Fritz 0 wrote: ↑Thu Jun 16, 2022 10:19 am
Dragon at 2805 Elo should move instantly. You are doing something wrong if it takes 2-3 seconds per move.
Whenever you set Komodo Dragon Elo = to whatever , most of the times it moves every second, but at certain point it takes 2 to 3 seconds, therefore, I only post the maximum amount of time. Anybody here on this forum can take a chronometer and measure how many seconds it take for Komodo Dragon to move, as average I would say 2 seconds. I noticed that when Komodo Dragon time control is down to 20 seconds, it moves instantly, like a fraction of a second per move specially during the endgame
Note: in Most game regardless of the T/C that you give to the opponent engine Komodo Dragon for the entire game only take around 60 to 100 seconds depending how many moves. There was a game with 257 moves that it took Komodo Dragon only around 4 minutes to finish the entire game and Shredder8 the entire T/C of 15/10 which is about 17 minutes total with only 10 seconds left on the clock when they both reached a draw position. Before you make any comments why don't you set Komodo Dragon Elo = whatever against another engine and since most of the time Komodo Dragon does not predict the opponent next move it will take at least 1 to 2 seconds to respond, whereas, if it is Komodo Dragon playing versus itself then the moves are played within fraction of a seconds since Dragon already knows the best answer according to the calculation done by the same engine
I have to agree with "Fritz 0" here; with a setting of Elo 2805 (or anywhere in the human range), Komodo Dragon (any version with Elo settings) should move "instantly", meaning in less than a tenth of a second, and I confirmed this on my own laptop with a quick game against it myself. I used the Fritz GUI, and it never took any time detectible to me for any move, so my guess is that there is some issue with the specific GUI you are using or else with the computer you are using. It should never take anywhere near even one second for any move under these conditions. This is with Threads = 1. If you used many threads, it would take more time, but still should be under half a second per move consistently. There were some issues with certain GUIs not handling the Elo properly in 2.6.1 so we reverted to the 2.6 method in Dragon 3, but apparently it is using the Elo or else it would take similar time to the opponent. In test games with GM Lenderman, with a setting of 2600 it averaged only about a quarter of a second per GAME (not move!); at 2800 it might be about one second per GAME. But each GUI has its own overhead so it won't be this fast on all GUIs.
Any plans to introduce a UCI_Sleep feature in future versions similar to what MByrne did with Honey?
The idea is to make it "sleep" for a variable time before moving when on weakened settings to avoid the jarring (and very unrealistic for training) insta-response every time you move against it.
Wouldn't that give the human pondering time and make Dragon relatively weaker than intended at a certain Elo setting?
Does it matter? The human SHOULD have pondering time. Just as they do in a real game with a real opponent (which is what the bloody thing is supposed to be simulating!).
A real 2200 player as an example, isn't blitzing his moves out at breakneck speed.
So if Dragon is REALLY replicating a 2200 opponent, there should also be "pondering" time included.
I am just saying that with included pondering time Dragon Elo values would have to be recalibrated.
Personally, I am not a fun of engines' artificial delay. I have better things to do with my time than to wait for the engine which is doing nothing.
Well, you're probably more of a "computer chess" guy (tinkerer) than a "chess" guy. That's fine.
Most serious chess players (as in, tournament players) if trying to replicate a human classical chess encounter with an engine, want it to be as true to life as possible.
Just as playing a REAL tournament game takes time, so does a training game.
Whether you "have better things to do" or not.
I have seen several CHESS players here who have shared my sentiment in the past, hence my query.
Fritz 0 wrote: ↑Thu Jun 16, 2022 10:19 am
Dragon at 2805 Elo should move instantly. You are doing something wrong if it takes 2-3 seconds per move.
Whenever you set Komodo Dragon Elo = to whatever , most of the times it moves every second, but at certain point it takes 2 to 3 seconds, therefore, I only post the maximum amount of time. Anybody here on this forum can take a chronometer and measure how many seconds it take for Komodo Dragon to move, as average I would say 2 seconds. I noticed that when Komodo Dragon time control is down to 20 seconds, it moves instantly, like a fraction of a second per move specially during the endgame
Note: in Most game regardless of the T/C that you give to the opponent engine Komodo Dragon for the entire game only take around 60 to 100 seconds depending how many moves. There was a game with 257 moves that it took Komodo Dragon only around 4 minutes to finish the entire game and Shredder8 the entire T/C of 15/10 which is about 17 minutes total with only 10 seconds left on the clock when they both reached a draw position. Before you make any comments why don't you set Komodo Dragon Elo = whatever against another engine and since most of the time Komodo Dragon does not predict the opponent next move it will take at least 1 to 2 seconds to respond, whereas, if it is Komodo Dragon playing versus itself then the moves are played within fraction of a seconds since Dragon already knows the best answer according to the calculation done by the same engine
I have to agree with "Fritz 0" here; with a setting of Elo 2805 (or anywhere in the human range), Komodo Dragon (any version with Elo settings) should move "instantly", meaning in less than a tenth of a second, and I confirmed this on my own laptop with a quick game against it myself. I used the Fritz GUI, and it never took any time detectible to me for any move, so my guess is that there is some issue with the specific GUI you are using or else with the computer you are using. It should never take anywhere near even one second for any move under these conditions. This is with Threads = 1. If you used many threads, it would take more time, but still should be under half a second per move consistently. There were some issues with certain GUIs not handling the Elo properly in 2.6.1 so we reverted to the 2.6 method in Dragon 3, but apparently it is using the Elo or else it would take similar time to the opponent. In test games with GM Lenderman, with a setting of 2600 it averaged only about a quarter of a second per GAME (not move!); at 2800 it might be about one second per GAME. But each GUI has its own overhead so it won't be this fast on all GUIs.
Any plans to introduce a UCI_Sleep feature in future versions similar to what MByrne did with Honey?
The idea is to make it "sleep" for a variable time before moving when on weakened settings to avoid the jarring (and very unrealistic for training) insta-response every time you move against it.
Wouldn't that give the human pondering time and make Dragon relatively weaker than intended at a certain Elo setting?
Does it matter? The human SHOULD have pondering time. Just as they do in a real game with a real opponent (which is what the bloody thing is supposed to be simulating!).
A real 2200 player as an example, isn't blitzing his moves out at breakneck speed.
So if Dragon is REALLY replicating a 2200 opponent, there should also be "pondering" time included.
I am just saying that with included pondering time Dragon Elo values would have to be recalibrated.
Personally, I am not a fun of engines' artificial delay. I have better things to do with my time than to wait for the engine which is doing nothing.
I agree with you here, although there is something to be said for a tiny (like one second) delay in moving, or for making it something the user can set. We might do one of these things, but the rating would only be accurate with no or minimal delay. We're not going to test it against humans on an equal time use basis, as it would take twice as long to test, and because I don't believe that most users would want more than a small delay. I think that the bots on chess.com use something like a one second delay, which presumably was found to be most pleasant for users.
