Difference between Computer and humans chess players
Moderators: hgm, Dann Corbit, Harvey Williamson
-
Chessqueen
- Posts: 5482
- Joined: Wed Sep 05, 2018 2:16 am
- Location: Moving
- Full name: Jorge Picado
Difference between Computer and humans chess players
Most computer chess engines have a constant chess rating that can be within 50 Elo higher or lower, but if you take for instance a human GM rated around 2650 to 2700, their rating could be up or down several hundreds Elo, for instance in one game they might perform around 2200 or lower if they blunder and in other games they might perform around 2800 or even higher.
Forget about memorization of Opening Theories https://www.youtube.com/watch?v=DN3381sdcdY
-
Cornfed
- Posts: 511
- Joined: Sun Apr 26, 2020 11:40 pm
- Full name: Brian D. Smith
Re: Difference between Computer and humans chess players
So...
Seriously, I thought you were going to say one has fingers and one does not.
Seriously, I thought you were going to say one has fingers and one does not.
-
Uri Blass
- Posts: 10101
- Joined: Thu Mar 09, 2006 12:37 am
- Location: Tel-Aviv Israel
Re: Difference between Computer and humans chess players
I disagree.Chessqueen wrote: ↑Thu Jan 27, 2022 11:41 pm Most computer chess engines have a constant chess rating that can be within 50 Elo higher or lower, but if you take for instance a human GM rated around 2650 to 2700, their rating could be up or down several hundreds Elo, for instance in one game they might perform around 2200 or lower if they blunder and in other games they might perform around 2800 or even higher.
What is the definition of performance in a game?
There is one definition based on result and you can say that if you lose against 2600 you get a performance of 2200 and if you win against 2600 you get a performance of 3000 and in this case computers with rating 2600 may also win against 2600 in a single game and lose against 2600 in a single game.
Second possible definition is based on the quality of the moves and in this case the question is how you measure quality of the moves.
It is possible that in one opening some engine with 2600 has quality of moves that is equivalent to 2200 humans and in another opening the quality of the moves of it is better than carlsen.
2600 chess engines may avoid tactical blunders assuming no bugs that is not always the case but
they can do big positional blunders in some openings that allow relatively weak players to beat them if they prepare to play the right openings.
I can add that in human games it is not something common to see 2650 to 2700 players lose against 2300 or 2200.
Blunders are not something random and usually the stronger player play moves that increase the chances of the weaker player to blunder.
-
Chessqueen
- Posts: 5482
- Joined: Wed Sep 05, 2018 2:16 am
- Location: Moving
- Full name: Jorge Picado
Re: Difference between Computer and humans chess players
Uri Blass wrote: ↑Fri Jan 28, 2022 12:44 amI disagree.Chessqueen wrote: ↑Thu Jan 27, 2022 11:41 pm Most computer chess engines have a constant chess rating that can be within 50 Elo higher or lower, but if you take for instance a human GM rated around 2650 to 2700, their rating could be up or down several hundreds Elo, for instance in one game they might perform around 2200 or lower if they blunder and in other games they might perform around 2800 or even higher.
What is the definition of performance in a game?
There is one definition based on result and you can say that if you lose against 2600 you get a performance of 2200 and if you win against 2600 you get a performance of 3000 and in this case computers with rating 2600 may also win against 2600 in a single game and lose against 2600 in a single game.
Second possible definition is based on the quality of the moves and in this case the question is how you measure quality of the moves.
It is possible that in one opening some engine with 2600 has quality of moves that is equivalent to 2200 humans and in another opening the quality of the moves of it is better than carlsen.
2600 chess engines may avoid tactical blunders assuming no bugs that is not always the case but
they can do big positional blunders in some openings that allow relatively weak players to beat them if they prepare to play the right openings.
I can add that in human games it is not something common to see 2650 to 2700 players lose against 2300 or 2200.
Blunders are not something random and usually the stronger player play moves that increase the chances of the weaker player to blunder.
