Why there is no interest in Computer with odds Vs Humans match?

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

Moderators: Harvey Williamson, bob, hgm

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
lkaufman
Posts: 4340
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by lkaufman » Mon Jul 01, 2019 5:33 am

adams161 wrote:
Mon Jul 01, 2019 4:26 am
Another problem with odds games is standard chess programs like Stockfish are not typically optimized to play them. The Alpha Beta search wants to find the best move and if that is trading queens it will do it in a heart beat. But the odds player usually uses some non best move intelligence like i need to get into certain positions, avoid to many trades etc which until you have chess programs programming for these types of games its a bit haphazard to say well give up a knights probably 2700. Would depend on how it wound up playing and what strategies that the human was using against it that without a heavy amount of code changes the engine would not defend against.
bo

Well you are quite correct that a program with no concept of how to maximize chances in a losing position won't play well giving knight odds. But Komodo, Stockfish, and Houdini all have a Contempt setting that tells them to value their own pieces more highly than the opponents', and Komodo also certainly knows to avoid equal trades when down in material even without a Contempt setting. Really, this is just basic chess knowledge, and any engine lacking this knowledge is an anachronism. Some of the Lc0 networks are very good at giving material handicaps, while others are atrocious at it; basically as long as the network chosen "thinks" that it still has at least a few percent winning chance at the specified handicap, it can play it well. Handicap play is a major part of chess history; something like 1/3 of recorded Morphy games were handicap games. Handicap play was still popular during the playing career of my first chess teacher, in the 1800s (crazy but true). Bobby Fischer played hundreds of handicap games for wagers (none recorded that I know of though), and even Gary Kasparov played a well-publicized high stakes two pawn handicap match. As for those who claim that handicap play isn't interesting, the Komodo vs Nakamura and MVL handicap matches had thousands of viewers, so apparently quite a few think it is interesting. Regarding the question of whether time odds or weak hardware odds would be more interesting than material odds, they are both of interest, but the former only tells us something about computers, whereas the later is a measure of how far from perfection even the best human players are. I find that to be quite an interesting question, although of course it only puts a lower bound on the answer, as we don't know how imperfect the best engines themselves may be.
Komodo rules!

Uri Blass
Posts: 8796
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by Uri Blass » Mon Jul 01, 2019 7:34 am

lkaufman wrote:
Mon Jul 01, 2019 5:33 am
adams161 wrote:
Mon Jul 01, 2019 4:26 am
Another problem with odds games is standard chess programs like Stockfish are not typically optimized to play them. The Alpha Beta search wants to find the best move and if that is trading queens it will do it in a heart beat. But the odds player usually uses some non best move intelligence like i need to get into certain positions, avoid to many trades etc which until you have chess programs programming for these types of games its a bit haphazard to say well give up a knights probably 2700. Would depend on how it wound up playing and what strategies that the human was using against it that without a heavy amount of code changes the engine would not defend against.
bo

Well you are quite correct that a program with no concept of how to maximize chances in a losing position won't play well giving knight odds. But Komodo, Stockfish, and Houdini all have a Contempt setting that tells them to value their own pieces more highly than the opponents', and Komodo also certainly knows to avoid equal trades when down in material even without a Contempt setting. Really, this is just basic chess knowledge, and any engine lacking this knowledge is an anachronism. Some of the Lc0 networks are very good at giving material handicaps, while others are atrocious at it; basically as long as the network chosen "thinks" that it still has at least a few percent winning chance at the specified handicap, it can play it well. Handicap play is a major part of chess history; something like 1/3 of recorded Morphy games were handicap games. Handicap play was still popular during the playing career of my first chess teacher, in the 1800s (crazy but true). Bobby Fischer played hundreds of handicap games for wagers (none recorded that I know of though), and even Gary Kasparov played a well-publicized high stakes two pawn handicap match. As for those who claim that handicap play isn't interesting, the Komodo vs Nakamura and MVL handicap matches had thousands of viewers, so apparently quite a few think it is interesting. Regarding the question of whether time odds or weak hardware odds would be more interesting than material odds, they are both of interest, but the former only tells us something about computers, whereas the later is a measure of how far from perfection even the best human players are. I find that to be quite an interesting question, although of course it only puts a lower bound on the answer, as we don't know how imperfect the best engines themselves may be.
For the last question you can try to find it by material odd of computers against computers when the side that gives odd use learning so does not lose the same game twice.

For example
Suppose one side is deterministic engine with 1 core with 1,000,000 nodes per move and no pondering.
Can you draw against it or even win against it without knight b1?

In order to reply the question you can use komodo without a knight b1 with 100,000,000 nodes per move.
If Komodo lose then play again and again when the idea is that komodo that gives the odds learn from the games and try to change the lines that it played.

The version of komodo that gives the odd can also extend in the search lines that the opponent played and prune lines that the opponent did not play because the assumption is that the opponent has to repeat lines that it played.


If even after 1000 games you have 1000:0 result then try lower material odd or lower number of nodes per move.
If you do not get 1000:0 result then you may try higher material odd or higher number of nodes per move.

Uri Blass
Posts: 8796
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by Uri Blass » Mon Jul 01, 2019 7:39 am

You can also set a competition between humans when the target is to beat or draw deterministic version of komodo with the maximal odd when participants simply need only to send the games.

