The most human-like engine?

[Frank Quisinsky]

Hi Dietrich,

you like to give "bla, bla ... not easy for me" in this thread!

Why Wasp is my favorite engine?
It must be a reason!

Why Phalanx and AnMon are my favorits in the past.
Or later Spark and Hakka, Junior, ETChess, Aristarch, Patzer ...

What I wrote about Wasp is fact!
But Wasp isn't the once again with such strengths.
Velvet is the topic!!

You have to read messages from myself with more attention.
If I can see strengths and weaknesses from others I will give directly the information with stats and other examples. So in my tourneys in the past are allways all of TOP-40 or TOP-60 engines.

So, what you do here is to use every chance to try to fight your own game vs. others persons.
But this have nothing to do with computer chess. This have more to do with your character.

OK for me, many others do that if own projects are in the foreground or people like to give the information:
I am a hero ... that is not new for me!

I like to speak from facts! This is much more interesting!

I made not many MCTS analyzes.
All what I can see is that Komodo with MCTS lost to many games to fast.
That's is maybe human like on a 2300 Elo level, or 2200 Elo level.
From 2.000 games vs. a group of attackers, Komodo 14.1 MCTS lost over 300 to fast.
That's blunder chess, nothing to do with human-like-style!

Feel free to make the same experiment and you have more information before you set such a message!

Again, the game started with openings and openings are the biggest argument from humans ... what is human-like!
The long-plan in openings are the biggest argument from grandmaster vs. computer chess. So, we have to give the strongest players in the World material to see how strong chess engines are today in openings. Most interesting topic around computerchess in my humble opinion!

Best
Frank

[dkappe]

I made not many MCTS analyzes.

In other words, “I don’t have a decent GPU.”

I’m, unfortunately, playing into your hijacking of this thread. For this I apologize.

P.S. The good news for those interested in human-like play, you can use these small leela-style nets at low nodes quite comfortably on a CPU.

[Frank Quisinsky]

cost for electricity is more the problem.
In two from my "chess systems" are RTX-3700 Ti cards!
Hardware in "Deep-Sleeping"

Interesting is Devre.
I have to read the information from the programmer in detail.
I believe the programmer is also speaking from Leela Nets.
Not testing so far ...

Lc0 with 40 moves in 20 minutes participate in my last official FCP Tourney-KI and is a bit weaker as Stockfish and Dragon by Komodo. Newer nets as the version from February 2022 I never tested. But with the results from Lc0 GPU I am very happy. Playing style is great, move average is great ... all stats are really very good.

I am looking more in detail for the different complicated opening systems.
Here the stats from Dragon by Komodo are on highest level (CPU-Engines).

[dkappe]

Getting back to the topic, AB engines with “human” modes always feel like playing a drunk grandmaster. You can have a look at the lower ranks of CCRL, like sub-2000, but these engines usually have poor evaluations and play nothing like human beings.

The Maia weights can be found here: https://github.com/CSSLab/maia-chess/tr ... ia_weights

Of my nets, I’d suggest Mean Girl and Tiny Girl.

https://github.com/dkappe/leela-chess-w ... -style-net
https://github.com/dkappe/leela-chess-w ... i/Bad-Gyal (Last in the list.)

You can play around with how many nodes to search

Code: Select all

go nodes 500

, etc.

[Frank Quisinsky]

istolacio ...

Other opinion: Most important are the first gaming phase.
Here Dragon by Komodo 3.1 NN is more interesting as Stockfish.
Can be see very easy with "fast won" games / "fast lost" games.

Today we have the possibility to use more as one engine for analyzes.
Interesting are 4 or 8 engines with different strengths and one core.

This combination of engines for analyzes is better as a chess tutor with one engine only.
The main reason why "Playing styles" of engines are all the time important.

Engines for a chess tutor are engines like to play all playing phases with around the same level without bigger weaknesses.
Here RubiChess or Ethereal are good examples. The styles are very balanced.
On other way is to use the strongest available engine!

Example:
I am using for analyzing openings:

- Dragon by Komodo 3.1 NN
- Slowchess Blitz 2.9 NN
- Revenge 3.0 NN
- Wasp 6.00 NN
- Velvet 6.00 NN
- RubiChess Aug. NN
- often Junior for speculative combinations!

RubiChess is very strong with unbalanced material with many pieces on board.

Best
Frank

PS: If you searching an easy chess tutor ...
Pharaon by Frank ZIBI have this as UCI feature ... I believe v2.56 or v2.58.
Or you can use Fritz GUI.