I did not meant it in that way, but for instance in the World championship in some games Neo played so bad that according to GM Caruana, he played worse than most 2200 players, NOT that his rating would drop to 2200, and most high rated players also blunder terribly in some games that the mistake is so horribly that can only be compared to a 2200 player, and the other way around, and that is when they play against players that are close to their ratings not that a 2600 players or 2700 would play against a 2300 or 2200 players in that same tournament, but it could happen too. Also sometimes a 2600 player would play so perfect that only Stockfish, or Komodo would agree with most of their moves, and not all engines moves are the best either, as I saw you posting some positions that stockfish can NOT find the Draw even after thinking for 2 hours, like this one that you recently posted. That point is that even if stockfish does not see the draw, it would not lose playing against another top engine or to Carsen with this position that you posted. [fen]2k5/2p5/1q1p4/pPpPp1pp/N1P1Pp2/P4PbP/KQ4P1/8 w - - 7 46[/fen]
Forget about memorization of Opening Theories https://www.youtube.com/watch?v=DN3381sdcdY
-
lkaufman
- Posts: 5942
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
Re: Difference between Computer and humans chess players
I think it depends a lot on the engine. If you pick a classic engine like an old Fritz or Shredder or Rebel (or a remake like Benjamin) that is around 2700 level, it will play tactically like 3000+ and positionally like maybe 2400 and so you would be correct. But if for example you pick Dragon with Elo set to 2700, it will make many more tactical errors but far fewer positional errors compared to these classic engines, and so should act much more like a 2700 human. True it will never make a blunder so obvious that you (for example) would recognize as such immediately, but it will make blunders that a GM would call blunders, missing some fairly simple tactical point a couple moves down the line. This would be even more true if you set Lc0 to look at only one node (or a few nodes) per move; it might average 2500 (for example), but would make elementary blunders sometimes and play 3000+ other times.Chessqueen wrote: ↑Thu Jan 27, 2022 11:41 pm Most computer chess engines have a constant chess rating that can be within 50 Elo higher or lower, but if you take for instance a human GM rated around 2650 to 2700, their rating could be up or down several hundreds Elo, for instance in one game they might perform around 2200 or lower if they blunder and in other games they might perform around 2800 or even higher.
Komodo rules!
-
Frank Quisinsky
- Posts: 6808
- Joined: Wed Nov 18, 2009 7:16 pm
- Location: Gutweiler, Germany
- Full name: Frank Quisinsky
Re: Difference between Computer and humans chess players
Hi Larry,
very esay for me give an answere in perfect english
I will give points from 1-10 to the four playing phases for strongest humans and strongest engines.
1. Earlier mid-game after openings:
Humans: 8 of 10
Engines: 7 of 10
Yes, the big and only chance for the strongest chess players for a small advantage!
2. Mid-game
Humans: 7 from 10
Engines: 8 from 10
3. Passage into the endgame
Humans: 5 of 10
Engines: 9 of 10
4. Endgames
Humans: 6 from 10
Engines: 10 from 10
The other more easy view will be ...
Humans: 08 / 07 / 05 / 06
Engines: 07 / 08 / 09 / 10
The information I gave a long time ...
Spark or Wasp or Fizbo plays computerchess human like.
These programs are much stronger in die first gaming phases as in the last gaming phases.
Very intelligent is to give a game draw if the game is draw!
Human like!!
Not that what Komodo do in version 2.6 with all the games over 300 moves (no other engine produced so many games over 300 moves Komodo like to do). I like more the way Wasp go, but Wasp can do that because Wasp don't have the strength in endgames Komodo have.
Example:
If a very small advantage in late mid-game ... search the way to a fast draw.
Again, this is human like!
Never Carlsen like to play chess with houndreds of moves if the position is draw.
I will give you an example:
0.025% of games (10 from 41.000) goes over 300 moves without resign mode (FCP Tourney-2021).
Dragon 2.61 NN (Komodo) have now 11 from 880 alone = 1,25%!
1,25% to 0,025% is a big, a very big different.
That's not a human like style.
So I can't understand why you search a discussion about such a topic if your program search the way to an "Anti-Human-Style".
