Dragon 3.1 Released at KomodoChess.com

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

CornfedForever
Posts: 648
Joined: Mon Jun 20, 2022 4:08 am
Full name: Brian D. Smith

Re: Dragon 3.1 Released at KomodoChess.com

Post by CornfedForever »

lkaufman wrote: Sun Aug 07, 2022 6:12 am
The first move is a big deal in chess. I would say the "c" pawn is the most neutral one to remove; its removal aids the queen's development (and arguably also the knight's development, since its natural square c6 won't block the "c" pawn if it is not there!), but it does split the remaining pawns into two groups, so I would say it's very close to a pure one pawn loss. So averaging removing the "c2" pawn and the "c7" pawn evals (from the point of view of the superior side) is the most appropriate figure to compare to the ideal value of 1.00 (ideal by the definition in UCI). For Dragon 3.1 on 16 threads searching 30 ply, I get 1.58 and 2.18, averaging 1.88 advantage. For a very recent Stockfish (July 24) doing same, I get 1.68 and 2.74, averaging 2.21. So even dividing by 2 still puts the eval on the high side of the desired 1.0, but close enough, while for Dragon taking 60% would make the evals similar to SF evals divided by 2.
Lastest SF Dev gives over 2.0 (c7 removed)...not as fancy a machine as yours but deep enough on 2 threads.

The first move is a big deal...but when you are trying to find an otherwise 'equal' position, removal of 1 pawn from the starting position is clearly the most logical as if you did it in the middle of any game, there are all sorts of other variables to throw into the mix. Even then, pick any pawn and it's always going to be far more than +1 for the benefiting side...and it should be.

Either way, going back to Eduards comment, I really cannot help but think removing a single pawn from most any normal position is likely to show the evaluation for the benefiting side go up '1 point'. Neural Networks 'feel things' simple human eval cannot in any case. I would think for anyone to try and make an argument that removing a pawn should = +1 for the benefiting side, would need to go back to hand crafted evals...bad ones (likely) that cannot see all the extra edge one actually gets from that extra pawn.

But then, chess is hard and those who make these evaluation complaints seem to be (more often than not) people who largely switch off their brains and push engines around with an eye on the numbers vs others doing the same thinking it is chess instead of a modern version of rockem/sockem robots or...whatever one wants to call what they do.

Don't think I have any more to say on this point...and congrats on the continued progress of Dragon! I hope to jump back in once you have made it possible to actually play it where you can set the engine to not 'auto move'...that's too disconcerting and spoils the experience- especially when one starts to get lower on time.
cc2150dx
Posts: 390
Joined: Sat Nov 30, 2013 9:51 am
Location: Canada
Full name: Jason Coombs

Re: Dragon 3.1 Released at KomodoChess.com

Post by cc2150dx »

"Dragon 3.1 fixes some issues in Dragon 3 (rare time losses in standard mode and crashes in MCTS mode), adds a new feature Skill Time (adjustible minimum time per move when using the LimitStrength (Elo) option), and with a new net and search improvements is around 10 to 20 elo stronger depending on type of book used and time control."
CornfedForever
Posts: 648
Joined: Mon Jun 20, 2022 4:08 am
Full name: Brian D. Smith

Re: Dragon 3.1 Released at KomodoChess.com

Post by CornfedForever »

cc2150dx wrote: Sun Aug 07, 2022 11:22 pm "Dragon 3.1 fixes some issues in Dragon 3 (rare time losses in standard mode and crashes in MCTS mode), adds a new feature Skill Time (adjustible minimum time per move when using the LimitStrength (Elo) option), and with a new net and search improvements is around 10 to 20 elo stronger depending on type of book used and time control."
Ah! So you can set a skill level and set the engine to move no fewer than (lets say) 10 sec after you move?!!!?
lkaufman
Posts: 6229
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA
Full name: Larry Kaufman

Re: Dragon 3.1 Released at KomodoChess.com

Post by lkaufman »

CornfedForever wrote: Mon Aug 08, 2022 2:33 am
cc2150dx wrote: Sun Aug 07, 2022 11:22 pm "Dragon 3.1 fixes some issues in Dragon 3 (rare time losses in standard mode and crashes in MCTS mode), adds a new feature Skill Time (adjustible minimum time per move when using the LimitStrength (Elo) option), and with a new net and search improvements is around 10 to 20 elo stronger depending on type of book used and time control."
Ah! So you can set a skill level and set the engine to move no fewer than (lets say) 10 sec after you move?!!!?
Yes, but the max was supposed to be 10 seconds, but somehow it won't take more than 4 seconds even when set to 10. So we should fix it in the future, but four seconds max is probably enough delay for most users, though certainly not all. Personally I prefer just a one second delay, which is what chess.com uses for their bots.
Komodo rules!
Fritz 0
Posts: 150
Joined: Fri Mar 11, 2022 12:10 pm
Full name: Branislav Đošić

