Dragon 2.5 Skill levels

[lkaufman]

We got some data that suggests that skill levels in the 16 thru 25 range in Dragon 2.5 perform better than comparably rated humans would in Rapid play in terms of having their moves analyzed by a much stronger level of Stockfish (equal to Skill 32 on Dragon). The levels above and below this range may be roughly ok. But I don't know if this really translates to results against humans. So far I'm not getting feedback from users who have played the indicated Skill levels (based on FIDE rating, or if you don't have one, chess.com Rapid rating would be close enough). This may be because it is human nature to want to talk about winning and not losing. Anyway, if any of you can confirm (or refute) that the levels are too tough for you (at 15' + 10") in the 16 to 25 range, that would be helpful.

[Odd Gunnar Malin]

Hi.

I can play games against different levels. The problem is that I only have Fide rating for classical (90 +30 and longer) games, since I did start playing as adult my blitz/rapid rating is probably alot lower (missing the auto moving thing).
I have set up a bot (PolarChessBot) on LiChess to play against Dragon where I will play my games. My username on LiChess is OddGM. My current classical LiChess rating is 2115 and my Fide rating is 1714, so the table you refered to in another post isn't correct for me.
I have set my bot to only play unrated games since I change the level often and would not comprimise LiChess rating system. I put the the configuration in the chat window for each game.
I will start playing tonight, the games you find under the PolarChessBot account now are only test games.
I have my own books for the engine that are a compilation of lichess classical and rapid games in the rating range of the level. Rating range 1600-1799, 1800-1999, 2000-2199, 2200-2499 and 2500->. The book are restricted to moves played atleast 500 times (100 for the 2200 book and 50 for the 2500 book). Dragons book features will kick in for the lower range when using the 1600 book. I include blitz games in the 2500 book to get enough games.

By the way the bot are only online when I'm playing against it, if someone else would try the have to give me a notice either here or on LiChess (OddGM) with the configuration the want.

The reason I will play comp. games on LiChess is so Aimchess could get hold on it for their analysis. I play with the DGT board for classical chess so I find LiChess easier than Chess.com. I also play like this on ICC that have the best integration of the DGT board, but AimChess can't get hold on those games.

[lkaufman]

Hi.

I can play games against different levels. The problem is that I only have Fide rating for classical (90 +30 and longer) games, since I did start playing as adult my blitz/rapid rating is probably alot lower (missing the auto moving thing).
I have set up a bot (PolarChessBot) on LiChess to play against Dragon where I will play my games. My username on LiChess is OddGM. My current classical LiChess rating is 2115 and my Fide rating is 1714, so the table you refered to in another post isn't correct for me.
I have set my bot to only play unrated games since I change the level often and would not comprimise LiChess rating system. I put the the configuration in the chat window for each game.
I will start playing tonight, the games you find under the PolarChessBot account now are only test games.
I have my own books for the engine that are a compilation of lichess classical and rapid games in the rating range of the level. Rating range 1600-1799, 1800-1999, 2000-2199, 2200-2499 and 2500->. The book are restricted to moves played atleast 500 times (100 for the 2200 book and 50 for the 2500 book). Dragons book features will kick in for the lower range when using the 1600 book. I include blitz games in the 2500 book to get enough games.

By the way the bot are only online when I'm playing against it, if someone else would try the have to give me a notice either here or on LiChess (OddGM) with the configuration the want.

The reason I will play comp. games on LiChess is so Aimchess could get hold on it for their analysis. I play with the DGT board for classical chess so I find LiChess easier than Chess.com. I also play like this on ICC that have the best integration of the DGT board, but AimChess can't get hold on those games.

That should be helpful. Your LiChess Rapid rating (the relevant one as we are talking about Rapid games with Dragon) is 1993. The chart that I referenced elsewhere has 1995 LiChess Rapid = FIDE 1720, so 1993 would be about 1718, just four points above your actual FIDE rating! So I think we can consider your FIDE rating to be accurate enough for playing Rapid. So if the Skill levels were perfect on Dragon 2.5, you should be a close match (slight favorite) against Skill level 17 at 15' + 10". But if the results of matching moves with a strong Stockfish are meaningful you will lose rather badly, and may even have a tough time with level 16. We'll see. Regarding the opening book, I think that at these Skill levels Dragon will play only its first six moves from the book, so book details aren't so important.

[lkaufman]

Hi.