If you wish do produce with Dragon a human like style it is very easy to do that.
1. The strenght in playing phase 1 from Dragon is TOP ...
not that was Don Daily created with his more or less passiv style in Doch and first Komodo versions.
2. The strenght in playing phase 2 from Dragon is TOP also!
Earlier mid-game / mid-game ... nothing to do!
But you have to reduce the strength in playing phase 3 and strenght in playing phase 4.
Komodo have to give faster a draw.
The next bad point in Dragon by Komodo 2.60.
Why your program calculated a move with 5 pieces or 6 pieces on board with endgame bases.
Do you know how many time I lost with all the 300 moves games if Komodo calculate nonsensical moves?
Not good what I see here what Komodo do in endgames!
Have a look here ...
Looks good what Dragon by Komodo 2.61 do in the first games phases.
For a human like playing style Komodo need UCI parameters to reduce the playing strength for endgames.
Example: 1% or 2% of the nomal speed can help here.
Best
Frank
very esay for me give an answere in perfect english
I will give points from 1-10 to the four playing phases for strongest humans and strongest engines.
1. Earlier mid-game after openings:
Humans: 8 of 10
Engines: 7 of 10
Yes, the big and only chance for the strongest chess players for a small advantage!
2. Mid-game
Humans: 7 from 10
Engines: 8 from 10
3. Passage into the endgame
Humans: 5 of 10
Engines: 9 of 10
4. Endgames
Humans: 6 from 10
Engines: 10 from 10
The other more easy view will be ...
Humans: 08 / 07 / 05 / 06
Engines: 07 / 08 / 09 / 10
The information I gave a long time ...
Spark or Wasp or Fizbo plays computerchess human like.
These programs are much stronger in die first gaming phases as in the last gaming phases.
Very intelligent is to give a game draw if the game is draw!
Human like!!
Not that what Komodo do in version 2.6 with all the games over 300 moves (no other engine produced so many games over 300 moves Komodo like to do). I like more the way Wasp go, but Wasp can do that because Wasp don't have the strength in endgames Komodo have.
Example:
If a very small advantage in late mid-game ... search the way to a fast draw.
Again, this is human like!
Never Carlsen like to play chess with houndreds of moves if the position is draw.
I will give you an example:
0.025% of games (10 from 41.000) goes over 300 moves without resign mode (FCP Tourney-2021).
Dragon 2.61 NN (Komodo) have now 11 from 880 alone = 1,25%!
1,25% to 0,025% is a big, a very big different.
That's not a human like style.
So I can't understand why you search a discussion about such a topic if your program search the way to an "Anti-Human-Style".
If you wish do produce with Dragon a human like style it is very easy to do that.
1. The strenght in playing phase 1 from Dragon is TOP ...
not that was Don Daily created with his more or less passiv style in Doch and first Komodo versions.
2. The strenght in playing phase 2 from Dragon is TOP also!
Earlier mid-game / mid-game ... nothing to do!
But you have to reduce the strength in playing phase 3 and strenght in playing phase 4.
Komodo have to give faster a draw.
The next bad point in Dragon by Komodo 2.60.
Why your program calculated a move with 5 pieces or 6 pieces on board with endgame bases.
Do you know how many time I lost with all the 300 moves games if Komodo calculate nonsensical moves?
Not good what I see here what Komodo do in endgames!
Have a look here ...
