Originally posted by diskamylThe strength of the engine should not be a factor. If engine games are truly distinguishable from human games even a weak engine should be distinguishable. If not, then anyone using a weak engine to cheat would be undetectable using current methods. I suspect even a weak engine would be sufficiently strong to get to the upper echelons on this site.
yes but they only included in the list because it was the one Kasparov had played. It's a very old version. engines were a lot weaker back then.
Originally posted by diskamylIsn't Fritz the most human-like engine rather than rybka? Those percentages tell you nothing if the games were positional GM-style games, since engines are good in tactical games. If the games are games with endings, the endgame tablebases and so on kicks in resulting in lower matchups.
here's a list of otb GM matchups with rybka 2.3.2 at 14 ply (it should be longer than 30 seconds) for the first choice of the engine, taken from http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?pid=103791
it's very interesting to see morphy have a higher matchup than kasparov, but this could have to do with the selection of games:
Shirov 59,0 ...[text shortened]... "best moves" by Rybka 2.3.2 at 14 Ply Forwards, from more than 1400 positions from 34 games.
On the other hand given Chessmaster's ability to play like Capablanca, a clever programmer could program his engine to play like Fischer or Kasparov and then people will claim engine Z plays like Kasparov or Fischer.
Originally posted by wormwoodI would think that the match up rates from pre computer players would be artificially higher than matchups should be now...
well, anyway you decide to go, a good sanity check is to do the exact same test for the greats, like capablanca, and see how they do. that way, whatever parameters your specific system has, the threshold will be adjusted accordingly.
best would be to run through a number of pre-computer era CC champs. but of the otb players I believe capablanca has had the highest matchup rate.
at least it should give you a some kind of a rule of thumb.
If you were writing a chess program, wouldn't you try to make it play like the great players of the past? The older players might not use an engine, but an engine might use them?
Originally posted by KeplerI see your point now, I agree with that.
The strength of the engine should not be a factor. If engine games are truly distinguishable from human games even a weak engine should be distinguishable. If not, then anyone using a weak engine to cheat would be undetectable using current methods. I suspect even a weak engine would be sufficiently strong to get to the upper echelons on this site.
Originally posted by KeplerWord! I suspect it's being done, even as we speak. Running an antique edition of CM, say, on a fast chip could get you 2100-rated moves, at least in the middle game, plus a low match-up, at least on first choices, with the newer, stronger software the mods test against.
The strength of the engine should not be a factor. If engine games are truly distinguishable from human games even a weak engine should be distinguishable. If not, then anyone using a weak engine to cheat would be undetectable using current methods. I suspect even a weak engine would be sufficiently strong to get to the upper echelons on this site.
Originally posted by KeplerThe strength of engines compared with each other depends on what is happening at the end of their search trees. The main differences between engines are to do with their evaluation functions and how they implement various tricks at the leaf nodes. I think you are right that for most tactical situations the main effect comes from the 10 - 14 ply depth search, the extra stuff at the nodes is by and large only going to be important when one engine cheat is playing against another.
The strength of the engine should not be a factor. If engine games are truly distinguishable from human games even a weak engine should be distinguishable. If not, then anyone using a weak engine to cheat would be undetectable using current methods. I suspect even a weak engine would be sufficiently strong to get to the upper echelons on this site.
I think as far as detection goes most of the difference in engine output is what order they put their top few moves in, so it don´t think different engine settings and so forth will make that much difference as far as detecting cheats goes.
Originally posted by DeepThoughtI have been trying various times for the analysis ranging from 5 seconds per move to 1 minute per move. Although time has an effect on match ups it is not predictable. For instance, one game I analysed with an engine called Sigma produces the following first choice match up percentages:
The strength of engines compared with each other depends on what is happening at the end of their search trees. The main differences between engines are to do with their evaluation functions and how they implement various tricks at the leaf nodes. I think you are right that for most tactical situations the main effect comes from the 10 - 14 ply depth ...[text shortened]... ent engine settings and so forth will make that much difference as far as detecting cheats goes.