They of course can use engines for producing the games because the target is not to show what humans can do only by themselves.

peter
Posts: 2160
Joined: Sat Feb 16, 2008 6:38 am
Full name: Peter Martan

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by peter » Mon Jul 01, 2019 9:39 am

Chessqueen wrote:
Fri Jun 28, 2019 4:47 am
There should be more computer with 2 pawns or Knight vs pawn odds that will bring back the challenge of Humans GMs vs top Engines. :shock:
I believe that top GMs like Nakamura or MVL can bring back the excitement of Computer Vs Humans.
I support your quest.
I'd find it even more interesting in times of a great new engine that's said to be playing much more "human- like" than the old ones did.
So it would be of even more interest to me to see, if this was true as for direct human- engine- matches of Leela with odds, (of time, pawns, figures, moves, whatever).
Would be fun to see, if LC0 really scores as well with e.g. Knight- odds against strong human players at least as well as komodo or SF would.
Need not be explicitely 2500+ GMs, as Larry already said, those guys are rather expensive and they often have to "fear" about their reputation even more than less famous ones do.

And then I'd find it most interesting too, to see a bookless Leela play corr.-chess without any odds against some corr.- IMs or GMs.

With defined and to the oppenent known hardware- time of Leela's, I guess good corr.- players with support of engines and databases of their choice should be well able to show up LC0's "human- like style" especially in opening, which is said to be the biggest strength of it.
Peter.

jp
Posts: 1406
Joined: Mon Apr 23, 2018 5:54 am

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by jp » Mon Jul 01, 2019 11:22 am

Nordlandia wrote:
Mon Jul 01, 2019 3:07 am
Another viable option is to use equilavent elo engine on generally obsolete hardware, in the 10-15 year range +/- CPU
Yes, not literally playing engines on old hardware, but trying to minimize the computing power needed to play well.

User avatar
Nordlandia
Posts: 2664
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by Nordlandia » Mon Jul 01, 2019 3:21 pm

Maybe odds matches in Fischer Random is harder for the human side. Humans are used to the standard opening setup. The unfamiliarity of Chess960 may give the engine the upper hand?

Maybe more odds is needed to equalize the chances in Fischer Random.

What is your thought on this one Larry?

lkaufman
Posts: 4340
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by lkaufman » Mon Jul 01, 2019 5:03 pm

Nordlandia wrote:
Mon Jul 01, 2019 3:21 pm
Maybe odds matches in Fischer Random is harder for the human side. Humans are used to the standard opening setup. The unfamiliarity of Chess960 may give the engine the upper hand?

Maybe more odds is needed to equalize the chances in Fischer Random.

What is your thought on this one Larry?
Well we already "proved" that this is so, John Meyer won 3 to 1 at regular knight odds vs Komodo, but lost 0-4 at frc knight odds, both 45' + 15". Komodo has no chance giving knight odds in normal chess to strong GMs at rapid (not blitz) time controls, but may have some chance at frc knight odds, depending on the time control and the rating of the GM. But even with frc, knight odds is a lot to give to a GM with enough time to think.
Komodo rules!

peter
Posts: 2160
Joined: Sat Feb 16, 2008 6:38 am
Full name: Peter Martan

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by peter » Mon Jul 01, 2019 5:19 pm

Nordlandia wrote:
Mon Jul 01, 2019 3:21 pm
Maybe odds matches in Fischer Random is harder for the human side. Humans are used to the standard opening setup. The unfamiliarity of Chess960 may give the engine the upper hand?
Of course, Jon.
I still think the easiest to organize and finance was the corr.- thing with the bookless LC0- engine on to the opponent known hardware.

TC could be lowered more and more down to freestyle, I think a bookless known engine- entity would have handicap enough against good centaurs, at least from 60'+15" upwards, at 120/40 the bookless to the opponent known engine, on which hardware ever running, should have much worse chances against manpower+good hardware+databases.
Peter.

Vinvin
Posts: 4767
Joined: Thu Mar 09, 2006 8:40 am
Full name: Vincent Lejeune

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by Vinvin » Mon Jul 01, 2019 7:55 pm

Some past matches Nakamura vs Komodo : https://www.youtube.com/results?search_ ... amura+odds

Chessqueen
Posts: 1079
Joined: Wed Sep 05, 2018 12:16 am
Full name: Nancy M Pichardo

Re: Why there is no interest in Computer with odds Vs Humans match?

Post by Chessqueen » Tue Jul 02, 2019 3:29 am

lkaufman wrote:
Mon Jul 01, 2019 5:03 pm
Nordlandia wrote:
Mon Jul 01, 2019 3:21 pm
Maybe odds matches in Fischer Random is harder for the human side. Humans are used to the standard opening setup. The unfamiliarity of Chess960 may give the engine the upper hand?

Maybe more odds is needed to equalize the chances in Fischer Random.

What is your thought on this one Larry?
Well we already "proved" that this is so, John Meyer won 3 to 1 at regular knight odds vs Komodo, but lost 0-4 at frc knight odds, both 45' + 15". Komodo has no chance giving knight odds in normal chess to strong GMs at rapid (not blitz) time controls, but may have some chance at frc knight odds, depending on the time control and the rating of the GM. But even with frc, knight odds is a lot to give to a GM with enough time to think.
I was also wondering what would be the outcome of pitting Komodo versus The fritz version that played versus Kramnik about 10 years ago Deep Fritz Bahrain with the f3 pawn or the f7 pawn odds in 50 games game in 30 minutes TC. What would your prediction be ?

Post Reply