Code: Select all
January 23rd, 2022
Short games statistic, v1.2
After round 20 out of 30 = 800 games per engine
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
672 of 16.400 = 4,097% : Miniatures, mate below 60 moves
656 of 16.400 = 4,000% : Fast draws below 40 moves
won59 lost59 draw39 won59w won59b lost59w lost59b
01. Dragon 2.6 NN (Komodo) 68 0 12 57 11 0 0
02. Stockfish 311221 NN dev 64 0 22 37 27 0 0
03. SlowChess Blitz 2.8 NN 29 0 16 17 12 0 0
04. Revenge 2.0 NN 29 1 28 20 9 0 1
05. Arasan 23.2 NN 28 1 43 20 8 1 0
06. Fire 8 MC.3 NN 25 0 23 22 3 0 0
07. Ethereal 13.25 NN 25 2 23 17 8 0 2
08. Berserk 8.5.1 NN 25 3 37 16 9 0 3
09. Booot 6.5 25 5 41 17 8 2 3
10. Igel 3.0.10 NN 25 7 44 20 5 1 6
11. Rodent 1.0 NN 24 4 62 19 5 1 3
12. Schooner 2.2 XB 20 10 29 11 9 3 7
13. Wasp 5.20 NN 19 2 45 15 4 0 2
14. Weiss 2.0 19 6 17 12 7 1 5
15. rofChade 2.313 NN dev 18 2 37 13 5 0 2
16. Lc0 0.28.2 CPU 18 30 48 14 4 4 26
17. RubiChess 2021 NN 17 0 36 13 4 0 0
18. Xiphos 0.6 17 9 35 14 3 2 7
19. Koivisto 7.9 NN 16 8 22 13 3 0 8
20. Fritz 18 (Ginkgo) 15 19 19 9 6 3 16
21. Nemorino 6.09 NN dev 12 8 31 9 3 0 8
22. Chiron 5 11 22 23 9 2 6 16
23. DanaSah 9.0 NN 11 28 41 9 2 10 18
24. Minic 3.17 NN 11 30 37 8 3 4 26
25. Clover 2.4 9 13 38 8 1 1 12
26. Andscacs 0.95.123 dev 9 23 30 5 4 9 14
27. Seer 2.4.0 NN 8 6 36 7 1 2 4
28. Wasp 4.50 8 8 22 6 2 3 5
29. Laser 1.7 8 12 28 5 3 3 9
30. Shredder 13 8 22 34 6 2 5 17
31. GullChess 3.0 Sy 8 43 20 7 1 14 29
32. Fizbo 2.0 7 12 33 4 3 3 9
33. Defenchess 2.3 dev 7 13 31 5 2 1 12
34. Marvin 5.2.0 NN 7 17 33 5 2 6 11
35. Zahak 9.0 6 45 26 6 0 25 20
36. Dark Toga 1.1 NN 4 17 24 4 0 1 16
37. Halogen 10 NN 3 49 36 2 1 17 32
38. Orion 0.8 NN 3 51 41 3 0 15 36
39. Winter 0.9 3 54 19 2 1 18 36
40. Stash 32 1 35 30 1 0 13 22
41. Gogobello 3.0 NN 1 56 60 0 1 11 45
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
169 of 16.400 = 1,030% : Miniatures, mate below 50 moves
215 of 16.400 = 1,310% : Fast draws below 30 moves
won49 lost49 draw29 won49w won49b lost49w lost49b
01. Dragon 2.6 NN (Komodo) 23 0 3 18 5 0 0
02. Stockfish 311221 NN dev 20 0 10 13 7 0 0
03. Revenge 2.0 NN 9 0 9 7 2 0 0
04. SlowChess Blitz 2.8 NN 7 0 7 6 1 0 0
05. Booot 6.5 7 0 14 3 4 0 0
06. Wasp 5.20 NN 7 0 17 7 0 0 0
07. Fire 8 MC.3 NN 6 0 6 5 1 0 0
08. Xiphos 0.6 6 1 13 5 1 1 0
09. Rodent 1.0 NN 6 1 25 5 1 0 1
10. Weiss 2.0 6 2 10 2 4 0 2
11. Lc0 0.28.2 CPU 6 8 9 5 1 2 6
12. RubiChess 2021 NN 4 0 10 4 0 0 0
13. Igel 3.0.10 NN 4 0 13 2 2 0 0
14. rofChade 2.313 NN dev 4 0 16 3 1 0 0
15. Arasan 23.2 NN 4 1 10 4 0 1 0
16. Seer 2.4.0 NN 4 1 11 4 0 1 0
17. Laser 1.7 4 3 10 2 2 0 3
18. Fritz 18 (Ginkgo) 4 8 6 3 1 2 6
19. GullChess 3.0 Sy 4 14 8 3 1 2 12
20. Berserk 8.5.1 NN 3 0 12 3 0 0 0
21. Minic 3.17 NN 3 9 10 2 1 1 8
