Help with Komodo 1.0 personalities.

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

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
Ovyron
Posts: 2031
Joined: Tue Jul 03, 2007 2:30 am

Re: Help with Komodo 1.0 personalities.

Post by Ovyron » Fri Jan 05, 2018 12:22 pm

leavenfish wrote:Seriously, spend more time getting better at the game instead of pushing engines around
I have done both and have improved 650 elo in 10 years. I have found that a big chunk of this improvement was thanks to my analysis and insights of correspondence chess, with engines, and I reckon that if my involvement with chess didn't include engines my improvements would only have been about 300 elo or something.

Sadly, I have found the ceiling and apparently the only way to go forwards is to change my style and play more solid :( - whenever I open the position with e4 and play my gambits stronger players just destroy me, if I close it with d4 and keep as much material as possible on the board they tend to panic (specially when they Berserk against me on lichess), force an attack that doesn't work, sack material or try anything, and if I hold, I win, even if they're 200 or 300 elo stronger. So apparently my next 300 elo will be like this, play solid, avoid blunders, and games will be won.

I don't get why you bring up the topic of my strength into this, though.
I'm serious - Argue all you want....it's a bad move, plain and simple...by any measure of what constitutes a 'good' or 'bad' move.
I'm not claiming that the move is good. I'm claiming that players like Tal used to play such bad moves, and win games with them. When games get psychological it's possible advancing your passed pawn when an opponent attacks one of your pieces is going to make them move all thier pieces to try to stop it letting you win the game, even though there's some deep tactic that would have lost the pawn, so objectively the move is bad, but if no other move would have won you the game then it's clearly a great choice to make.
a horrible move is a horrible move, regardless of who plays it.
No, a horrible move depends on who is it played against. The only reason this 35. Re5 lost is because the opponent was strong enough to refute it, but the world is full of chess entities that would have lost against 35. Re5, so the move is playable, it doesn't really matter what a perfect chess entity would say about the move, since we don't have it here to tell us about the quality of moves.

User avatar
Ovyron
Posts: 2031
Joined: Tue Jul 03, 2007 2:30 am

Re: Help with Komodo 1.0 personalities.

Post by Ovyron » Fri Jan 05, 2018 12:25 pm

Rebel wrote:[Confusion = 100]

Doing a mouse-over this option in the Personality Creator and it states -

Confusion is an improved version of the ANTI-ANAND algorithm we used playing Vishy Anand back in 1998. The algorithm tries to complicate the position for tactical shots and then profit from the created chaos by outsearchng the human opponent. The default value of [100] means the algorithm is inactive, other values than [100] will activate the slgorithm. Values that make sense are in the range of 25-50-75-125-150. A value of [500] will make wild sacrifices and are fun to watch but bad for the overall playing strength.
That's very cool!

I'm still using ProDeo 2.2, with this new info I feel as if I was living in the stone age 8-)

BrendanJNorman
Posts: 1262
Joined: Sun Feb 07, 2016 11:43 pm
Full name: Brendan J Norman

Re: Help with Komodo 1.0 personalities.

Post by BrendanJNorman » Fri Jan 05, 2018 1:30 pm

Ovyron wrote:
BrendanJNorman wrote:Shall we produce our own ratings list, including learning engines and personalities as part of the mix? :twisted:
I'd much rather have a list that measured an engine's style instead of its strength... it could be instead of the elo rating, the... uly rating :lol:

But seriously, it could be trivial and fast, because an engine's style shows, we don't need 1000 games to measure it, a few wins against another engine can suffice, we extract the score from the best looking moves that it makes for attractive games.
Maybe all we need is a table with scores given based on themes present in the game - something like:

Tier 1 (100 points) - Sacrifices:
  • Queen Sacrifices

    Piece Sacrifices

    Exchange Sacrifices

    Pawn Sacrifices
Tier 2 (100 points) - Aesthetics:
  • 2 or more pieces hanging at once (covered by tactics)

    King march with queens still on board

    Underpromotion
Tier 3 (100 points) - Thematic:
  • Rook Lift

    Menacing Pawn Storm (not from book moves, i.e Dragon Sicilian or similar)

    Dynamic domination of color complex
Something like this with rating "points" only scored from won or drawn games is how it should be IMO.

