Zygalski wrote on 09/23/10 at 17:25:59:
Markovich wrote on 09/23/10 at 16:55:40:
Zygalski wrote on 09/23/10 at 15:47:52:
Smyslov:
Yes you're quite right, in a single or a few games the match up rates can be very high indeed. Unasissted players simply cannot seem to sustain this engine-like play in many games over time though.
I've yet to see benchmarks from any player which exceed the 60/75/85% for more than 20 games with 500+ non-database moves. Another member of chess.com has also generated many top-quality player benchmarks & had virtually identical findings to mine.
If I were to analyse an unasissted player & find much higher match rates than the thresholds in 20 objectively chosen games with 500+ non-database moves, I would wonder if yes, some of the lower matching players who've been banned may be innocent!
60/75/85% for top 1/2/3 (all 3 figures=/+) and I would submit & maybe expect a ban, maybe not.
65/80/90% (all 3 figures=/+) and I would expect that player to be banned.
Start with a sample set that includes all available corr games...
What- from all years?
Even ones where computers/engines have no doubt been used. What will that prove exactly?
I'm not spending the next 10 years generating useless benchmarks for someone who has no understanding of what they're talking about & is simply an internet troll.
Sorry!
No do you see, I said "pre-computer?"
Markovich wrote on 09/23/10 at 16:55:40:
Start with a sample set that includes all available corr games and only games played pre-computer by players of some minimum strength. Do not include any OTB games. Calculate matchup percentages for both players. Be careful to apply an objective criterion of what constitutes a move elligible for consideration, and apply the same criterion to each move. Be careful to use the same engine and engine settings across all games.
As to whether I know what I am talking about, I have repeatedly said that not enough of chess.com's method's is public to enable anyone to know if they know what they are doing. So to that extent, no, I don't know about these methods.
But I modestly do assert that I understand something about statistics; I buttered my bread pretty well with it for 30 years, anyway. I have not seen from you or from chess.com's other defenders here anything resembling a appreciation for the statistical aspects of this problem, which are fundamental. Instead I have just seen people pointing at numbers and saying "See?"
I don't think it would take 10 years to develop an estimate of the distribution of computer mathup percentage, following the methods I outlined. I think it would take about two months. But someone would have to write a Python script so as to interract with the engine and the game set, as well possibly with a db so as to decide which moves are worthy of being considered. Or you could just take all moves between the 20th and the appearance of less than seven pieces, as a pretty good shortcut, I would think.
In any case, that a proper analysis would consume time is hardly an argument against it. A proper analysis would take too much time, so let us accuse people of cheating based on an improper analysis?
@drogo: I don't think it is especially idiotic for some well-meaning people to have tried to draw conclusions without a proper statistical analysis. I do think it's idiotic to persist in this error, having been informed of it. I don't live in a world surrounded by idiots, just in a world where those defending chess.com here appear to be idiots.
Pages: 
Yelena Dembo on Chess.com (Read 99705 times)
