purechess wrote: ↑Mon Jan 24, 2022 5:17 pm
Ok 2.6.1 introduced a UCI_Elo bug for some GUI's which ignore UCI_Elo when it's set to 0 which is the default setting?!
Please fix it.
Maybe I don't understand the problem. If setting elo to 0 really just sets it to default 3500 (which is full strength), it just means that the range of elos that work is 1 to 3500, with 0 meaning "off". That seems perfectly fine if that's the case, just set elo to 1 for minimum strength, which would only be appropriate if you are having it play a six year old kid who just learned the moves of the pieces (roughly speaking). If I misunderstand the problem, please explain.
But if Komodo is trying to emulate FIDE rated players, it should start with a minimum of 1000, since the lowest FIDE rating that can be achieved is 1000, because of the rating floor in FIDE's rating system, 1000 is the lowest attainable rating. A handful of players have this rating, most of them kids and the Maximum 2950 which I doubt that Carlsen ever will achieve in his lifetime
Yes, FIDE 1000 is the minimum, but it is not supposed to mean beginner level. Chess.com ratings are in general reasonably close to FIDE ratings (below GM level anyway), give or take a couple hundred elo, but the AVERAGE rating there is something like 1200, with the minimum close to zero. My guess is that the average adult who admits to knowing how to play chess would not be able to earn a FIDE rating of 1000 given the appropriate opportunity. So just consider 1000 and above as intended to match FIDE, and ratings below that more or less comparable to chess.com or perhaps USCF scholastic ratings. At the high end, remember these are Rapid ratings, so assuming that 2850 is correct for Rapid, Carlsen would need to set the Elo to something like 3000 for an even challenge at Classical chess, and if he wants to play a somewhat stronger opponent at classical against whom he would still have good drawing chances when playing White, a setting of 3200 might be appropriate. Or to simulate how Carlsen would play at one move per day correspondence chess, maybe 3400 or so is about right. So there are valid reasons for the entire range of Elo settings, although for most of our customers 1000 to 2700 would be sufficient.
If the majority of your customers are rated between 1000 to 2700, then it should be the target or even from 800 to 2700.
Note: From Iceland you should be able to find the ideal Komodo Dragon 2.6.x Next Knight Odds opponent, since the range of Fide rated players makes them ideal. Probably some of the best 8 players from Iceland https://ratings.fide.com/rankings.phtml?country=ISL
purechess wrote: ↑Mon Jan 24, 2022 5:17 pm
Ok 2.6.1 introduced a UCI_Elo bug for some GUI's which ignore UCI_Elo when it's set to 0 which is the default setting?!
Please fix it.
Maybe I don't understand the problem. If setting elo to 0 really just sets it to default 3500 (which is full strength), it just means that the range of elos that work is 1 to 3500, with 0 meaning "off". That seems perfectly fine if that's the case, just set elo to 1 for minimum strength, which would only be appropriate if you are having it play a six year old kid who just learned the moves of the pieces (roughly speaking). If I misunderstand the problem, please explain.
But if Komodo is trying to emulate FIDE rated players, it should start with a minimum of 1000, since the lowest FIDE rating that can be achieved is 1000, because of the rating floor in FIDE's rating system, 1000 is the lowest attainable rating. A handful of players have this rating, most of them kids and the Maximum 2950 which I doubt that Carlsen ever will achieve in his lifetime
Yes, FIDE 1000 is the minimum, but it is not supposed to mean beginner level. Chess.com ratings are in general reasonably close to FIDE ratings (below GM level anyway), give or take a couple hundred elo, but the AVERAGE rating there is something like 1200, with the minimum close to zero. My guess is that the average adult who admits to knowing how to play chess would not be able to earn a FIDE rating of 1000 given the appropriate opportunity. So just consider 1000 and above as intended to match FIDE, and ratings below that more or less comparable to chess.com or perhaps USCF scholastic ratings. At the high end, remember these are Rapid ratings, so assuming that 2850 is correct for Rapid, Carlsen would need to set the Elo to something like 3000 for an even challenge at Classical chess, and if he wants to play a somewhat stronger opponent at classical against whom he would still have good drawing chances when playing White, a setting of 3200 might be appropriate. Or to simulate how Carlsen would play at one move per day correspondence chess, maybe 3400 or so is about right. So there are valid reasons for the entire range of Elo settings, although for most of our customers 1000 to 2700 would be sufficient.
If the majority of your customers are rated between 1000 to 2700, then it should be the target or even from 800 to 2700.
Note: From Iceland you should be able to find the ideal Komodo Dragon 2.6.x Next Knight Odds opponent, since the range of Fide rated players makes them ideal. Probably some of the best 8 players from Iceland https://ratings.fide.com/rankings.phtml?country=ISL
The target range is 1000 to 2700, that's the range at which I hope we have parity with FIDE. But there is no disadvantage to offering the range 1 to 3500 for novice players or those who want to simulate Carlsen at various time controls, just don't take the Elo too seriously outside that range. As for players for future Knight odds matches, the main criteria aside from appropriate elo and GM title are 1. Experience with streaming while playing on chess.com, which guarantees something people can watch with minimum likelihood of technical problems, and also pretty much rules out cheating, since a streamer would not want to be called a cheater 2. Good personality, meaning someone people like and who will not make unreasonable excuses for a bad result or misrepresent a good result. I think this applies to most if not all of our opponents in recent years. Nationality doesn't matter as long as they are fluent in English (well actually our first knight odds GM match was in French, so even that isn't an absolute).