This list here would be the "checklist" for rating aggressive engines, whilst a similar tri-tiered checklist for positional engines would include things like Superior Minor Piece, Crushing Bishop Pair Use, Dominating Outpost, Intelligent Maneuvering of a Piece (such as after a...Bxf3 capture, playing Nd7-f8-e6-d4 maneuver to occupy a weak d4 square) and things like this.

Again, you use a theme checklist when rating games and assign a ratio score.

For example OpenTal might have a 350/12 "style rating" which would mean he has accumulated 350 "style points" from 12 games.

This seems consistent, and easy over the long-term to see which engines are stylish and which aren't.

Of course lots of other themes can be added, these were just examples.

BrendanJNorman
Posts: 1262
Joined: Sun Feb 07, 2016 11:43 pm
Full name: Brendan J Norman

Re: Help with Komodo 1.0 personalities.

Post by BrendanJNorman » Fri Jan 05, 2018 1:33 pm

Rebel wrote:
Ovyron wrote:We could finally be able to tell if the Anti-GM setting of ProDeo really does improve its style or if it's only useful against humans.
Just for the record the Anti-GM setting doesn't exist any longer. It has been replaced with something better -

[Confusion = 100]

Doing a mouse-over this option in the Personality Creator and it states -

Confusion is an improved version of the ANTI-ANAND algorithm we used playing Vishy Anand back in 1998. The algorithm tries to complicate the position for tactical shots and then profit from the created chaos by outsearchng the human opponent. The default value of [100] means the algorithm is inactive, other values than [100] will activate the slgorithm. Values that make sense are in the range of 25-50-75-125-150. A value of [500] will make wild sacrifices and are fun to watch but bad for the overall playing strength.
Yep, I've made some interesting personalities using this feature in ProDeo 2.6 - will share as soon as I get a chance to do a final test on them. ;)

BrendanJNorman
Posts: 1262
Joined: Sun Feb 07, 2016 11:43 pm
Full name: Brendan J Norman

Re: Help with Komodo 1.0 personalities.

Post by BrendanJNorman » Fri Jan 05, 2018 1:39 pm

Ovyron wrote:...specially when they Berserk against me on lichess
What's your liChess handle? Let's play a few games one day :)

User avatar
Ovyron
Posts: 2031
Joined: Tue Jul 03, 2007 2:30 am

Re: Help with Komodo 1.0 personalities.

Post by Ovyron » Fri Jan 05, 2018 2:41 pm

BrendanJNorman wrote:Maybe all we need is a table with scores given based on themes present in the game - something like:

Tier 1 (100 points) - Sacrifices:
  • Queen Sacrifices

    Piece Sacrifices

    Exchange Sacrifices

    Pawn Sacrifices
Tier 2 (100 points) - Aesthetics:
  • 2 or more pieces hanging at once (covered by tactics)

    King march with queens still on board

    Underpromotion
Tier 3 (100 points) - Thematic:
  • Rook Lift

    Menacing Pawn Storm (not from book moves, i.e Dragon Sicilian or similar)

    Dynamic domination of color complex
Yes, I like it! The great thing about starting with such a list is that, after we put it in practice we can tweak it after we see the results it outputs. It should be pretty obvious when some amazing engine doesn't get as many points or when some dull engine somehow abuses the system (say, Komodo 10 topping the list even though we all agree its style isn't that interesting :shock: )

It should also be pretty obvious when the engine makes some very attractive move and we're not rewarding it, what are the themes that we're missing.
BrendanJNorman wrote:Something like this with rating "points" only scored from won or drawn games is how it should be IMO.
Okay, what about only awarding half the points on draws? I can imagine a group of four engines, the first two play nice but draw all their games, while the third and fourth wins and loses half of them, this would balance things out because we don't care about results.

