3 edits
Originally posted by no1marauderI'm tiresome! What about you? You constantly misrepresent me.
You're tiresome. You imply things then quickly deny you meant them. You constantly lie about what I have said. Here is a typical DF lie:
but you are also including move C and D, E and F when the difference in evaluation suits you.
No, I am not and you know it. I specifically stated what my methods were; you won't find any nalyzed well over a dozen of this suspect's games. The results are all consistent.
I have simply never changed my stance and I'm not about to now.
I am not suggesting nothing be done. I have made logical suggestions on how they be selected, I have no fault with the analysis you suggest then narrowing the field down to the most likely candidates but am merely suggesting that those candidates then be looked at in more detail for "typical engine moves". I have not defined those "typical engine moves" but gave some examples of what I considered could fit into that category but when David Tebb gave a reason why some of my suggested "typical engine moves" could be made by a human player I immediately acknowledged that he could indeed be correct. There are, however, other "typical engine moves" that will help confirm your analysis. I don't believe these would be that hard to find when the candidates have been narrowed down.
I don't recollect ever saying humans don't blunder. Please can you tell me where I said this. Of course humans blunder but strong humans do not blunder in every game nor even once in every 20 games. I have in fact said that, in my opinion, over a sufficiently large sample the absence of blunders is highly indicitive. However you need caution even here as if (for example) we define a blunder as a difference of 3 pawns a move that changes the assessement from +8.00 to +4.00 would be a blunder but this could be done deliberately as it doesn't affect the result of the game at all, whereas a move that changes the assessment from +0.50 to -0.50 is almost certainly a blunder.
The individual you refer to may very well be an engine user, and in fact probably is and that is not in dispute but over a long period (before the mods were disbanded) they clearly did not agree with you so must have been lacking some sort of evidence.
You imply you know exactly how the mods think but you don't as you are not a mod although I concede you may have a reasonable idea.
My only concern is that your method is too blunt an instrument. Relying on match ups alone may very well identify lots of cheats but how many innocent people could it possibly also identify. If that is 5% is that acceptable? 10%? 15%? You may have successfully reported many cheaters to the mods and for that I commend you but can you advise how many "cheaters" you reported who were not subsequently banned and what percentage of those you reported do they represent?
We don't disagree on the fundamentals, not even on this match up idea. I don't know why you persist in misunderstanding and misrepresenting me.
Originally posted by no1marauderThis is very supicious.
Rittner - Kaurenen
Fritz 10 at 30 seconds at 512MB
37 non-database moves
1st: 18
2nd: 6
3rd: 2
Total: 26/37 or 70%
11 non-matches of which 9 didn't match any of Fritz's first five recommendations.
OR
...[text shortened]... .10 equivalents
36/38 or 95%
Quite a difference.
I have never said or implied otherwise.
Originally posted by no1marauderInteresting. I'll take a look later (at work).
Rittner - Kaurenen
Fritz 10 at 30 seconds at 512MB
37 non-database moves
1st: 18
2nd: 6
3rd: 2
Total: 26/37 or 70%
11 non-matches of which 9 didn't match any of Fritz's first five recommendations.
OR
...[text shortened]... .10 equivalents
36/38 or 95%
Quite a difference.
Could you PM me the game id of your suspect game(s) - thanks.
1 edit
Originally posted by Dragon FireDF: Relying on match ups alone may very well identify lots of cheats but how many innocent people could it possibly also identify.
I'm tiresome! What about you? You constantly misrepresent me.
I have simply never changed my stance and I'm not about to now.
I am not suggesting nothing be done. I have made logical suggestions on how they be selected, I have no fault with the analysis you suggest then narrowing the field down to the most likely candidates but am merely suggesting dea. I don't know why you persist in misunderstanding and misrepresenting me.
None. No "innocent person" is going to match up with an engine's recommendations in game after game at 85-90-or more %.
A lot of analysis and study has went into this. There are a certain amount of forced or obvious moves in each game. I long ago studied hundreds of my own games and reported stats of match ups with players at different levels; I seem to remember 1200-1300 being somewhere around 30-40%; 1500's being in the 50% range, etc. Let us say that a strong human player could be expected to match up on RHP about 75% to one of Fritz's first three choices (I think Gate gave that number somewhere as well). So that's a 1 in four chance that a strong player, on average, won't match up with any of Fritz's top three recommendations.