22. Wasp 4.50 2 0 5 2 0 0 0
23. Fizbo 2.0 2 0 9 1 1 0 0
24. Ethereal 13.25 NN 2 0 11 2 0 0 0
25. Schooner 2.2 XB 2 2 9 1 1 0 2
26. Defenchess 2.3 dev 2 2 15 2 0 0 2
27. Andscacs 0.95.123 dev 2 3 8 0 2 0 3
28. Nemorino 6.09 NN dev 2 3 9 2 0 0 3
29. Chiron 5 2 4 5 2 0 2 2
30. DanaSah 9.0 NN 2 8 14 2 0 5 3
31. Winter 0.9 2 14 4 2 0 4 10
32. Shredder 13 1 1 8 1 0 0 1
33. Clover 2.4 1 1 11 1 0 0 1
34. Dark Toga 1.1 NN 1 3 6 1 0 1 2
35. Marvin 5.2.0 NN 1 4 10 0 1 1 3
36. Koivisto 7.9 NN 1 5 11 1 0 0 5
37. Orion 0.8 NN 1 11 13 1 0 2 9
38. Halogen 10 NN 1 16 9 1 0 2 14
39. Stash 32 0 9 14 0 0 4 5
40. Zahak 9.0 0 16 8 0 0 6 10
41. Gogobello 3.0 NN 0 18 22 0 0 4 14Looks good what Dragon by Komodo 2.61 do in the first games phases.
For a human like playing style Komodo need UCI parameters to reduce the playing strength for endgames.
Example: 1% or 2% of the nomal speed can help here.
Best
Frank
-
Frank Quisinsky
- Posts: 6808
- Joined: Wed Nov 18, 2009 7:16 pm
- Location: Gutweiler, Germany
- Full name: Frank Quisinsky
Re: Difference between Computer and humans chess players
Hi Larry,
Here a "nogo" ZIP file for you (CBH database).
Check that ...
https://www.amateurschach.de/download/f ... __nogo.zip (all collected games, goes over 300 moves).
And please no chances to the first playing phases of games!
This is just great!!
Best
Frank
Here a "nogo" ZIP file for you (CBH database).
Check that ...
https://www.amateurschach.de/download/f ... __nogo.zip (all collected games, goes over 300 moves).
And please no chances to the first playing phases of games!
This is just great!!
Best
Frank
-
Frank Quisinsky
- Posts: 6808
- Joined: Wed Nov 18, 2009 7:16 pm
- Location: Gutweiler, Germany
- Full name: Frank Quisinsky
Re: Difference between Computer and humans chess players
Check that ...
https://www.amateurschach.de/t-ki/common/ki-ratings.jpg
Have a look on move average ... not on the great Elo Dragon have.
Dragon 2.6 is 7 moves higher as Dragon 2.5??
What do you do here??
Elo alone isn't important for the decision ... I like or dislike an engine!
And this is a statistic without all the 300 moves games (such games will be replayed for my tourneys).
This is really an "Anti-Human-Style to 100%".
Sorry, very hard for a program I really like but it's fact!!
Best
Frank
https://www.amateurschach.de/t-ki/common/ki-ratings.jpg
Have a look on move average ... not on the great Elo Dragon have.
Dragon 2.6 is 7 moves higher as Dragon 2.5??
What do you do here??
Elo alone isn't important for the decision ... I like or dislike an engine!
And this is a statistic without all the 300 moves games (such games will be replayed for my tourneys).
This is really an "Anti-Human-Style to 100%".
Sorry, very hard for a program I really like but it's fact!!