Quote from Elo: "The rating scale itself - its range of numbers - is, like any scale without reproducible fixed points, necessarily an open-ended floating scale."
Then he says that you should arrange this scale so no rating ever goes negative.
One of the problem with Fide and all national 'Elo' lists are rapid improvers like youngsters. I don't know if this is one of the thing a floor could help, but there is a problem with floors too, you can't measure improvement among beginners. A chessprogram for personal use don't need to use floor at all and should measure progress of play from the very beginner too.
Fide's floor was once 2200 I believe and are going down evenly.
I don't think the majority of chess players are above 1000, the majority that plays in club is, but I would guess 90% of pure home players are below 1000 in strength.
Ozymandias wrote: ↑Tue Jan 25, 2022 9:06 am
Curiously enough, a search at https://ratings.fide.com/ gives me a result of 179 players with an Elo of 1001, no one with 1000.
Yes
It seems that there is a bug in the rating list.
The rules say
"The rating must be at least 1000." so 1000 should be an option but I guess the program that calculates the rating has a bug so players who get exactly 1000 drop out of the list.
Here is how the rating was among Norwegian club players before we discarded this and only use Fide rating. I think we had a hidden floor at 400 then.
US still use a national list with a low floor (200?), maybe someone with access could look it up.
Odd Gunnar Malin wrote: ↑Tue Jan 25, 2022 10:09 amOne of the problem with Fide and all national 'Elo' lists are rapid improvers like youngsters. I don't know if this is one of the thing a floor could help, but there is a problem with floors too, you can't measure improvement among beginners. A chessprogram for personal use don't need to use floor at all and should measure progress of play from the very beginner too.
Fide's floor was once 2200 I believe and are going down evenly.
I don't think the majority of chess players are above 1000, the majority that plays in club is, but I would guess 90% of pure home players are below 1000 in strength.
The problem with "home players" is: they stay at home. Some (most?) don't even play online. They like the game, may have played it at some point, but never really got into it. What's their level? How many of them are there?
When I started in the 90's, I recall the floor was at 1800, and that still worked OK. A decade later, ratings where still consistent. Now the whole system has been thrown out of whack with constantly fluctuating "entities". Players who start at 1000 and improve hundreds of points in a matter of months. I'm sure Arpad didn't have such players in mind when devising the formula, and don't even know if one can be written to properly deal with them.
Uri can say something a little more informed on the matter (he finished high in the kaggle competition 11 years ago).
Odd Gunnar Malin wrote: ↑Tue Jan 25, 2022 10:39 am
Here is how the rating was among Norwegian club players before we discarded this and only use Fide rating. I think we had a hidden floor at 400 then.
US still use a national list with a low floor (200?), maybe someone with access could look it up.
Using only Fide rating did not cause many players in norway to lose their rating and have no rating?
Does it mean that many chess tournaments between children in norway do not go for rating because nobody has a fide rating?
Odd Gunnar Malin wrote: ↑Tue Jan 25, 2022 10:09 am
Quote from Elo: "The rating scale itself - its range of numbers - is, like any scale without reproducible fixed points, necessarily an open-ended floating scale."
Then he says that you should arrange this scale so no rating ever goes negative.
One of the problem with Fide and all national 'Elo' lists are rapid improvers like youngsters. I don't know if this is one of the thing a floor could help, but there is a problem with floors too, you can't measure improvement among beginners. A chessprogram for personal use don't need to use floor at all and should measure progress of play from the very beginner too.
Fide's floor was once 2200 I believe and are going down evenly.
I don't think the majority of chess players are above 1000, the majority that plays in club is, but I would guess 90% of pure home players are below 1000 in strength.
I am not sure about the majority that plays in club and I think that it is dependent on the country.
I guess that the majority of players with USCF rating (that are not home players) are below 1000 fide rating in strength but I do not live there so I am not sure.
In israel it is not clear(the lowest league has many children who are below 1000 fide rating in strength but they may not be a majority of club players)
I am also not sure about the strength of 1000 fide players and basically the rating list at the low level is not fair in the meaning that often the stronger player is not the player with the higher rating.
I believe that the strength difference between the average player with fide rating 1001 and the average player with fide rating 1400 is clearly smaller than the average difference between the average player with fide rating 2001 and the average player with fide rating 2400.
Some people on Infinity server have had timeouts with Komodo 2.6.1, where engine stopped responding after 40-50 moves and will run out of time. In all cases they had AMD 32-64 cores machine. Also on playchess server some have reported a bug where engine will come up with exception error and game will be lost by engine resigning in the middle of the game. In all cases users had AMD machines. Do Komodo team think this engine bug is related to AMD machines ?