User avatar
Ovyron
Posts: 2031
Joined: Tue Jul 03, 2007 2:30 am

Re: Help with Komodo 1.0 personalities.

Post by Ovyron » Fri Jan 05, 2018 2:42 pm

BrendanJNorman wrote:
Ovyron wrote:...specially when they Berserk against me on lichess
What's your liChess handle? Let's play a few games one day :)
Hehe, my nick appears on your post... I'm Ovyron at lichess! Sure we can play some games :)

leavenfish
Posts: 248
Joined: Mon Sep 02, 2013 6:23 am

Re: Help with Komodo 1.0 personalities.

Post by leavenfish » Sat Jan 06, 2018 3:21 pm

BrendanJNorman wrote:
leavenfish wrote:Seriously, spend more time getting better at the game instead of pushing engines around and you might be able to judge a blunder from a decent move.
Ok.

Can I ask a couple of questions Brian?

1. How strong are you at chess? (Just curious given the "get better at chess" request from you which came across as a little condescending).

2. Why would you presume that Lev Aronian uses an engine that is "several years old" because it "suits his style"?

3. Why was Anand using Hiarcs in an age when Rybka, Stockfish and Komodo were available?

4. Why are these elite players not aware that the engines they choose, play "weak chess"?

Just curious Brian, no need to get all angry.
No one (me anyway) is 'angry here'...just pointing out what I see as the reality of one of the situations noted.

1. How strong are you at chess? (Just curious given the "get better at chess" request from you which came across as a little condescending). Well, I passed 2100 USCF about 25 yrs ago and stayed over 2000 most of that. I have to say that all the 'quick play' at the local club had taken a meat cleaver to my rating of late...if this is what you mean by 'strong'. My ICCF was 2399 about 13 yrs ago when I quit because of the rampant engine use.
My 'get better at chess' comment is directed to anyone who in their heart of hearts things that Rook sacrifice had anything going for it. It's just bad.


2. Why would you presume that Lev Aronian uses an engine that is "several years old" because it "suits his style"? How does he use it (well any GM I suppose), would be to check for novelties, or check them out and let them crunch away while they do other things. You can clearly keep using a 'several year old' engine as the calculation strength is there to find such things...heck, remember how Kaspy and his team use them way back in the day? I don't think Aronian would consider ...Re5 'blundering a piece' as anything akin to 'style'.

3. Why was Anand using Hiarcs in an age when Rybka, Stockfish and Komodo were available? See the answer above! Fast and strong is fast and strong...we don't all need to drive Ferrari's you know. Many engines can get the job done...the job a GM would use them for especially

4. Why are these elite players not aware that the engines they choose, play "weak chess"?
I can't even fathom why you ask this leading question. None of them are using the particular Komodo engine under discussion. Many 'free engines' will get them what they need - looking for novelties or how difficult it is likely to be for an opponent to handle a novelty. There is probably an inverse correlation between what a top level GM 'needs' and the engine he needs...maybe not, but if you think about it, they are already super GM's...playing their same openings over and over again, they already have more than a 'feel' for them...but they look for any way they can to get a little edge but looking for things their (often the same) opponents might be ready for

So, I've answered your questions. And a bad move is still a bad move.

It's...lets just say 'lazy thinking' to think of Re5 as being a product of style....and I all too often see people point to such moves and shout "style!"...the difference between how a Capablanca and Alekhine might approach a postion can rightly be said to involve conflicting styles (or Kasparov and Karpov for another example)...but a bad move is just a bad move.

BrendanJNorman
Posts: 1262
Joined: Sun Feb 07, 2016 11:43 pm
Full name: Brendan J Norman

Re: Help with Komodo 1.0 personalities.

Post by BrendanJNorman » Sat Jan 06, 2018 3:55 pm

leavenfish wrote:Well, I passed 2100 USCF about 25 yrs ago and stayed over 2000 most of that.
So you're still weaker than me, so please stop with the insulting "get better at chess" type comments.