I can play games against different levels. The problem is that I only have Fide rating for classical (90 +30 and longer) games, since I did start playing as adult my blitz/rapid rating is probably alot lower (missing the auto moving thing).
I have set up a bot (PolarChessBot) on LiChess to play against Dragon where I will play my games. My username on LiChess is OddGM. My current classical LiChess rating is 2115 and my Fide rating is 1714, so the table you refered to in another post isn't correct for me.
I have set my bot to only play unrated games since I change the level often and would not comprimise LiChess rating system. I put the the configuration in the chat window for each game.
I will start playing tonight, the games you find under the PolarChessBot account now are only test games.
I have my own books for the engine that are a compilation of lichess classical and rapid games in the rating range of the level. Rating range 1600-1799, 1800-1999, 2000-2199, 2200-2499 and 2500->. The book are restricted to moves played atleast 500 times (100 for the 2200 book and 50 for the 2500 book). Dragons book features will kick in for the lower range when using the 1600 book. I include blitz games in the 2500 book to get enough games.

By the way the bot are only online when I'm playing against it, if someone else would try the have to give me a notice either here or on LiChess (OddGM) with the configuration the want.

The reason I will play comp. games on LiChess is so Aimchess could get hold on it for their analysis. I play with the DGT board for classical chess so I find LiChess easier than Chess.com. I also play like this on ICC that have the best integration of the DGT board, but AimChess can't get hold on those games.

That should be helpful. Your LiChess Rapid rating (the relevant one as we are talking about Rapid games with Dragon) is 1993. The chart that I referenced elsewhere has 1995 LiChess Rapid = FIDE 1720, so 1993 would be about 1718, just four points above your actual FIDE rating! So I think we can consider your FIDE rating to be accurate enough for playing Rapid. So if the Skill levels were perfect on Dragon 2.5, you should be a close match (slight favorite) against Skill level 17 at 15' + 10". But if the results of matching moves with a strong Stockfish are meaningful you will lose rather badly, and may even have a tough time with level 16. We'll see. Regarding the opening book, I think that at these Skill levels Dragon will play only its first six moves from the book, so book details aren't so important.

I see you lost the first four Rapid games with the bot, then drew, then won the final game. Unfortunately it doesn't tell us what Skill levels you were playing against. My guess based on the games and the stats is that you started at level 17 as suggested, and after losing 3 or 4 you switched to level 16 which was more balanced. How is my guessing?

[Cornfed]

By the way the bot are only online when I'm playing against it, if someone else would try the have to give me a notice either here or on LiChess (OddGM) with the configuration the want.

The reason I will play comp. games on LiChess is so Aimchess could get hold on it for their analysis. I play with the DGT board for classical chess so I find LiChess easier than Chess.com. I also play like this on ICC that have the best integration of the DGT board, but AimChess can't get hold on those games.

As I own Aimchess and have played on lichess (mostly on chess.com and ICC - just because I can get the games automatically saved in a folder on my hard drive), I wanted to ask you about this...as I really do not understand bots or the fascination with them.

So, you put a 'beatable' skill level bot of Dragon 2.5 on lichess and play against it and then get Aimchess to fetch the games and extract the statistics from them?

If so, that's interesting and I had not thought about it. Is there some advantage of doing this compared to simply playing other humans and then pulling that information into Aimchess?

Also I REALLY think Aimchess should have an option to upload games from ones slower Over The Board tournament games! I know it would not be able to deal with 'time management' but that's no big deal. It would be a great way to chart certain things about my play over...well, my past 30+ years of tournament play. Online play tends to be for 'blitz chess' mostly and therefore a bit superficial.

Also, I think they should build out the Tactics part of the program to dissect different types of tactics - to pinpoint the types of tactics one needs to get a better grip on. This COULD be done. They could do something similar for the more 'positional' aspects of the game. Granted that would be a bit harder to program.


As is, Aimchess is a great idea...but just doesn't go far enough as a self-diagnostic tool. And the "personalized lessons'....well, they leave a lot to be desired.

[Odd Gunnar Malin]

Hi.

The nice thing about aimchess is that it set up a task for you to do every day. In here among others are tactical tasks devided into a few sub task (tactic, endgame, blunder prevent and defender). There also something they call 360 Trainer wher you don't know if it is tactical or strategic. In addition there is som other tasks that is not related to your game (I think).
To create those tactical and strategical tasks they try to use as much of you games on Lichess, Chess.com and Chess24.com as they can. How they create these other tasks, I don't know. Maybe it is based on theme from your errors (would be nice if they do).
Here it comes in to get they be able to collect your games, the more games they find the more tasks are from your games.