[dkappe]

For a flavor of what MeanGirl on a Raspberry PI 3+ does at 4000 nodes:

[pgn]
[Event "Casual Rapid game"]
[Site "https://lichess.org/RBZ3IVvv"]
[Date "2022.08.22"]
[Round "?"]
[White "Cavallo-Matto"]
[Black "MiniHuman"]
[Result "0-1"]
[WhiteElo "2027"]
[BlackElo "2541"]
[ECO "C00"]
[TimeControl "1200+3"]
[UTCDate "2022.08.22"]
[UTCTime "14:12:31"]
[BlackTitle "BOT"]
[Variant "Standard"]
[Opening "French Defense: Schlechter Variation"]
[Termination "Normal"]
[Annotator "lichess.org"]

1. e4 {[%eval 0.33] [%clk 0:20:00]} 1... e6 {[%eval 0.13] [%clk 0:20:00]} 2. d4
{[%eval 0.3] [%clk 0:20:02]} 2... d5 {[%eval 0.25] [%clk 0:20:03]} 3. Bd3
{C00 French Defense: Schlechter Variation} 3... dxe4
{[%eval -0.08] [%clk 0:20:06]} 4. Bxe4 {[%eval -0.17] [%clk 0:20:04]} 4... Nf6
{[%eval 0.0] [%clk 0:20:09]} 5. Bd3 {[%eval -0.36] [%clk 0:20:05]} 5... c5
{[%eval -0.13] [%clk 0:20:11]} 6. c3 {[%eval -0.67] [%clk 0:20:04]} 6... Nc6 $6
{[%eval -0.09] [%clk 0:20:14]} (6... cxd4 7. cxd4 g6 8. Nf3 Bg7 9. O-O O-O 10.
Nc3 Nc6 11. Be3 ) Nf3 {[%eval -0.21] [%clk 0:20:04]} Be7
{[%eval 0.32] [%clk 0:20:17]} 8. O-O $6 {[%eval -0.27] [%clk 0:20:06]} (8. dxc5
Bxc5 9. Qe2 e5 10. Bg5 h6 11. Bxf6 Qxf6 12. Nbd2 Bb6 13. Nc4 Bc7 14. O-O O-O )
8... cxd4 {[%eval -0.24] [%clk 0:20:20]} 9. cxd4 {[%eval -0.11] [%clk 0:20:08]}
9... O-O {[%eval -0.24] [%clk 0:20:22]} 10. a3 {[%eval -0.29] [%clk 0:20:01]}
10... Nxd4 $4 {[%eval 5.62] [%clk 0:20:24]} (10... Qd7 11. Ne5 ) Nxd4 Qxd4 $6
{[%eval 7.69] [%clk 0:20:26]} (11... Qd7 12. Nf3 ) Bxh7+
{[%eval 7.83] [%clk 0:20:02]} Nxh7 {[%eval 7.99] [%clk 0:20:29]} 13. Qxd4
{[%eval 7.91] [%clk 0:20:03]} 13... Rd8 {[%eval 7.76] [%clk 0:20:32]} 14. Qe4
{[%eval 7.69] [%clk 0:19:39]} 14... Nf6 {[%eval 7.6] [%clk 0:20:35]} 15. Qf3
{[%eval 7.91] [%clk 0:19:41]} 15... Bd7 {[%eval 7.84] [%clk 0:20:36]} 16. Qxb7
{[%eval 7.91] [%clk 0:19:40]} 16... Rab8 {[%eval 7.61] [%clk 0:20:38]} 17. Qxa7
{[%eval 7.0] [%clk 0:19:41]} 17... Ra8 {[%eval 8.07] [%clk 0:20:40]} 18. Qe3
{[%eval 8.05] [%clk 0:19:37]} 18... Bb5 {[%eval 8.12] [%clk 0:20:43]} 19. b4 $6
{[%eval 5.82] [%clk 0:19:30]} (19. Re1 Bb4 20. Bd2 Rd3 21. Qg5 Rd5 22. Qf4 Bd6
23. Qf3 Rd8 24. a4 Rf5 25. axb5 Rxf3 ) 19... Bxf1 {[%eval 6.12] [%clk 0:20:45]}
20. Kxf1 {[%eval 5.85] [%clk 0:19:31]} 20... Rd1+ {[%eval 6.27] [%clk 0:20:47]}
21. Ke2 {[%eval 6.39] [%clk 0:19:32]} 21... Rg1 {[%eval 6.09] [%clk 0:20:50]}
22. g3 {[%eval 5.75] [%clk 0:19:33]} 22... Nd5 {[%eval 5.74] [%clk 0:20:51]} 23.
Qf3 $4 {[%eval -4.87] [%clk 0:19:29]} (23. Qd2 ) Rxc1
{[%eval -5.0] [%clk 0:20:53]} 24. Kd2 {[%eval -5.71] [%clk 0:19:19]} 24... Bg5+
{White resigns.}
0-1
[/pgn]

[Odd Gunnar Malin]

Sorry if this question has already been asked here, but which is, in your opinion, the most human-like chess engine?