10s 75%
20s 66%
30s 80%
40s 70%
50s 73%
60s 65%
Other engines produce completely different patterns. It appears to me that the combination of time per move and engine is critical. That is why I asked earlier in this thread for information on the reasoning behind the 30 seconds per move that is generallyaccepted as the standard here. So far I have had no answer so will have to just assume it is an arbitrary choice.
Originally posted by KeplerI´d imagine that the 30s per move figure is practical. If you have 25 games to analyse that´ll be somewhere between 500 and 1,000 non-opening moves to crunch. Increasing the depth of an engine search by 1 ply increases the time to calculate by a factor of around 8, so the next step up after 30s is about 5 minutes per move, They probably found that 30s per move gives the engine enough time to hit a reasonable search depth and for the chance that the engine will change it´s mind if you give it longer to be acceptable, without making the analysis take forever.
I have been trying various times for the analysis ranging from 5 seconds per move to 1 minute per move. Although time has an effect on match ups it is not predictable. For instance, one game I analysed with an engine called Sigma produces the following first choice match up percentages:
10s 75%
20s 66%
30s 80%
40s 70%
50s 73%
60s 65%
Other en ...[text shortened]... tandard here. So far I have had no answer so will have to just assume it is an arbitrary choice.
Originally posted by DeepThoughtI am thinking that the question I should be asking is not why 30s but rather what search depth the 30s is based on. Anyone got any ideas?
I´d imagine that the 30s per move figure is practical. If you have 25 games to analyse that´ll be somewhere between 500 and 1,000 non-opening moves to crunch. Increasing the depth of an engine search by 1 ply increases the time to calculate by a factor of around 8, so the next step up after 30s is about 5 minutes per move, They probably found that 30s ...[text shortened]... ange it´s mind if you give it longer to be acceptable, without making the analysis take forever.
Originally posted by KeplerIf strongest legit players have been analysed with the same time then there is no legit reason why the suspect has higher matchup.
I have been trying various times for the analysis ranging from 5 seconds per move to 1 minute per move. Although time has an effect on match ups it is not predictable. For instance, one game I analysed with an engine called Sigma produces the following first choice match up percentages:
10s 75%
20s 66%
30s 80%
40s 70%
50s 73%
60s 65%
Other en ...[text shortened]... tandard here. So far I have had no answer so will have to just assume it is an arbitrary choice.
Originally posted by KeplerIn the thread on who the most creative player is Thread 103777 someone posted this game of Tebb´s Game 617581, Tebb won with a nice build up against black´s weak f-pawn.
I am thinking that the question I should be asking is not why 30s but rather what search depth the 30s is based on. Anyone got any ideas?
I went through it with Crafty on an old Semperon based single core 1.5 GHz machine, just seeing how deep its search got to after 30s or so. On most moves Crafty was getting to 12 ply quickly but didn´t normally get to 13 ply. On a few moves it found it harder and only just got to 12 ply in the 30s, on several the search would get to 14 or 15 ply.
In some positions the search tree is more amenable to pruning than others, so ply depth isn´t the right thing to be looking at. Using a modern quad core machine you´d expect the search to be 4 times faster and get 1 or 2 ply deeper.
Originally posted by KorchSuspect? What suspect? The list of match up rates was for differing times on a single game to show that results vary wildly (and unpredictably) according to time per move used. I know both players in that game were engines, it was game between Rybka and Shredder from the 16th World Computer Chess Championship.
If strongest legit players have been analysed with the same time then there is no legit reason why the suspect has higher matchup.
1 edit
Originally posted by KeplerI repeat - analysis of strongest legit players has been made using the same time per move as suspect. And if you still did not understand - statistics is made from more than single game.
Suspect? What suspect? The list of match up rates was for differing times on a single game to show that results vary wildly (and unpredictably) according to time per move used. I know both players in that game were engines, it was game between Rybka and Shredder from the 16th World Computer Chess Championship.