Best
Frank
-
lkaufman
- Posts: 5942
- Joined: Sun Jan 10, 2010 6:15 am
- Location: Maryland USA
Re: Difference between Computer and humans chess players
I think that Komodo plays drawn endgames out forever due to Contempt. If you don't like to see this, just set Contempt to zero. As for human-like play, this is only relevant when Elo is set to something in the human range; full strength Komodo is pretty much by definition not human-like. You talk about setting the speed to 1 or 2%, but the Elo settings do limit nodes searched to tiny percentages of normal, they just do it instantly (we could add an artificial delay as some have requested). So the question is, is Dragon with say a 2600 Elo setting more or less human-like than a full strength engine that happens to play at 2600 level?Frank Quisinsky wrote: ↑Fri Jan 28, 2022 8:29 am Hi Larry,
very esay for me give an answere in perfect english
I will give points from 1-10 to the four playing phases for strongest humans and strongest engines.
1. Earlier mid-game after openings:
Humans: 8 of 10
Engines: 7 of 10
Yes, the big and only chance for the strongest chess players for a small advantage!
2. Mid-game
Humans: 7 from 10
Engines: 8 from 10
3. Passage into the endgame
Humans: 5 of 10
Engines: 9 of 10
4. Endgames
Humans: 6 from 10
Engines: 10 from 10
The other more easy view will be ...
Humans: 08 / 07 / 05 / 06
Engines: 07 / 08 / 09 / 10
The information I gave a long time ...
Spark or Wasp or Fizbo plays computerchess human like.
These programs are much stronger in die first gaming phases as in the last gaming phases.
Very intelligent is to give a game draw if the game is draw!
Human like!!
Not that what Komodo do in version 2.6 with all the games over 300 moves (no other engine produced so many games over 300 moves Komodo like to do). I like more the way Wasp go, but Wasp can do that because Wasp don't have the strength in endgames Komodo have.
Example:
If a very small advantage in late mid-game ... search the way to a fast draw.
Again, this is human like!
Never Carlsen like to play chess with houndreds of moves if the position is draw.
I will give you an example:
0.025% of games (10 from 41.000) goes over 300 moves without resign mode (FCP Tourney-2021).
Dragon 2.61 NN (Komodo) have now 11 from 880 alone = 1,25%!
1,25% to 0,025% is a big, a very big different.
That's not a human like style.
So I can't understand why you search a discussion about such a topic if your program search the way to an "Anti-Human-Style".
If you wish do produce with Dragon a human like style it is very easy to do that.
1. The strenght in playing phase 1 from Dragon is TOP ...
not that was Don Daily created with his more or less passiv style in Doch and first Komodo versions.
2. The strenght in playing phase 2 from Dragon is TOP also!
Earlier mid-game / mid-game ... nothing to do!
But you have to reduce the strength in playing phase 3 and strenght in playing phase 4.
Komodo have to give faster a draw.
The next bad point in Dragon by Komodo 2.60.
Why your program calculated a move with 5 pieces or 6 pieces on board with endgame bases.
Do you know how many time I lost with all the 300 moves games if Komodo calculate nonsensical moves?
Not good what I see here what Komodo do in endgames!
Have a look here ...