Re: Dragon 3.1 Released at KomodoChess.com

Post by Fritz 0 »

lkaufman wrote: Mon Aug 08, 2022 7:13 am
CornfedForever wrote: Mon Aug 08, 2022 2:33 am
cc2150dx wrote: Sun Aug 07, 2022 11:22 pm "Dragon 3.1 fixes some issues in Dragon 3 (rare time losses in standard mode and crashes in MCTS mode), adds a new feature Skill Time (adjustible minimum time per move when using the LimitStrength (Elo) option), and with a new net and search improvements is around 10 to 20 elo stronger depending on type of book used and time control."
Ah! So you can set a skill level and set the engine to move no fewer than (lets say) 10 sec after you move?!!!?
Yes, but the max was supposed to be 10 seconds, but somehow it won't take more than 4 seconds even when set to 10. So we should fix it in the future, but four seconds max is probably enough delay for most users, though certainly not all. Personally I prefer just a one second delay, which is what chess.com uses for their bots.
Is it supposed to play a bit stronger at a certain Elo setting with this 4 seconds delay, because of a little pondering time it allows, or it is irrelevant?
lkaufman
Posts: 6229
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA
Full name: Larry Kaufman

Re: Dragon 3.1 Released at KomodoChess.com

Post by lkaufman »

Fritz 0 wrote: Mon Aug 08, 2022 9:14 am
lkaufman wrote: Mon Aug 08, 2022 7:13 am
CornfedForever wrote: Mon Aug 08, 2022 2:33 am
cc2150dx wrote: Sun Aug 07, 2022 11:22 pm "Dragon 3.1 fixes some issues in Dragon 3 (rare time losses in standard mode and crashes in MCTS mode), adds a new feature Skill Time (adjustible minimum time per move when using the LimitStrength (Elo) option), and with a new net and search improvements is around 10 to 20 elo stronger depending on type of book used and time control."
Ah! So you can set a skill level and set the engine to move no fewer than (lets say) 10 sec after you move?!!!?
Yes, but the max was supposed to be 10 seconds, but somehow it won't take more than 4 seconds even when set to 10. So we should fix it in the future, but four seconds max is probably enough delay for most users, though certainly not all. Personally I prefer just a one second delay, which is what chess.com uses for their bots.
Is it supposed to play a bit stronger at a certain Elo setting with this 4 seconds delay, because of a little pondering time it allows, or it is irrelevant?
It plays at exactly the same strength with or without the extra seconds. If it actually used those seconds to think, the elo would shoot up by a thousand points or so.
Komodo rules!
Fritz 0
Posts: 150
Joined: Fri Mar 11, 2022 12:10 pm
Full name: Branislav Đošić

Re: Dragon 3.1 Released at KomodoChess.com

Post by Fritz 0 »

lkaufman wrote: Mon Aug 08, 2022 6:58 pm
Fritz 0 wrote: Mon Aug 08, 2022 9:14 am
lkaufman wrote: Mon Aug 08, 2022 7:13 am
CornfedForever wrote: Mon Aug 08, 2022 2:33 am
cc2150dx wrote: Sun Aug 07, 2022 11:22 pm "Dragon 3.1 fixes some issues in Dragon 3 (rare time losses in standard mode and crashes in MCTS mode), adds a new feature Skill Time (adjustible minimum time per move when using the LimitStrength (Elo) option), and with a new net and search improvements is around 10 to 20 elo stronger depending on type of book used and time control."
Ah! So you can set a skill level and set the engine to move no fewer than (lets say) 10 sec after you move?!!!?
Yes, but the max was supposed to be 10 seconds, but somehow it won't take more than 4 seconds even when set to 10. So we should fix it in the future, but four seconds max is probably enough delay for most users, though certainly not all. Personally I prefer just a one second delay, which is what chess.com uses for their bots.
Is it supposed to play a bit stronger at a certain Elo setting with this 4 seconds delay, because of a little pondering time it allows, or it is irrelevant?
It plays at exactly the same strength with or without the extra seconds. If it actually used those seconds to think, the elo would shoot up by a thousand points or so.
I meant that it perhaps should play a little stronger due to the opponent's pondering time. Of course that Dragon would crush anybody with 4 seconds per move.
lkaufman
Posts: 6229
Joined: Sun Jan 10, 2010 6:15 am
Location: Maryland USA
Full name: Larry Kaufman