I don't look at the statistics too often, just to get an overview now and then, but it looks nice in their advertising. It compare you with players of similar strength in different aspect of the game so I maybe could use it more and focus on it in my training. There is other thing in there that are more useful, especially the training room.

Of course it is better to play against humans than against computers, The problem is that you mostly play blitz and maybe a few rapid games. This isn't general training of your chess. Maybe could be used (as a training tool) to train openings a things like that. Else it is pure entertainment (Magnus Carlsen).
Why is this? Most is afraid the opponent use a computer, maybe.
I have played some 90+30 games on Lichess (one every sunday), and have never had any suspicion that the opponent had cheated in any of the games. I stream my games to twitch (no banter) so anybody can check for them self if they suspect me for cheating.

Here it comes the interest, not to play much against computers, but getting a better raiting estimation of engines and especially Dragon since it is what I use at home.
Why this? I try to create a training program (GUI) for me that use Dragon as an engine for playing games and need a, as good as can get, rating of different levels and peronalyties of Dragon.

In real, 15+10 games are way too fast for me, but if this is what what they need now, I give it a try and maybe later play som 90+30 games to check the difference. We will see, I guess this forum demand 1000 games to estimate the rating, so it take som times to go through all levels and personalities. I played 6 games last night, and could get some overall pinpoints of my weakness, so it helps my chess too.

I asked Aimchess about OTB games, and here is the answer (27/8):
"There are many users who are looking to import their OTB games and we are working hard on this feature, and believe we will be able to deliver it to production soon."

[Odd Gunnar Malin]

I see you lost the first four Rapid games with the bot, then drew, then won the final game. Unfortunately it doesn't tell us what Skill levels you were playing against. My guess based on the games and the stats is that you started at level 17 as suggested, and after losing 3 or 4 you switched to level 16 which was more balanced. How is my guessing?

Hi.

Hmm. I see,
The bot software ( I use the official python scripts) put a comment (hello message) in the chat room on the left side of the game when the game start. I thought other users did see the game when they click on it, but it is not so, only the involved players comes to the game when clicking on it. Other users are going directly into the analysis room. I didn't find a way to see the actual game for other users so I will for now copy the hello message into the analysis room after each game. Hopefully they who look at game live could see this game chat.
Here is the hello message for all 6 games:
"Hi! I'm PolarChessBot. Current setup: Dragon 2.5, Level 16, Personality: default, Book: Pc1600.bin"

I don't know what happend in the first four or even all six games, I was blundering as a mad man. Normaly my centipawn loss is around 30 but here it was much higher. I will try to give a more serious shot on it for the next games. And maybe reduse it to 4 or even 2 game each night. It is too much afterwork with 6 games.
But anyhow, I see a trend in my blunder so something to work on.

[Cornfed]

Hi.

The nice thing about aimchess is that it set up a task for you to do every day. In here among others are tactical tasks devided into a few sub task (tactic, endgame, blunder prevent and defender). There also something they call 360 Trainer wher you don't know if it is tactical or strategic. In addition there is som other tasks that is not related to your game (I think).
To create those tactical and strategical tasks they try to use as much of you games on Lichess, Chess.com and Chess24.com as they can. How they create these other tasks, I don't know. Maybe it is based on theme from your errors (would be nice if they do).
Here it comes in to get they be able to collect your games, the more games they find the more tasks are from your games.

I don't look at the statistics too often, just to get an overview now and then, but it looks nice in their advertising. It compare you with players of similar strength in different aspect of the game so I maybe could use it more and focus on it in my training. There is other thing in there that are more useful, especially the training room.

Of course it is better to play against humans than against computers, The problem is that you mostly play blitz and maybe a few rapid games. This isn't general training of your chess. Maybe could be used (as a training tool) to train openings a things like that. Else it is pure entertainment (Magnus Carlsen).
Why is this? Most is afraid the opponent use a computer, maybe.
I have played some 90+30 games on Lichess (one every sunday), and have never had any suspicion that the opponent had cheated in any of the games. I stream my games to twitch (no banter) so anybody can check for them self if they suspect me for cheating.