Picture this: You have an ordinary deck of cards; your chance of drawing something beside a spade is 75%. Assume that the cards are reshuffled after one card is exposed.
Now draw a card 38 times in a row. What are the chances you will not draw a single spade? Very low, perhaps infinitesimally low. Repeat this experiment again, perhaps drawing a card 30 times; say this time you only draw 3 spades when the expected result would be 7.5. Say you do this over and over and you never draw spades in any one run at a 25% rate but over hundreds of draws you average 10% of spades being drawn. Wouldn't you conclude that something was affecting the shuffling of the deck to make it so you wouldn't get spades?
Obviously, the "spades" are non-match ups with an engine; the "runs" are each game of the suspect and the thing affecting the outcome is that the suspect is "looking at the cards".
Perhaps someone with a better knowledge of standard deviation calculations can explain it better than I can. The point is that we would expect to see a range of outcomes in a subject's games that varied considerably. I have games where I match up in the 30% and I have games (short tactical ones) where I match up in the 90s%. That's what will happen with a human. That isn't what happens with an engine cheat.
EDIT: I realize I have been a little too harsh in this thread in dealings with some of the other posters. I apologize for this.
Originally posted by no1marauderI think that you're explanation had all the essentials and was well made. The only thing really missing was some actual probability calculations but you did a good job explaining things without having to resort to that.
Perhaps someone with a better knowledge of standard deviation calculations can explain it better than I can. The point is that we would expect to see a range of outcomes in a subject's games that varied considerably. I have games where I match up in the 30% and I have games (short tactical ones) where I match up in the 90s%. That's what will happen with ...[text shortened]... le too harsh in this thread in dealings with some of the other posters. I apologize for this.
EDIT: I realize I have been a little too harsh in this thread in dealings with some of the other posters. I apologize for this.
And congrats on this stance of yours.
I posted this on the 4th of July 2005:
From my 640 games, opponents percentage matches at different levels on this site (book opening moves not included):
-1200 31%
1201-1300 33%
1301-1400 39%
1401-1500 42%
1501-1600 48%
1601-1700 51%
1701-1800 61%
1801-1900 63%
1901-2000 69%
2001-2100 77%
2100+ 79%
Two caveats: This was there average rating when they played their games against me (rating at start + rating at finish/2); some players who I played early in their RHP careers were only 1400-1500's but wound up being 2000+. Second, obviously the top brackets (and some of the others) are skewed by cheaters and probably significantly overstate a true "average" matchup. One of the top players on the site I played a couple of games with who is clearly not an engine user, matched up 43/59 or 73%. For the record, I match up 64%.
http://www.timeforchess.com/board/showthread.php?threadid=25853&page=5
Those results are using "blunder check" set at 0 but disregarding any moves not a first recommendation above .04 centipawns' difference. There have been some concerns expressed that a "blunder check" would yield higher matchup rates. But surely I was being too conservative in regarding .05 centipawn differences as non-matchups. I suspect these two factors basically cancel out and the numbers are roughly useful guides.
Originally posted by no1marauderBlunder check can be biased towards winning positions it has seen previously. But your opponents are more often than not on the losing side (you win most of your games), so blunder checking for them will be biased towards avoiding the actual game continuation, and hence biased towards reducing match ups for them - not increasing them.
I posted this on the 4th of July 2005:
From my 640 games, opponents percentage matches at different levels on this site (book opening moves not included):
-1200 31%
1201-1300 33%
1301-1400 39%
1401-1500 42%
1501-1600 48%
1601-1700 51%
1701-1800 61%
1801-1900 63%
1901-2000 69%
2001-2100 77%
2100+ 79%
Two cavea ...[text shortened]... uspect these two factors basically cancel out and the numbers are roughly useful guides.
Combined with your 0.04 threshold (rather than 0.1), both factors reduce the match up rate – not cancel each other out. By how much I don’t know.
But maybe you lost more against the higher rated players and therefore maybe those values are less affected?!