Code: Select all
January 23rd, 2022 Short games statistic, v1.2 After round 20 out of 30 = 800 games per engine +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 672 of 16.400 = 4,097% : Miniatures, mate below 60 moves 656 of 16.400 = 4,000% : Fast draws below 40 moves won59 lost59 draw39 won59w won59b lost59w lost59b 01. Dragon 2.6 NN (Komodo) 68 0 12 57 11 0 0 02. Stockfish 311221 NN dev 64 0 22 37 27 0 0 03. SlowChess Blitz 2.8 NN 29 0 16 17 12 0 0 04. Revenge 2.0 NN 29 1 28 20 9 0 1 05. Arasan 23.2 NN 28 1 43 20 8 1 0 06. Fire 8 MC.3 NN 25 0 23 22 3 0 0 07. Ethereal 13.25 NN 25 2 23 17 8 0 2 08. Berserk 8.5.1 NN 25 3 37 16 9 0 3 09. Booot 6.5 25 5 41 17 8 2 3 10. Igel 3.0.10 NN 25 7 44 20 5 1 6 11. Rodent 1.0 NN 24 4 62 19 5 1 3 12. Schooner 2.2 XB 20 10 29 11 9 3 7 13. Wasp 5.20 NN 19 2 45 15 4 0 2 14. Weiss 2.0 19 6 17 12 7 1 5 15. rofChade 2.313 NN dev 18 2 37 13 5 0 2 16. Lc0 0.28.2 CPU 18 30 48 14 4 4 26 17. RubiChess 2021 NN 17 0 36 13 4 0 0 18. Xiphos 0.6 17 9 35 14 3 2 7 19. Koivisto 7.9 NN 16 8 22 13 3 0 8 20. Fritz 18 (Ginkgo) 15 19 19 9 6 3 16 21. Nemorino 6.09 NN dev 12 8 31 9 3 0 8 22. Chiron 5 11 22 23 9 2 6 16 23. DanaSah 9.0 NN 11 28 41 9 2 10 18 24. Minic 3.17 NN 11 30 37 8 3 4 26 25. Clover 2.4 9 13 38 8 1 1 12 26. Andscacs 0.95.123 dev 9 23 30 5 4 9 14 27. Seer 2.4.0 NN 8 6 36 7 1 2 4 28. Wasp 4.50 8 8 22 6 2 3 5 29. Laser 1.7 8 12 28 5 3 3 9 30. Shredder 13 8 22 34 6 2 5 17 31. GullChess 3.0 Sy 8 43 20 7 1 14 29 32. Fizbo 2.0 7 12 33 4 3 3 9 33. Defenchess 2.3 dev 7 13 31 5 2 1 12 34. Marvin 5.2.0 NN 7 17 33 5 2 6 11 35. Zahak 9.0 6 45 26 6 0 25 20 36. Dark Toga 1.1 NN 4 17 24 4 0 1 16 37. Halogen 10 NN 3 49 36 2 1 17 32 38. Orion 0.8 NN 3 51 41 3 0 15 36 39. Winter 0.9 3 54 19 2 1 18 36 40. Stash 32 1 35 30 1 0 13 22 41. Gogobello 3.0 NN 1 56 60 0 1 11 45 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 169 of 16.400 = 1,030% : Miniatures, mate below 50 moves 215 of 16.400 = 1,310% : Fast draws below 30 moves won49 lost49 draw29 won49w won49b lost49w lost49b 01. Dragon 2.6 NN (Komodo) 23 0 3 18 5 0 0 02. Stockfish 311221 NN dev 20 0 10 13 7 0 0 03. Revenge 2.0 NN 9 0 9 7 2 0 0 04. SlowChess Blitz 2.8 NN 7 0 7 6 1 0 0 05. Booot 6.5 7 0 14 3 4 0 0 06. Wasp 5.20 NN 7 0 17 7 0 0 0 07. Fire 8 MC.3 NN 6 0 6 5 1 0 0 08. Xiphos 0.6 6 1 13 5 1 1 0 09. Rodent 1.0 NN 6 1 25 5 1 0 1 10. Weiss 2.0 6 2 10 2 4 0 2 11. Lc0 0.28.2 CPU 6 8 9 5 1 2 6 12. RubiChess 2021 NN 4 0 10 4 0 0 0 13. Igel 3.0.10 NN 4 0 13 2 2 0 0 14. rofChade 2.313 NN dev 4 0 16 3 1 0 0 15. Arasan 23.2 NN 4 1 10 4 0 1 0 16. Seer 2.4.0 NN 4 1 11 4 0 1 0 17. Laser 1.7 4 3 10 2 2 0 3 18. Fritz 18 (Ginkgo) 4 8 6 3 1 2 6 19. GullChess 3.0 Sy 4 14 8 3 1 2 12 20. Berserk 8.5.1 NN 3 0 12 3 0 0 0 21. Minic 3.17 NN 3 9 10 2 1 1 8 22. Wasp 4.50 2 0 5 2 0 0 0 23. Fizbo 2.0 2 0 9 1 1 0 0 24. Ethereal 13.25 NN 2 0 11 2 0 0 0 25. Schooner 2.2 XB 2 2 9 1 1 0 2 26. Defenchess 2.3 dev 2 2 15 2 0 0 2 27. Andscacs 0.95.123 dev 2 3 8 0 2 0 3 28. Nemorino 6.09 NN dev 2 3 9 2 0 0 3 29. Chiron 5 2 4 5 2 0 2 2 30. DanaSah 9.0 NN 2 8 14 2 0 5 3 31. Winter 0.9 2 14 4 2 0 4 10 32. Shredder 13 1 1 8 1 0 0 1 33. Clover 2.4 1 1 11 1 0 0 1 34. Dark Toga 1.1 NN 1 3 6 1 0 1 2 35. Marvin 5.2.0 NN 1 4 10 0 1 1 3 36. Koivisto 7.9 NN 1 5 11 1 0 0 5 37. Orion 0.8 NN 1 11 13 1 0 2 9 38. Halogen 10 NN 1 16 9 1 0 2 14 39. Stash 32 0 9 14 0 0 4 5 40. Zahak 9.0 0 16 8 0 0 6 10 41. Gogobello 3.0 NN 0 18 22 0 0 4 14