Here it comes the interest, not to play much against computers, but getting a better raiting estimation of engines and especially Dragon since it is what I use at home.
Why this? I try to create a training program (GUI) for me that use Dragon as an engine for playing games and need a, as good as can get, rating of different levels and peronalyties of Dragon.

In real, 15+10 games are way too fast for me, but if this is what what they need now, I give it a try and maybe later play som 90+30 games to check the difference. We will see, I guess this forum demand 1000 games to estimate the rating, so it take som times to go through all levels and personalities. I played 6 games last night, and could get some overall pinpoints of my weakness, so it helps my chess too.

I asked Aimchess about OTB games, and here is the answer (27/8):
"There are many users who are looking to import their OTB games and we are working hard on this feature, and believe we will be able to deliver it to production soon."

Thanks for the info. Yes, sometimes it is hard to find a good human opponent in the 2000+ range so your idea has some merit. I had not thought of it. I may give it a try but would need to get Dragon 2.5 or some other such engine with a 'tweakable rating'.

Yes, Blitz is largely 'pure entertainment'...so the ability to import/upload a pgn file of one's OTB games is largely what it is missing - and those are already played - serious chess, serious time controls. I am happy to hear they are waiting on it as I have over 1,000 OTB tournament games I could sort by time control that I would love to feed into Aimchess.

[lkaufman]

I see you lost the first four Rapid games with the bot, then drew, then won the final game. Unfortunately it doesn't tell us what Skill levels you were playing against. My guess based on the games and the stats is that you started at level 17 as suggested, and after losing 3 or 4 you switched to level 16 which was more balanced. How is my guessing?

Hi.

Hmm. I see,
The bot software ( I use the official python scripts) put a comment (hello message) in the chat room on the left side of the game when the game start. I thought other users did see the game when they click on it, but it is not so, only the involved players comes to the game when clicking on it. Other users are going directly into the analysis room. I didn't find a way to see the actual game for other users so I will for now copy the hello message into the analysis room after each game. Hopefully they who look at game live could see this game chat.
Here is the hello message for all 6 games:
"Hi! I'm PolarChessBot. Current setup: Dragon 2.5, Level 16, Personality: default, Book: Pc1600.bin"

I don't know what happend in the first four or even all six games, I was blundering as a mad man. Normaly my centipawn loss is around 30 but here it was much higher. I will try to give a more serious shot on it for the next games. And maybe reduse it to 4 or even 2 game each night. It is too much afterwork with 6 games.
But anyhow, I see a trend in my blunder so something to work on.

So these were all against level 16. I assume you set Threads to 1, is that correct? If so then it's pretty clear that the levels are stronger than intended (at least 16 and up). Even if you had a bad day, you wouldn't lose 4.5 to 1.5 to a 1600 player if you are over 1700 FIDE (and estimated FIDE from your Rapid rating). My own games with Dragon also suggest that the levels are stronger than intended around 21 and 22. But the low levels, like 1 thru 13, seem to be no stronger than the indicated ratings, as far as I can tell. I can easily beat level 13 giving it queen odds; I don't think I can expect to defeat a 1300 FIDE rating human at queen odds in Rapid. Somehow the steps from 13 to 16 must be too large.

[Odd Gunnar Malin]

So these were all against level 16. I assume you set Threads to 1, is that correct? If so then it's pretty clear that the levels are stronger than intended (at least 16 and up). Even if you had a bad day, you wouldn't lose 4.5 to 1.5 to a 1600 player if you are over 1700 FIDE (and estimated FIDE from your Rapid rating). My own games with Dragon also suggest that the levels are stronger than intended around 21 and 22. But the low levels, like 1 thru 13, seem to be no stronger than the indicated ratings, as far as I can tell. I can easily beat level 13 giving it queen odds; I don't think I can expect to defeat a 1300 FIDE rating human at queen odds in Rapid. Somehow the steps from 13 to 16 must be too large.

All setting except these are default:
uci_options: # Arbitrary UCI options passed to the engine.
Log File: "dragon.log" # Logging
Book File: "C:/Engines/LiChessBot/engines/pc1600.bin" # Book to be used 1600,1800,2000,2200 or 2500
Skill: 16 # Skill * 100 = elo

I will play some more games before I conclude. Looking at the games there are some bad blunders. Mate in 2 after +6 eval. Mixing moveorder in one game, and other stupid blunders. I have hope to play a litle better when I get used to this timecontrol. And also play more games against human opponent with this timecontrol to get my rating more established.