INTERESTING Engines with Unique Styles (Unlike SF)

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

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Cornfed
Posts: 137
Joined: Sun Apr 26, 2020 9:40 pm
Full name: Brian D. Smith

Re: INTERESTING Engines with Unique Styles (Unlike SF)

Post by Cornfed » Sun Oct 11, 2020 9:16 pm

"Interesting" generally means 'not very good'. Sorry.

User avatar
AdminX
Posts: 5605
Joined: Mon Mar 13, 2006 1:34 pm
Location: Acworth, GA
Contact:

Re: INTERESTING Engines with Unique Styles (Unlike SF)

Post by AdminX » Sun Oct 11, 2020 9:26 pm

I am still a fan of MChess

Last edited by AdminX on Sun Oct 11, 2020 9:35 pm, edited 1 time in total.
"Good decisions come from experience, and experience comes from bad decisions."
__________________________________________________________________
Ted Summers

Ferdy
Posts: 4459
Joined: Sun Aug 10, 2008 1:15 pm
Location: Philippines

Re: INTERESTING Engines with Unique Styles (Unlike SF)

Post by Ferdy » Sun Oct 11, 2020 9:32 pm

BrendanJNorman wrote:
Sun Oct 11, 2020 3:01 pm
This is a fresh thread for Thorsten's idea of making a list of all "interesting" engines.

For me, in no particular order, they are:

Teki 2
Thinker 5.1c Passive
Baron 1.8.1
Gandalf 6
WChess 1.06
Vajolet 2.03
ProDeo 1.0-1.6 (assuming no personality used)
ZuriChess Graubuenden
Alfil 8.11
Stash 21.0 (new discovery...nice style to play against)
CM9000 (King 3.23)
Lambchop
Gromit 2.2
Horizon 4.3
Wyldchess 1.5.1
Detrich Kappe's "Distilled" Leela Nets like Bad Gyal etc
Winter
Zarkov
Trace

Tons more. Let's see your opinions. ;)
Could you list what makes an engine interesting? Like, likes to attack opponent's king than defends it own. Prefers mobile pieces in exchange for material, an engine that blunders like human giving away its queen or rook in a complicated position, etc. Perhaps there is rating range. Interesting engine in the range 1000 to 1200, and so on. With a given list hopefully programmers will make an attempt to create such interesting engine.

At certain level, it is fun to play an engine with limited depth. It may hold in middle as humans is not that accurate, but have issues when there is mate threat or in ending because it could not think deep enough.

carldaman
Posts: 2048
Joined: Sat Jun 02, 2012 12:13 am

Re: INTERESTING Engines with Unique Styles (Unlike SF)

Post by carldaman » Mon Oct 12, 2020 2:43 am

We need to define what "unlike SF" means. Is OpenTal like SF? You didn't mention it... :mrgreen:

BrendanJNorman
Posts: 2357
Joined: Sun Feb 07, 2016 11:43 pm
Full name: Brendan J Norman

Re: INTERESTING Engines with Unique Styles (Unlike SF)

Post by BrendanJNorman » Mon Oct 12, 2020 4:01 am

mvanthoor wrote:
Sun Oct 11, 2020 3:19 pm
BrendanJNorman wrote:
Sun Oct 11, 2020 3:01 pm
This is a fresh thread for Thorsten's idea of making a list of all "interesting" engines.

For me, in no particular order, they are:

Teki 2
Thinker 5.1c Passive
Baron 1.8.1
Gandalf 6
WChess 1.06
Vajolet 2.03
ProDeo 1.0-1.6 (assuming no personality used)
ZuriChess Graubuenden
Alfil 8.11
Stash 21.0 (new discovery...nice style to play against)
CM9000 (King 3.23)
Lambchop
Gromit 2.2
Horizon 4.3
Wyldchess 1.5.1
Detrich Kappe's "Distilled" Leela Nets like Bad Gyal etc
Winter
Zarkov
Trace

Tons more. Let's see your opinions. ;)
Most of those are either old, very old, or new engines that are still in development and have not reached the stage where the author starts to automatically tune the eval-function. Maybe I was right when I said that the hand-crafted evaluation function determines the engine's personality :) I hope that at some point, my engine will be in that list as well. (I actually intend to have a hand-crafted evaluation and a tuned one, at some point, for playing against people, or computers. IIRC, DanaSah does something similar.)