It's my right to say that Re5 is a stylish move, so telling people of this opinion to "man up" and agree with your assertion is impolite.

Would you like me to show you some of the moves Tal, Kasparov and Nakamura have played in blitz?

Moves that even Crafty would poke holes in?

Why do these GMs like these guys get to keep the label as "stylish", even when they make dubious moves, but Komodo Kinghunter doesn't?
leavenfish wrote:Remember how Kaspy and his team use them way back in the day?
Yes, I do indeed. Kaspy used first Deep Junior 6 and then Deep Junior 7, two "stylish" engines which happen to be about 200 Elo points weaker than the Komodo Kinghunter personality.
leavenfish wrote:I don't think Aronian would consider ...Re5 'blundering a piece' as anything akin to 'style'.
He makes this type of move regularly in blitz, do you think he hates his own playing style?
leavenfish wrote:I can't even fathom why you ask this leading question.
Because you aluded that Komodo Kinghunter plays weak chess, despite the fact that it is hundreds of Elo stronger than the engines Kasparov used in his prime.

If he (Kaspy) had Komodo Kinghunter in 1999-2001, Deep Junior 6/7 would have been thrown in the Recycling Bin.
leavenfish wrote:It's...lets just say 'lazy thinking' to think of Re5 as being a product of style....and I all too often see people point to such moves and shout "style!"...
So, to follow this logic further, could we also say that Mikhail Tal played weak moves, and that most of his sacrifices were not "style", but poor chess?

His moves often give SF a heart attack much worse than Re5 does.
leavenfish wrote:The difference between how a Capablanca and Alekhine might approach a postion can rightly be said to involve conflicting styles (or Kasparov and Karpov for another example)...but a bad move is just a bad move.
How? Revisit my question about Tal... Despite being known as perhaps the greatest attacker in history, shall we now relegate him to the weak players category? Alekhine too played sacrifices as well, that very often cannot pass the scrutiny of a strong chess engine, so shall we write him off too?

Let's see a description of this STYLE in action...
With such intuitive sacrifices, he (Tal) created vast complications, and many masters found it impossible to solve all the problems he created over the board, though deeper post-game analysis found flaws in some of his conceptions. Although this playing style was scorned by ex-World Champion Vasily Smyslov as nothing more than "tricks", Tal convincingly beat every notable grandmaster with his trademark aggression.
All weak chess, right?

matejst
Posts: 153
Joined: Mon May 14, 2007 6:20 pm

Re: Help with Komodo 1.0 personalities.

Post by matejst » Sat Jan 06, 2018 3:57 pm

The question of style, in chess engines, is very complicated. Graham, e.g., thinks Gaviota is an aggressive engine, an attacker, but I use him mostly in simple, endings-like positions, where it excels. In a certain way, we are both right: Gaviota, although with a heavy eval, won't hesitate to throw its pawns forward and to play an active -- and probably risky -- brand of chess.

What is certain, about style, is the following: I have noticed that some engines find books moves in openings, moves a human would play in middlegames, others do not. I can also notice that some engines are more aware of tactical threats than others at lower depths.

I tried Frenzee these days, and I mostly like what I saw: his moves make sense to me. But its game is very active, and I don't have a shadow of doubt than, in an adequate opening, it will find very sharp lines.

It's different from the world of humans: engines don't avoid complications. They don't choose the kind of positions they would like to play -- we do. While I can tweak a few engines to play insane attacking chess, I can't tweak them to choose sharp or quiet positions from the get go. When I managed to change just a bit Zarkov 6.55 to play something that looks like Karpov's style, I worked in the dark. It worked well, although I didn't have a clue about what I did.

There are only two fully tweakable engines: Pro Deo and Rodent, with the ability to really have personalities. But it includes the possibility to choose a repertoire, and there are so many parameters that I managed to create a Spielman while trying to make a Rubinstein.

For the rest: the distinction has to be really... nuanced, well thought of. And I don't think that we can make trully good personalities: the engine authors can, since they are the only ones to fully understand what they have created.

Post Reply