Originally posted by no1marauderIt seems to me that you are selecting here match ups within .05 centipawns rather than the 1st 3 engine moves presumably after giving an engine a fixed time to think about each move.
I posted this on the 4th of July 2005:
From my 640 games, opponents percentage matches at different levels on this site (book opening moves not included):
-1200 31%
1201-1300 33%
1301-1400 39%
1401-1500 42%
1501-1600 48%
1601-1700 51%
1701-1800 61%
1801-1900 63%
1901-2000 69%
2001-2100 77%
2100+ 79%
Two cavea ...[text shortened]... uspect these two factors basically cancel out and the numbers are roughly useful guides.
If that is the case I have absolutely no problem with this. My problem was with always counting as a match the 1st 3 moves and possibly more.
I also agree that there does not seem to be anything significantly wrong with the % you quote provided the sample of games chosen is sufficiently large and typical, i.e. you should not select only the games that a player has won or only games against certain categories of opponents. It is hopefully pretty obvious that any blunders will occur in games lost and most players tend to raise there game against stronger opposition. I would suggest a moving average of say 30 consecutive games would be a reasonable way to do this. This would identify trends and point to circumstances when computers are / are not used. You would expect such a moving average to be pretty steady.
Looking at the figures you give I would say a reasonable deviation is +/- 5% so a moving average moving outside that range should ring alarm bells.
Just for the record, although I no longer use it, I DO have a tertiary qualification in Statistics so am fully familiar with these concepts and understand the concept of standard deviations as well as the need to get an appropriate sample. If used carelessly statistics can be used to prove anything.
Originally posted by no1marauderSo can I take it that every person you have reported as a possible cheat has been positively identified as being a cheater and banned from the site.
DF: Relying on match ups alone may very well identify lots of cheats but how many innocent people could it possibly also identify.
None. No "innocent person" is going to match up with an engine's recommendations in game after game at 85-90-or more %.
If not what is the approximate % of those reported who have not been banned?
Originally posted by Dragon FireNo1 has said he's reported names that were obvious cheats in his mind, but not removed by the admins.
So can I take it that every person you have reported as a possible cheat has been positively identified as being a cheater and banned from the site.
If not what is the approximate % of those reported who have not been banned?
P-
Originally posted by PhlabibitExactly! So by implication someone does not agree that his method is fool proof. Somewhere there is doubt.
No1 has said he's reported names that were obvious cheats in his mind, but not removed by the admins.
P-
I personally doubt know what is causing that doubt except that one persons obvious cheat is clearly not so obvious to another!
Perhaps I should also ask what percentage of the reported cheats were found to be cheaters by the game mods and what percentage of those were banned. This will be interesting for us all to know.
Originally posted by Dragon Fireit's been more of a problem with the site owners refusing from acting based on clear evidence, than no1 being wrong. I can't remember a single case where there were any reasonable doubt of the guilt.
Exactly! So by implication someone does not agree that his method is fool proof. Somewhere there is doubt.
I personally doubt know what is causing that doubt except that one persons obvious cheat is clearly not so obvious to another!
Perhaps I should also ask what percentage of the reported cheats were found to be cheaters by the game mods and what percentage of those were banned. This will be interesting for us all to know.
as far as I can say, his track record has been phenomenal.
Originally posted by Dragon FireNo1 is much better at spotting cheats than the admin who hits the kill button. The game mods may even still be up and running right now were he on the team.
Exactly! So by implication someone does not agree that his method is fool proof. Somewhere there is doubt.
I personally doubt know what is causing that doubt except that one persons obvious cheat is clearly not so obvious to another!
Perhaps I should also ask what percentage of the reported cheats were found to be cheaters by the game mods and what percentage of those were banned. This will be interesting for us all to know.
P-
Originally posted by Katastroof"Excuse me waiter. This beef is dry, the potatoes are cold, my salad is brown and wilted. For the record, I'm not criticizing this meal or the chef".
Why of course!How can he be wrong?You guys worked out a system you think is sound,he applies it,you accept the results as true.That's always going to be a 100% score.
For the record,I'm not criticising the system nor no1marauder.
P-