17 Jul 08
2 edits
Originally posted by !~TONY~!Tony, I share your approach.
Ah that's why. I didn't get to use Analyzer 1 that much because of my laptop explosion, and when I got back Analyzer 2 was already in use. And to think my brilliant Ro # was already the Gt #! Damn it!
I think that more mathematics, etc. need to be applied. I like the sound of your Ro# but I need to ponder on it some more but I make up my mind on the extent of its usefulness. I’m not aware of what ideas have already been tried in practice. But I give the game mods, etc. the benefit of the doubt and respect them as intelligent individuals.
I’ve only thought about cheat detection in recent weeks, so I’m still gathering my thoughts… they may change…
- I don’t see much value in move rankings. I believe that paying attention to evaluations is more useful and makes any move ranking info redundant.
- I feel that lack of automation tools is forcing people to use compromised methods (again, I can’t speak of things unknown to me).
- I want to experiment with match ups using e.g. Rybka at a fixed ply of 1, 2, etc. My curiousity is this: is there any correlation between depth of search and human ability to find the best moves? If, e.g., is it feasible to assume a human can play like an engine limited to 1 ply? 2 ply? etc, then we can use that to remove all moves in a game which can be matched at the feasibly low ply level? After other types of removal too, then we could examine the match up for the remaining moves, i.e. those which required the engine to run at higher depths of search. I accept that this experimentation may turn out to be ineffective.
- I also want to experiment with match up for different phases of the game. For example, what are the chances that straight out of theory, someone is following analysis from a book or engine analysis they constructed earlier? Or, how do match ups vary in the middlegame compared to the endgame? Again, the answers may be useful or not.
- does a choice of opening or style of player affect match ups? For example, I’ve read that Rybka doesn’t understand the compensation in the Benko Gambit relatively well. How should match ups vary for open or closed positions?
- presently, I’m overwhelmed with the amount of factors/variables that can influence move match ups. Can we really simplify this to saying “if it’s greater than X% then it’s suspicious”? Analysing lots of games would help balance some of these factors, but what if a narrow opening repertoire is usually involved? Or a player has a frequent tendency to avoid endgames or not? etc.
- I acknowledge that some players like greenpawn believe that the cheating problem is being blown out of proportion. I’d just add that I’m not motivated by how big a problem it is or not, but just out of interest for mathematicals, automation, engines, etc. It’s an interesting but difficult problem. I’m not on a misson! 🙂
17 Jul 08
Originally posted by !~TONY~!he's like a kid who just found a list of logical fallacies in the web, went to bash some half-wit creationists with it, and now thinks he can actually debate. but the sad truth is that simply spilling jargon is a poor substitute for content. it's the pretentious version of sticking fingers in your ears and screaming lalalalala.
You are single handedly ruining this entire forum for those who actually want to talk about the subject at hand, instead of posting to engorge themselves at the thought of seeing their own posts in an internet forum. You still have not pointed out why my original post you attacked is at all incorrect or illogical. I also don't see why a game that is all the ...[text shortened]... analysis. If a child such as yourself could find the moves, why couldn't another human being?
I don't know why anybody bothers to read his brainfarts anymore. I simply skip all his posts.
17 Jul 08
Originally posted by KeplerBut we should not confuse “forced” with “obvious”.
I went with Robin Smith's suggestion that when analysing using an engine a move should be considered forced or clearly best in a positional game if it is the first choice and its score is greater than the second choice score by at least 1 pawn
Some positions require a very narrow and accurate way to win. e.g. in some endgames there is a long series of single winning moves, plus many sacrificial attacks require an exact followup. But such play may not be obvious to a human. An engine when given long enough will typically have a better chance of finding such “only moves”. So can we automatically remove them on the basis of a simple criteria (2nd move evaluation is significantly lower)…. I don’t think so. Too much removal would benefit a cheater.
Consider Smith’s suggestion when someone is cheating by using tablebases to play out a complex RP vs R endgame. We’d be disregarding all the “only moves”, which there may be many of. But are they obvious to a human? The 2nd part stills needs attention.
17 Jul 08
1 edit
Originally posted by Dragon Fire(Hits head against wall)
I have a problem with this couple of 100 moves.
It could constitute only 4 games, maybe less.
I could guarantee that I could find 4 games here for virtually any player with an 85% match up if I looked but another 4 games from the same player may have a 25% match up.
The problem is the selection of games. Say you decide to select the wins only. T raid I just do not know.
This is where I have a concern about "manipulation of statistics"
I get the feeling that you don't listen to a single thing I say. I'll try one more time to make things clear to you:
1) I said, and I meant, "representative" games. NO, that's not only wins; NO, that's not only wins against higher rated players. NO, I didn't say you could run games and then disregard the results and only use the suspect's "best games.
2) "Representative games" would mean a number of games with the approximate same percentage of wins, losses and draws. The one caveat that I have is that I don't want short games that might have been quickly won by obvious tactics (that would mean higher matchups, so this gives the suspect the benefit of the doubt). I like games with at least 20 non-database moves, preferably more. I like to have equal numbers of games with Black and White. Other things may apply too that I can't think of off the top of my head.
Your suggestion that a minimum of 30 games be run is unrealistic. As I and others have pointed out, to run the games requires at least an hour, often more. THEN you want the Game Mods to go over every single move in an attempt to find "engine moves" (which, of course, a centaur won't play). Say at least another half hour or so per game for this. THEN he has to report back to the other Mods and then the admins. So we're talking 50-60 hours of work on EACH CHEAT. By part time, unpaid volunteers!
How many cheaters will EVER be banned using your proposed method, DF?