Looks good what Dragon by Komodo 2.61 do in the first games phases.
For a human like playing style Komodo need UCI parameters to reduce the playing strength for endgames.
Example: 1% or 2% of the nomal speed can help here.
Best
Frank
Komodo rules!
-
Frank Quisinsky
- Posts: 6808
- Joined: Wed Nov 18, 2009 7:16 pm
- Location: Gutweiler, Germany
- Full name: Frank Quisinsky
Re: Difference between Computer and humans chess players
Hi Larry,
both version (2.51 and 2.6) are playing with the same default contempt settings.
So, it can't be contempt alone if I compare the move average and looking on all the 300 moves games Dragon 2.6 produced.
Furthermore, a lot of the "300-400" move games Dragon plays vs. Berserk. But this can not be the reason only. One of the last changes between 2.5 and 2.61 must be the reason.
To your question:
I think more "Human-like" is a question of endgame strength / playing phases.
The endgame strength from TOP engines is to high for best humans.
To reduce the strength generally to 2600, with nodes per second as example, will be not produce the wish result. In my opinion a more human like style is possible with clearly lesser strength in gaming phase 3 and 4 (in case of Dragon by Komodo).
Maybe with ...
If lesser as 16 pieces on board = 1-2% processor time / or x nodes per second.
Sure that different good ways in programming are possible.
Very important questions Larry!!!
I wrote since years that the secret in times today is to reduce Elo strength in combination with a human like style.
Chess programs are much more interesting for club players for selfplaying!
Example: Won vs. Dragon by Komodo with 2000 Elo ... people will be very happy and try as next 2100 Elo.
The secret is to reduce strength not with extra blunders!!
Best
Frank
both version (2.51 and 2.6) are playing with the same default contempt settings.
So, it can't be contempt alone if I compare the move average and looking on all the 300 moves games Dragon 2.6 produced.
Furthermore, a lot of the "300-400" move games Dragon plays vs. Berserk. But this can not be the reason only. One of the last changes between 2.5 and 2.61 must be the reason.
To your question:
I think more "Human-like" is a question of endgame strength / playing phases.
The endgame strength from TOP engines is to high for best humans.
To reduce the strength generally to 2600, with nodes per second as example, will be not produce the wish result. In my opinion a more human like style is possible with clearly lesser strength in gaming phase 3 and 4 (in case of Dragon by Komodo).
Maybe with ...
If lesser as 16 pieces on board = 1-2% processor time / or x nodes per second.
Sure that different good ways in programming are possible.
Very important questions Larry!!!
I wrote since years that the secret in times today is to reduce Elo strength in combination with a human like style.
Chess programs are much more interesting for club players for selfplaying!
Example: Won vs. Dragon by Komodo with 2000 Elo ... people will be very happy and try as next 2100 Elo.
The secret is to reduce strength not with extra blunders!!
Best
Frank