Re: Dragon 3.1 Released at KomodoChess.com

Post by lkaufman »

Fritz 0 wrote: Mon Aug 08, 2022 8:08 pm
lkaufman wrote: Mon Aug 08, 2022 6:58 pm
Fritz 0 wrote: Mon Aug 08, 2022 9:14 am
lkaufman wrote: Mon Aug 08, 2022 7:13 am
CornfedForever wrote: Mon Aug 08, 2022 2:33 am
cc2150dx wrote: Sun Aug 07, 2022 11:22 pm "Dragon 3.1 fixes some issues in Dragon 3 (rare time losses in standard mode and crashes in MCTS mode), adds a new feature Skill Time (adjustible minimum time per move when using the LimitStrength (Elo) option), and with a new net and search improvements is around 10 to 20 elo stronger depending on type of book used and time control."
Ah! So you can set a skill level and set the engine to move no fewer than (lets say) 10 sec after you move?!!!?
Yes, but the max was supposed to be 10 seconds, but somehow it won't take more than 4 seconds even when set to 10. So we should fix it in the future, but four seconds max is probably enough delay for most users, though certainly not all. Personally I prefer just a one second delay, which is what chess.com uses for their bots.
Is it supposed to play a bit stronger at a certain Elo setting with this 4 seconds delay, because of a little pondering time it allows, or it is irrelevant?
It plays at exactly the same strength with or without the extra seconds. If it actually used those seconds to think, the elo would shoot up by a thousand points or so.
I meant that it perhaps should play a little stronger due to the opponent's pondering time. Of course that Dragon would crush anybody with 4 seconds per move.
If the opponent (human) has the extra ponder time, he should do better so the effective rating of the engine for a given setting should be lower, not higher. Probably not significant for one second delay, maybe a small measurable weakening with several seconds.
Komodo rules!
Fritz 0
Posts: 150
Joined: Fri Mar 11, 2022 12:10 pm
Full name: Branislav Đošić

Re: Dragon 3.1 Released at KomodoChess.com

Post by Fritz 0 »

lkaufman wrote: Tue Aug 09, 2022 1:41 am
Fritz 0 wrote: Mon Aug 08, 2022 8:08 pm
lkaufman wrote: Mon Aug 08, 2022 6:58 pm
Fritz 0 wrote: Mon Aug 08, 2022 9:14 am
lkaufman wrote: Mon Aug 08, 2022 7:13 am
CornfedForever wrote: Mon Aug 08, 2022 2:33 am
cc2150dx wrote: Sun Aug 07, 2022 11:22 pm "Dragon 3.1 fixes some issues in Dragon 3 (rare time losses in standard mode and crashes in MCTS mode), adds a new feature Skill Time (adjustible minimum time per move when using the LimitStrength (Elo) option), and with a new net and search improvements is around 10 to 20 elo stronger depending on type of book used and time control."
Ah! So you can set a skill level and set the engine to move no fewer than (lets say) 10 sec after you move?!!!?
Yes, but the max was supposed to be 10 seconds, but somehow it won't take more than 4 seconds even when set to 10. So we should fix it in the future, but four seconds max is probably enough delay for most users, though certainly not all. Personally I prefer just a one second delay, which is what chess.com uses for their bots.
Is it supposed to play a bit stronger at a certain Elo setting with this 4 seconds delay, because of a little pondering time it allows, or it is irrelevant?
It plays at exactly the same strength with or without the extra seconds. If it actually used those seconds to think, the elo would shoot up by a thousand points or so.
I meant that it perhaps should play a little stronger due to the opponent's pondering time. Of course that Dragon would crush anybody with 4 seconds per move.
If the opponent (human) has the extra ponder time, he should do better so the effective rating of the engine for a given setting should be lower, not higher. Probably not significant for one second delay, maybe a small measurable weakening with several seconds.
That is exactly what I meant - Dragon should play stronger if the opponent has the pondering time for the Elo settings to be accurate.
magicianofriga
Posts: 3
Joined: Thu Aug 11, 2022 4:57 am
Full name: Tanmay Srinath

Re: Dragon 3.1 Released at KomodoChess.com

Post by magicianofriga »

Hi all,
I am new here, and I had a few questions for Larry and Mark. How strong/weak is Dragon MCTS compared to regular Dragon, SF and Leela? I had read a few years back that there was an objective to replace the AB Komodo with its MCTS version - is it still on? Also, how useful is Dragon MCTS for CC play? How do we use it correctly for the same? Does the regular hash matter as much as the MCTS hash?
Thanks a lot!