guess cheat detection at the top level is an art as well as a science.