Komodo should be the 1st with a strength function !

[Chessqueen]

I know for a fact that 85% of the club players are rated below 1600, but none of the top engines have functions to make it practical for those 85% of the players to enjoy or should I say to be able to practice against their own strength. Back in the 90's and even now most of the dedicated chess computers, like Millennium, Mephito, Novag, etc.... came with strength level function suitable for every playing level they offer infinite levels of difficulty and is fully adjustable: from beginners to international grandmasters. key where players of all different strengths could set their strength level and have a chance to enjoy playing against their dedicated chess computer, but when PC chess engine came and even now most top engines lack this function. Imagine how many more people would be purchasing top engines that could offer this function. https://www.chessusa.com/product/82-S11.html


I do NOT see this function at all
Mastermind GM Larry Kaufman introduces Dragon 2 by Komodo Chess

May 4, 2021

We have released Dragon 2 at komodochess.com. Dragon uses NNUE (Neural Network Updated Efficiently) technology, originally developed for the game of shogi. Komodo has a great deal of chess knowledge in its evaluation. Training an NNUE network based on this evaluation was both an advantage and a challenge, requiring experimentation with architectures and data generation of billions of positions. The reinforcement learning phase for Dragon is in its infancy, but is already showing great promise.

Dragon 2 is a huge strength improvement over Komodo 14.1, the last pre-dragon release, about 210 elo in standard mode, at CCRL blitz time control on one thread based on 10,000 game direct matches. With four threads, the gains were 198 elo in MCTS and 184 elo in standard mode. The improvement is due to the embedded neural network providing much more accurate evaluation and also in some sense gaining an extra ply or so of search by “seeing” some tactics that a normal eval won’t spot. Our testing also shows that Dragon is especially strong in Fischerandom (960) chess, which was confirmed when Dragon won the TCEC FRC 2021 championship. Dragon also won the 2021 TCEC Swiss championship (normal chess). We believe that Dragon will play in a more human-like style than standard Komodo since it relies on learning what actually works in games rather than just on pre-assigned values for eval terms.

Although Dragon 2 is only slightly stronger than the initial dragon version in standard mode, the MCTS mode when running under Windows is dramatically improved, about a hundred elo. Also the MultiPV play of standard mode is dramatically improved from the initial dragon release, again by about a hundred elo! Based on our testing, we now believe that Dragon 2 in standard mode is the strongest engine for analysis on most computers when displaying more than four lines of play at once. A new feature with Dragon 2 is “MCTS Optimism”, which allows the user to make Dragon “trust” its own judgments more while being less confident in predicting the opponent’s moves. Other features added since Komodo 13 include Armageddon mode, more levels, Personalities, and Auto-Skill. Standard mode is usually the stronger mode objectively, while MCTS mode is usually better at setting problems for human opponents as it does not assume perfect defense. Skill levels have been adjusted to use the Dragon NNUE without much change in the strength of a given skill level; we think this makes the skill levels more "human-like".

[Chessqueen]

I know for a fact that 85% of the club players are rated below 1600, but none of the top engines have functions to make it practical for those 85% of the players to enjoy or should I say to be able to practice against their own strength. Back in the 90's and even now most of the dedicated chess computers, like Millennium, Mephito, Novag, etc.... came with strength level function suitable for every playing level they offer infinite levels of difficulty and is fully adjustable: from beginners to international grandmasters. key where players of all different strengths could set their strength level and have a chance to enjoy playing against their dedicated chess computer, but when PC chess engine came and even now most top engines lack this function. Imagine how many more people would be purchasing top engines that could offer this function. The level function could be set up to 2800 just in case Carlsen would like to practice against it, and of course it does not have to be exact it might be off by 100 human FIDE Elo either up or down https://www.chessusa.com/product/82-S11.html


I do NOT see this function at all
Mastermind GM Larry Kaufman introduces Dragon 2 by Komodo Chess

May 4, 2021

We have released Dragon 2 at komodochess.com. Dragon uses NNUE (Neural Network Updated Efficiently) technology, originally developed for the game of shogi. Komodo has a great deal of chess knowledge in its evaluation. Training an NNUE network based on this evaluation was both an advantage and a challenge, requiring experimentation with architectures and data generation of billions of positions. The reinforcement learning phase for Dragon is in its infancy, but is already showing great promise.

Dragon 2 is a huge strength improvement over Komodo 14.1, the last pre-dragon release, about 210 elo in standard mode, at CCRL blitz time control on one thread based on 10,000 game direct matches. With four threads, the gains were 198 elo in MCTS and 184 elo in standard mode. The improvement is due to the embedded neural network providing much more accurate evaluation and also in some sense gaining an extra ply or so of search by “seeing” some tactics that a normal eval won’t spot. Our testing also shows that Dragon is especially strong in Fischerandom (960) chess, which was confirmed when Dragon won the TCEC FRC 2021 championship. Dragon also won the 2021 TCEC Swiss championship (normal chess). We believe that Dragon will play in a more human-like style than standard Komodo since it relies on learning what actually works in games rather than just on pre-assigned values for eval terms.

