![[us]](/mwf/v12/flags/us.png "United States")
I thought this subject was so interesting that it demands a thread started.(From a previous post)What do you think with all this talk about the various chess players,a thought came to my mind.Why don't you guys(Vas and Larry) implement some sort of "personalities" setting for Rybka having to choose between lets say a Fischer style,Capablanca,Karpov(positional genious") Kasparov,Tal(let the sacs commence).A personality setting in some version would be interesting though Rybka is fine as she is,just a thought.So then is the idea like this?For a Tal setting for example,would a programmer award points in a position if for example the pawns protecting the king are torn away by sacs?Vas how would one implement a personality setting lets say from a posiotional style player like a Karpov compared to more dynamic players like a Tal or Kasparov?What about a Capablanca setting(Super correct positional style). It seems of the "Correct" players Capa and Botvinnik would score high such that maybe a big change in a program for thjose players would not be such a drastic change.I am assuming of course I am no programmer,so what does everyone think about a personalities setting for Rybka in the future.we must first concentrate on version 3.0 of course.
FWCC
Future World Chess Champion
The Future Is Now
FWCC
Future World Chess Champion
The Future Is Now
It would be pretty easy to give the user an option to raise or lower the value of positional factors relative to material. A high value would mean playing like Tal, a low value would mean something more like Capablanca or Karpov. But of course there are many other factors besides willingness to sacrifice that distinguish one great player from another. Some of these have to do with opening choices (not our department, as the book is totally separate from Rybka) and some with factors that are very hard to describe or program.
Right, but putting together the historic opening repertoire of chess greats should, in theory, be a piece of cake.
As I see it you'd first create a basic book consisting of all games world champions (or the player set) ever played collectively, and then you'd create individual books from this basic mold for each "personality", sharply increasing the likelihood of lines each player actually played. (You'd need to do it this way because some greats may have less than 1000 recorded games and it could get very monotonous if they always played the same exact lines. Just because they didn't historically play certain lines doesn't mean they couldn't have, but to get the right flavor of things you would make it unlikely for "Steinitz" to play cutting-edge Sicilians.)
Then it would be a matter of designing easy-to-use parameters similar to Chessmaster, though you wouldn't need that many levers: the simpler, the better. (These personalities should play at roughly their peak historical strength level regardless of computer system being used--I refer you to the work of Jeff Sonas to get an appreciation of how historical players ranked. Meaning that if you pit Rybka at full strength against any GM personality Rybka would mop the floor with almost all of them.)
In my opinion this would be a big hit. While I am with you guys in that the top priority is and always must be to increase strength, the second priority ought to be commercialization, no? Personalities would be fascinating to play against each other in tournaments. I have to tell you, this is how I really got into computer chess many years ago--Chessmaster personalities hooked me.
As I see it you'd first create a basic book consisting of all games world champions (or the player set) ever played collectively, and then you'd create individual books from this basic mold for each "personality", sharply increasing the likelihood of lines each player actually played. (You'd need to do it this way because some greats may have less than 1000 recorded games and it could get very monotonous if they always played the same exact lines. Just because they didn't historically play certain lines doesn't mean they couldn't have, but to get the right flavor of things you would make it unlikely for "Steinitz" to play cutting-edge Sicilians.)
Then it would be a matter of designing easy-to-use parameters similar to Chessmaster, though you wouldn't need that many levers: the simpler, the better. (These personalities should play at roughly their peak historical strength level regardless of computer system being used--I refer you to the work of Jeff Sonas to get an appreciation of how historical players ranked. Meaning that if you pit Rybka at full strength against any GM personality Rybka would mop the floor with almost all of them.)
In my opinion this would be a big hit. While I am with you guys in that the top priority is and always must be to increase strength, the second priority ought to be commercialization, no? Personalities would be fascinating to play against each other in tournaments. I have to tell you, this is how I really got into computer chess many years ago--Chessmaster personalities hooked me.
Very interesting Hernandez.So then can we say that in my opinion a "personality" setting is in effect a collection of games from a famous GM with a formula applied to come up with a "most likelihood move" for the player selected,lets say a Karpov.We collect thousands of games that Karpov has played and then assign some sort of algorithm percentage to the moves he played in certain positions.What we would have is a collection of games stored in a program(Rybka in this case) and then apply the formula to come up with a move based on this,the problem is you would have to have stored a great amount of games from the selected players(collectively)you have in a personality profile which would eventually take up a lot of space ,it would be good if the games could be compressed to a bitbase format in the program.Maybe the move could be achieved through some sort of function that a tablebase or bitbase extension uses.So what i'm trying to say is maybe the moves chosen from a player selected would be achieved from a formula applied to a game selection of "that" player.
Regarding the work of Jeff Sonas, he makes no effort to compare chessplayers of different eras. Instead, he defines a rating as representing a certain percentile (something like 2600 means the top 1%, I forget the details). So he is not making any claim that a 2600 from the 1800s plays as well as a 2600 today, he doesn't even try to answer that question. I'm inclined to agree with Kasparov that chess has advanced tremendously, and that the top players of the past would have no chance against today's stars. Kasparov remarked that "even Ljubujevic (who had just finished last in a top tourney) would beat Capablanca in a match", thereby managing to insult both of them in one sentence! But I'm afraid he is right. The greats of the past were not lacking in talent, but we have learned from our predecessors. My feeling is that even Morphy, perhaps the greatest chess genius of them all, would have trouble against today's "weak" grandmasters, if he had to play with the knowledge he had during his life.