My personal picks would be Hiarcs (I only know the versions 10 and 12), and Fritz 10 and 11. (After 11, Fritz 12 and 13 got tuned to play more like Ryba, and suddenly Frans Morsch went on retirement, if I recall correctly.)
Well, for me an engine (outside of the Stockfish 12 and Leela world, which I use for analysis) needs to be weak enough to be a sparring partner that doesn't crush me like a bug.

I'm probably a little weaker than FM strength and very old engines tend to be more suitable, plus have handcrafted eval, so more "human-like" and with lower search depth (so they allow deep tactics) to boot.

I'm also a fan of the older Hiarcs and Fritz (Morsch versions)...a shame that they stopped development.

Wish I still had my Hiarcs 10 engine too...lost it! :(

BrendanJNorman
Posts: 2357
Joined: Sun Feb 07, 2016 11:43 pm
Full name: Brendan J Norman

Re: INTERESTING Engines with Unique Styles (Unlike SF)

Post by BrendanJNorman » Mon Oct 12, 2020 4:02 am

Madeleine Birchfield wrote:
Sun Oct 11, 2020 3:59 pm
I would like to add Minic NNUE with the Nascent Nutrient net trained by the Minic author himself using the Minic engine search and eval, it plays a very aggressive style that is very different from Stockfish according to Sylwy.
Sylwy wrote:
Fri Oct 09, 2020 10:21 am
Test conditions:
-TC=4'+2"
-Hash=256 MB
-GUI: Arena 3.5.1
-Book: Perfect_2010.abk for both engines
-default settings for both engines
-1 thread-CPU=Intel i5-7400-3GHz (Kaby Lake)
-OS: Windows 10 Home
-6-men Syzygy bases for both engines.

Texel 1.07 x64 has .......................3114 Elo points-CCRL Blitz
Minic 2.50 NNUE-Nascent Nutrient...around 3355 Elo points (3114+241)
Minic 2.50 NNUE-SV1520.............around 3458 (3451-Komodo 14 x64 +7)

Nascent Nutrient is about 100 Elo points below the best Segio Vieri nets.
A strong chess entity and an original play ! Congratulations !
xr_a_y wrote:
Fri Oct 09, 2020 10:33 am
Thanks a lot for testing.

Did you really see some "personalty" in this net ? Does it recall Minic playing style (if such thing exists ...) ? It is different from SV net playing style to your eyes ?
Sylwy wrote:
Fri Oct 09, 2020 11:29 am
xr_a_y wrote:
Fri Oct 09, 2020 10:33 am
It is different from SV net playing style to your eyes ?
Yes, it's definitely something different ! Almost all my NNUE tests used various versions of SV net. In Igel and Minic the difference is obvious. A very aggressive Minic. :wink:
I understand the frustration of the masters of handcrafted evaluation. A lot of years. A lot of work. But the finality is chess and here the progress is great......and inevitable. Someday the end of the new NNUE age will come. That's life !
xr_a_y wrote:
Fri Oct 09, 2020 11:45 am
This is interesting. A fear I have is that in time all nets may more or less converge to the same with LR.
This doesn't seem to be the case yet, but Nascent Nutrient hasn't be RL yet.
I'm glad this one shows some aggressiveness.
I think as NNUE authors stop using Stockfish to train their nets and start using their own engines, we would begin to see more diversity in engine playing style.
Is there a link to download? You're making me curious! :)

BrendanJNorman
Posts: 2357
Joined: Sun Feb 07, 2016 11:43 pm
Full name: Brendan J Norman

Re: INTERESTING Engines with Unique Styles (Unlike SF)

Post by BrendanJNorman » Mon Oct 12, 2020 4:05 am

mclane wrote:
Sun Oct 11, 2020 5:42 pm
Ok, I would like to begin with 2 dos engines:

Philidor by David broughton.
It runs in dosbox.
https://www.schachcomputer.info/forum/s ... php?t=6209


This dos engine for 8086 CPU was developed 1983 in assembler.