Although Dragon 2 is only slightly stronger than the initial dragon version in standard mode, the MCTS mode when running under Windows is dramatically improved, about a hundred elo. Also the MultiPV play of standard mode is dramatically improved from the initial dragon release, again by about a hundred elo! Based on our testing, we now believe that Dragon 2 in standard mode is the strongest engine for analysis on most computers when displaying more than four lines of play at once. A new feature with Dragon 2 is “MCTS Optimism”, which allows the user to make Dragon “trust” its own judgments more while being less confident in predicting the opponent’s moves. Other features added since Komodo 13 include Armageddon mode, more levels, Personalities, and Auto-Skill. Standard mode is usually the stronger mode objectively, while MCTS mode is usually better at setting problems for human opponents as it does not assume perfect defense. Skill levels have been adjusted to use the Dragon NNUE without much change in the strength of a given skill level; we think this makes the skill levels more "human-like".

[Paloma]

double posting (quoting), it does not make it any better.

[Chessqueen]

double posting (quoting), it does not make it any better.

It is not exactly the same I simply forgot to include this. The level function could be set up to 2800 just in case Carlsen would like to practice against it, and of course it does not have to be exact it might be off by 100 human FIDE Elo either up or down

[lkaufman]

Based on my latest findings, Komodo Dragon 2 on one thread playing Rapid (15' + 10") against human players using the Skill levels 1-24 should be roughly evenly matched if the human FIDE rating = 115 x Skill level. Are you saying that 115 elo points between levels is too wide, or just that we need to publicize this equation, or have the user enter his elo and have the engine divide by 115? For intermediate levels, using MP will raise the level by some fractional amount, though we don't have accurate data on just how much. Perhaps in the future we'll have MCTS Skill levels, in which case it would be easier to have smooth changes with increasing elo.

[Chessqueen]

Based on my latest findings, Komodo Dragon 2 on one thread playing Rapid (15' + 10") against human players using the Skill levels 1-24 should be roughly evenly matched if the human FIDE rating = 115 x Skill level. Are you saying that 115 elo points between levels is too wide, or just that we need to publicize this equation, or have the user enter his elo and have the engine divide by 115? For intermediate levels, using MP will raise the level by some fractional amount, though we don't have accurate data on just how much. Perhaps in the future we'll have MCTS Skill levels, in which case it would be easier to have smooth changes with increasing elo.

Yes what I meant to say is that the MCTS should have skill levels that are comparable to Human FIDE ratings, for instance if MCTS is set to play at 1600 Elo that a Human rated in between 1550 to 1650 could at least give it a good game, and if it is set to play at 2500 that any Human Fide rated between 2450 to 2550 could give it a good fighting game. As I mentioned before most dedicated computer back from the 90's up to now have this option which made it suitable for every playing level by offering infinite levels of difficulty and they are fully adjustable: from beginners to international grandmasters. key where players of all different strengths could set their strength level and have a chance to enjoy playing against their dedicated chess computer. for instance Harmon already have an option to play by Elo or predefined settings "Expert, Class A, Class B, Class C" etc. but when PC chess engine came and even now most top engines lack this function except Harmon. Adaptive play will automatically force the engine to play weaker if it gets ahead. Similar to variety, it is automatically turned off once the total piece count gets to 24 pieces. It can play up to 300 Elo or so weaker, and it can be used in conjunction with variety play which would weaken engine play .by another 75 to 100 Elo. Using both Adaptive Play and Variety together, at an Elo setting of 1300, Honey or Harmon. Elo play is obtained primarily by node count. This also saves on the power consumption. which may be of benefit on an Android device. Adaptive play was designed for beginner/weaker players who find the lowest Elo settings too hard. An average club player will find the Elo settings of 1500 to 1800 with variety turned on to be just about right (with Adaptive off). With variety turned on, you do not need an opening book as it is very unlikely to have repeat games, in the Android App, the internal book is on by default, but it can be turned off. To avoid instant move with the Honey engine, Honey has an engine option to sleep before making its move. It will automatically sleep for the appropriately time give the time control you selected. All of the Elo levels for the Honey engine factor in the time control, so it will search more nodes. as you select a longer time control for a game or if you select a higher elo setting. The default nps for an Android device is set around 500Knps, if your Android device is much slower than 500K/nps, you can adjust the average nps suitable for your device. The Elo settings are on the top left side menu. Adaptive play and Variety settings are in the engines settings Imagine how many more people would be purchasing top engines that could offer this function. https://www.chessusa.com/product/82-S11.html

[lkaufman]

Is your point that we need skill levels for mcts, or that users are unable to divide their rating by 115 to choose the proper skill level?

[Chessqueen]

Is your point that we need skill levels for mcts, or that users are unable to divide their rating by 115 to choose the proper skill level?

It is much easier to make Komodo MCTS play at different Elo ratings from Class E to World Champion =Elo 2835 like Honey. Sorry I do NOT know how to take a screenshot of Honey different skill Levels to show you here, probably somebody else can take a screenshot and post it here.
forum3/viewtopic.php?f=2&t=77658