Hi.
I don't have an answer to your question, but here are some of my observations.

When I spar in technical endgame I use Dragon in MCTS mode at full strength, I don't have any tablebases installed. This is for positions I know are winning or drawing. If it is a win I take the winning side and if it is a draw I take the side that should try to hold a draw. The objection to using an engine for this has been in the past that the engine will make an inhuman move in f.ex. a drawn R+P vs. R because all moves lead to a draw. A human would still try to win this. With a Monte Carlo engine like Dragon in MCTS mode it will still try to win those endings like a human would. By the way I play those sparring games with a timecontrol of 1+30 to simulate a real game, these endings are reached at the end when you don't have much time left.

I have been using Hiarcs 15 for a few games with tournament timecontrol (90/40 + 30/g with 30 sec. inc). The normal flow in these games is that I stand better in the opening and early middlegame but it trick me somewhat in the end. I like it because it lets me train on winning better position and keep my concentration up throughout the whole game. This doesn't qualify for human-like, but anyhow useful for training.

I don't have the latest Shredder but when Shredder 12 was released I let my son play with it on lower levels (around 1000). I was a little impressed by what I saw. Instead of blundering pieces directly, it set up a small combination to be found. I don't know how it plays at your level.

So you see, I select engines for what I want to learn today.


Odd Gunnar

[Odd Gunnar Malin]

I had to run to a webinar with GM Ankit Rajpara so I forgot to mention that the Maia engine with the setup on their site and on Lichess is probably too weak for you. I'm 1700+ and have no problem with it. I guess it is possible to do some adjustment to get it stronger since the engine is lz0. I guess 1900 is lichess rating so 1500+ in real rating.

[Fritz 0]

Sorry if this question has already been asked here, but which is, in your opinion, the most human-like chess engine?

Hi.
I don't have an answer to your question, but here are some of my observations.

When I spar in technical endgame I use Dragon in MCTS mode at full strength, I don't have any tablebases installed. This is for positions I know are winning or drawing. If it is a win I take the winning side and if it is a draw I take the side that should try to hold a draw. The objection to using an engine for this has been in the past that the engine will make an inhuman move in f.ex. a drawn R+P vs. R because all moves lead to a draw. A human would still try to win this. With a Monte Carlo engine like Dragon in MCTS mode it will still try to win those endings like a human would. By the way I play those sparring games with a timecontrol of 1+30 to simulate a real game, these endings are reached at the end when you don't have much time left.

I have been using Hiarcs 15 for a few games with tournament timecontrol (90/40 + 30/g with 30 sec. inc). The normal flow in these games is that I stand better in the opening and early middlegame but it trick me somewhat in the end. I like it because it lets me train on winning better position and keep my concentration up throughout the whole game. This doesn't qualify for human-like, but anyhow useful for training.

I don't have the latest Shredder but when Shredder 12 was released I let my son play with it on lower levels (around 1000). I was a little impressed by what I saw. Instead of blundering pieces directly, it set up a small combination to be found. I don't know how it plays at your level.

So you see, I select engines for what I want to learn today.


Odd Gunnar

I have Hiarcs and played a few games at 2000 Elo. In each game It made ridiculous blunders that a 2000 human player would make perhaps once in 10-20 games if he had a really bad day. So I see no point in using it. Maybe I should try it at a fixed depth of 5 ply or so and see if it's more useful.

I also downloaded Shredder trial version and it played much more reasonably.

[Fritz 0]

I had to run to a webinar with GM Ankit Rajpara so I forgot to mention that the Maia engine with the setup on their site and on Lichess is probably too weak for you. I'm 1700+ and have no problem with it. I guess it is possible to do some adjustment to get it stronger since the engine is lz0. I guess 1900 is lichess rating so 1500+ in real rating.

I played just one game against 1900 Maia at the time control of 10+5 or so and won rather easily, and I am pretty bad at those fast time controls. If they made something like 2100 Maia it might be a good sparring partner for me.