David Broughton got the championship title for dedicated chess computers with his chess champion MK V by scisys.
I like the human playing style.



The next dos Program is Gandalf 2.1 by steen suurballe.
I think I first met steen at the championship in Munich.
And hold contact in the years later.

http://talkchess.com/forum3/viewtopic.php?t=68797

Gandalf 2.1 is said to have 1 ply plus extensions.
Hi Thorsten, these Dos oldies look great. Do you have a direct download link? I'd love to download, challenge and review them. :)

BrendanJNorman
Posts: 2357
Joined: Sun Feb 07, 2016 11:43 pm
Full name: Brendan J Norman

Re: INTERESTING Engines with Unique Styles (Unlike SF)

Post by BrendanJNorman » Mon Oct 12, 2020 4:49 am

OliverBr wrote:
Sun Oct 11, 2020 8:32 pm
OliThink is certainly the most Unlike-SF engine. There are many others, too.
Okay you made me curious...

And so did Dan.
Dann Corbit wrote:
Fri Oct 09, 2020 7:09 am

Let's pause for a moment and consider Olithink.
It does not give a damn about King safety. You can pile wood around the king until the cows come home and he won't flinch.
Bishop pairs? Who cares?
All the eval terms we know and love are nothing.

And yet, it is about Glaurung strength. That is (quite frankly) absurd and flies in the face of chess knowledge painstakingly coded into most chess engines.
So I put 2 Olithink versions through a test and I also played a few blitz games myself.

Observations:

- Very unorthodox, but sound chess. Seems positional and materialistic, sort of like Karpov-style.

- Weak "king safety", but a great defender. Gives humans a feeling that we can get *close* to forcing checkmate, unlike "perfect" engines.

- Really likes to REDUCE opponent's mobility, just as much, if not more than to increase its own.

Examples:

First I'll share a blitz game I played as black vs Olithink (2479 CCRL version) which I could/should have won, but missed the critical moment.

I played in my usual dynamic style, as if I were playing a human, sacrificed an exchange in a rather thematic manner (color-complex domination, pawn and initiative in return, though still a bit speculative) and after a few attacking moves I was completely winning.

Then after I missed my moment, Oli defended like a champ (also very human like) and I had no chance.

Let's see...Don't forget, this was 3 2! lol :)



Key Positions:

Initial Exchange Sac

Is my move 22...Rd4 sound? Looks good to me! :)



The rook lift 24...Re5! is very strong, position is very dynamic, probably better for black.



I missed a clear win (which I'd easily see if it weren't blitz) with 25...Ne2+ and 26...Qf4 (with the idea of 27...Be5) and tried instead to pry open the kingside with my h-pawn.



After missing these opportunities, I fumbled around a bit and Oli shut me down effectively.

Much like a strong human would, once the moment of danger has passed.

Example 2: This was a game played against a newer version of Olithink, by a super human-like engine I am secretly working on with Pawel (Rodent author), but the spotlight in this instance, is on the way Oli slips out of the noose and manages to get into a decent queen an rook endgame.

Fantastic defense, despite losing in the end.



This game is self-explanatory, but there was a funny trap our "mystery engine" set, which poor Oli fell right into.

Playing 48.g4!? here looks ridiculous, but of course, with the outside passed pawn on the a-file we can afford to set such traps.