17 Jul 08
2 edits
Originally posted by no1marauderActually game mods have used program (so called Analyser) which allows to analyse large amount of games without too much exertion. With that Analyser analysing of 30 games would take maximum 1-2 days. So suggestion about 30 games is not so unrealistic. Searching of engine like moves in these 30 games might take 3-4 hours at maximum. Only problem is - if your computer is slow then analyser may make him more slower which may create problems to run another programs.
(Hits head against wall)
I get the feeling that you don't listen to a single thing I say. I'll try one more time to make things clear to you:
1) I said, and I meant, "representative" games. NO, that's not only wins; NO, that's not only wins against higher rated players. NO, I didn't say you could run games and then disre How many cheaters will EVER be banned using your proposed method, DF?
17 Jul 08
Originally posted by wormwoodMe too. Though we agreed on many points, now he took a run at me over something as simple as checking for forced moves.
he's like a kid who just found a list of logical fallacies in the web, went to bash some half-wit creationists with it, and now thinks he can actually debate. but the sad truth is that simply spilling jargon is a poor substitute for content. it's the pretentious version of sticking fingers in your ears and screaming lalalalala.
I don't know why anybody bothers to read his brainfarts anymore. I simply skip all his posts.
Even if they are ignored or counted, it doesn't really change the percentage that I see unless you were to count them in the total of evaluated moves but not count them as engine.
You only count evaluated moves, so either way you still get the same percentage and counting forced moves is a simple waste of time. (unless someone can really show me otherwise. As I said, if I came up with a huge match number, I might look into forced moves, but certainly not if I plan to run 15 or 20 games of any given user.)
P-
17 Jul 08
Originally posted by KorchYou see any benefit to finding forced moves in a game? I have not.
Actually game mods have used program (so called Analyser) which allows to analyse large amount of games without too much exertion. With that Analyser analysing of 30 games would take maximum 1-2 days. So suggestion about 30 games is not so unrealistic. Searching of engine like moves in these 30 games might take 3-4 hours at maximum. Only problem is - if your c ...[text shortened]... s slow then analyser may make him more slower which may create problems to run another programs.
P-
17 Jul 08
Originally posted by KorchNice. I really wish I had that toy. But I thought that Gate said you still have to be looking at the game move by move while doing the analysis?
Actually game mods have used program (so called Analyser) which allows to analyse large amount of games without too much exertion. With that Analyser analysing of 30 games would take maximum 1-2 days. So suggestion about 30 games is not so unrealistic. Searching of engine like moves in these 30 games might take 3-4 hours at maximum.
You do agree, however, that a centaur i.e. a pretty good player using an engine for guidance isn't going to make obvious engine moves, so that type of cheat (the strongest kind) isn't going to be caught by DF's methods. In fact, a 1,2,3 analysis would probably be best.
17 Jul 08
1 edit
Originally posted by no1marauderIt depends how high is that matchup - in my opinion there are degree of matchup which does not makes any doubts. On the other hand - there might be a cases when matchup is higher than average but not so high to be for 100% sure. Then some additional evidence is needed. And engine-like moves may be used as one of the possible ways to get additional evidence.
Nice. I really wish I had that toy. But I thought that Gate said you still have to be looking at the game move by move while doing the analysis?
You do agree, however, that a centaur i.e. a pretty good player using an engine for guidance isn't going to make obvious engine moves, so that type of cheat (the strongest kind) isn't going to be caught by DF's methods. In fact, a 1,2,3 analysis would probably be best.
17 Jul 08
1 edit
Originally posted by VarenkaI don't think my engine running on my pathetically slow machine for 30 seconds is going to find the pay off to a long range tactic. It simply will not go deep enough. I was actually trying to point out the need for care in such situations since it is possible to massage the figures to produce the desired result. It is also possible to apply the wrong criteria through ignorance of the true situation.
But we should not confuse “forced” with “obvious”.
Some positions require a very narrow and accurate way to win. e.g. in some endgames there is a long series of single winning moves, plus many sacrificial attacks require an exact followup. But such play may not be obvious to a human. An engine when given long enough will typicall ...[text shortened]... ch there may be many of. But are they obvious to a human? The 2nd part stills needs attention.
My thoughts on the matter of cheat detection using the method described are as follows. The method I used to examine Berliner's games is good enough to search for suspicious levels of engine match ups provided there are sufficient long enough (I am leaving "sufficient" and "long enough" for someone else to define) games to analyse. This process could be automated quite easily. However, the evidence of this process, no matter how high the match up rate produced, should not be taken as sufficient to condemn a person as a cheat. I found that I learned a lot from just watching Berliner's games unfold while the engine went through its analysis. I think that a certain number of the games analysed by automatic means should be examined by good human players before a decision is reached on guilt or innocence.
17 Jul 08
Originally posted by PhlabibitI don't either. Say that you have a game where a player matches up 28 out of 30 or 93%. Say there are 5 "forced" moves; he matches up 23 out of 25 or 92% if they are disregarded. What's the difference? If you disregard the forced moves in Rittner's or whatever CC player you are using as a comparative base, you'd have to drop a few percentage points from the base anyway. It's a waste of time if the sample is large enough and representative enough and if the sample isn't large enough and/or representative enough the data is useless anyway.
You see any benefit to finding forced moves in a game? I have not.
P-
17 Jul 08
Originally posted by KorchI agree wholeheartedly with this post.
It depends how high is that matchup - in my opinion there are degree of matchup which does not makes any doubts. On the other hand - there might be a cases when matchup is higher than average but not so high to be for 100% sure. Then some additional evidence is needed. And engine-like moves may be used as one of the possible ways to get additional evidence.