Since 48...Qxg4+ is losing, I guess white's move is also objectively decent as well.



Conclusion:

Olithink is a great engine and definitely being added to my preferred sparring partners.

Weak king safety combined with solid positional play and excellent defense (and falling into traps occasionally!) makes it the perfect human-like sparring partner.

(Is this enough detail, Thorsten? ;) )

BrendanJNorman
Posts: 2357
Joined: Sun Feb 07, 2016 11:43 pm
Full name: Brendan J Norman

Re: INTERESTING Engines with Unique Styles (Unlike SF)

Post by BrendanJNorman » Mon Oct 12, 2020 5:07 am

Cornfed wrote:
Sun Oct 11, 2020 9:16 pm
"Interesting" generally means 'not very good'. Sorry.
No need to be sorry, you misunderstood our intentions.

Firstly, 'not very good' is a subjective descriptor.

Perhaps, you mean ' not very strong' which would be true in a lot of cases in the context you stated.

But for us, good means something totally different.

We are the type of computer chess fan who actually looks at the games...

We actually play against the engines.

We care nothing for analyzing stats, code similarities, and excel spreadsheets full of results and "Elo gain".

We care nothing for playing thousands of "micro-bullet" games in the console.

In short, we are chess nerds, not computer nerds. :lol:

Anyway, thanks for your negative input, much appreciated.

BrendanJNorman
Posts: 2357
Joined: Sun Feb 07, 2016 11:43 pm
Full name: Brendan J Norman

Re: INTERESTING Engines with Unique Styles (Unlike SF)

Post by BrendanJNorman » Mon Oct 12, 2020 5:26 am

Ferdy wrote:
Sun Oct 11, 2020 9:32 pm
Could you list what makes an engine interesting? Like, likes to attack opponent's king than defends it own. Prefers mobile pieces in exchange for material, an engine that blunders like human giving away its queen or rook in a complicated position, etc. Perhaps there is rating range. Interesting engine in the range 1000 to 1200, and so on. With a given list hopefully programmers will make an attempt to create such interesting engine.
carldaman wrote:
Mon Oct 12, 2020 2:43 am
We need to define what "unlike SF" means. Is OpenTal like SF? You didn't mention it... :mrgreen:
Hi Carl and Ferdy,

For me, the ideal "sparring partner" is low depth (maybe around 8 ply or so, like strong humans) but has enough knowledge to make it judge postions in a human-like way.

Human-like means:

- Underestimating king safety issues (see: Olithink), so that you actually have chances to attack like a human.

- Having exaggurated postional preferences (rather than balanced, like an engine). This could mean being obsessed with pawn structure, outposts, better minor pieces...i.e strong bishop vs knight, and control of the center (for a positional opponent...I know tons of humans who play like this and miss tactics often due to myopia haha!) or being obsessed with creating a naked enemy king (so will sac like a madman to make it happen), or anything in between.

- Having a narrow opening book which resembles a human's (so for example, ONE suitable response to 1.d4 and 1.e4, ONE main repertoire as white, etc - easy to do with polyglot books).

- Variety. Not sure how to explain what I mean here, but Thinker used to have a feature where it would occasionally choose the second or third best move in PV, as long as it is within a certain centipawn range. This results is minor mistakes/innacuracies and more human-like/less "perfect" play. I believe Pawel did similar in Rodent with his "Evalblur" option. This kind of simulates the human analytical process where you're considering 2-3 seemingly equal options, you get low on time and think "screw it, I'll play x!".

- Contempt or Negative comtempt if you want to train how to keep it complicated agaist those annoying guys who try to play for a draw.

And so on. I basically think about my own tournament games, and think "how could we simulate this experience/feeling in computer play?"

Really, want to study programming, but finishing Mandarin first! :D

P.S. Carl, I didn't mention OpenTal because it would be like bragging, since I (and Pawel) gave birth to it. ;)

Post Reply