Hello Larry,im not so sure Larry.Capablanca for instance was a natural geniuos and I feel if he was alive today he would STILL be in the top 10 players in the world.His talent was so great that I would dare to say he would still have a chance to be world champion even today.Capa was a natural genious a child prodigy.Some of the greats from the past would even today beat our top GM's.Yes,even Morphy would be a WC contender.I'm talking about based on "natural talent" not the chess principles from the past.Fischer had great talent but it still took him some years to become WC.So talent plus work is the key.But there are some players from the past that would do well today.Maybe if we get this personality setting we could run fantasy matches.LOL
![[cr]](/mwf/v12/flags/cr.png "Costa Rica")
I think that Larry is right.
Capablanca was a great genius, but in nowadays chess is completely different of that time, and today, the opening theory has advanced too much. If he could return and play today, I doubt too much that he can win against today Top 10 players. He would be in a big big disadvantage, just from the beginning against practically any strong GM. You can have a great talent, but without study and enough good preparation, I strongly believe that today it is not possible to make that kind of prowesses. Through the time, the way in how humans play chess have changed a lot. This is why Fischer tried with his Fischer random game. Today in day, there are a lot of laboratory players and it is quite difficult to fight against them if you do not has enough and very good opening knowledge. It doesn't import, how big can be your talent.
Regards,
Gambito.
Capablanca was a great genius, but in nowadays chess is completely different of that time, and today, the opening theory has advanced too much. If he could return and play today, I doubt too much that he can win against today Top 10 players. He would be in a big big disadvantage, just from the beginning against practically any strong GM. You can have a great talent, but without study and enough good preparation, I strongly believe that today it is not possible to make that kind of prowesses. Through the time, the way in how humans play chess have changed a lot. This is why Fischer tried with his Fischer random game. Today in day, there are a lot of laboratory players and it is quite difficult to fight against them if you do not has enough and very good opening knowledge. It doesn't import, how big can be your talent.
Regards,
Gambito.
Nonetheless, Capablanca didn't have much of an openings repertoire and he put very little emphasis into studying openings. I think that most people agree that he was, as a player, much stronger than Alekhine, and yet he lost quite a long match in 1927. Thus, it shouldn't be difficult to imagine that he would also lose to a player who is weaker than Alekhine in raw talent but who has more chess knowledge accumulated over the generations. I think that the only player of previous generations who would not get demolished by each and every one of the top 20 in the world today is Fischer.
![[il]](/mwf/v12/flags/il.png "Israel")
I do not understand why do you think that most people will agree that Capablanca was stronger than Alekhine.
I think that most people will say simply that they do not know because they know too little about chess to have an opinion about it
and even if they have an opinion you cannot trust their opinion.
I remember that when I was a child somebody told me that he believes that another player is stronger than me simply because of the fact that this player won faster against him(both of us could win against him easily)
I think that it is nonsense and if 2000 player play 2 games against players when one of them has rating of 2500 and one of them has rating of 2700 he will not be able to know who is the stronger player based on his losses.
The same is correct for Capablanca and Alekhine.
Players who are at least 400 elo weaker than both of them has no basis to give an opinion who is stronger out of them.
I think that most people will say simply that they do not know because they know too little about chess to have an opinion about it
and even if they have an opinion you cannot trust their opinion.
I remember that when I was a child somebody told me that he believes that another player is stronger than me simply because of the fact that this player won faster against him(both of us could win against him easily)
I think that it is nonsense and if 2000 player play 2 games against players when one of them has rating of 2500 and one of them has rating of 2700 he will not be able to know who is the stronger player based on his losses.
The same is correct for Capablanca and Alekhine.
Players who are at least 400 elo weaker than both of them has no basis to give an opinion who is stronger out of them.
Nonetheless, from my experience, I think it's quite correct that the strong players view Capablanca as a stronger chess player than Alekhine.
![[de]](/mwf/v12/flags/de.png "Germany")
About Jeff Jonas rating list, you are not right. You can make rating lists over 25 years (and more?). Jaan Ehlvest was in this lists in the top ten in the the first eightiens (and it´s true!). Jaan would surely win against Jose Raul, because he is better. And you can be sure, that your hero Morphy will lose a match against 13 years old Hou, Yifan (WGM; 2523 Elo).
Just look at the responses, Larry. There is one camp that agrees with you and derides the gifts of the old masters and another that glorifies the oldies. I am neutral; I utterly lack the knowledge to side with one or the other. But this argument is the whole point, creation of personalities would stimulate debate and endless bickering! That has to be good for business.
Clearly you'd need to lay down some basic assumptions and parameters first. For instance, something like...
1. Personality strength is assumed to be at approximately their career peak; personalities should play in a 2650-2850 range
2. Openings played follow a three step sequence...
a. If position was played historically by this GM, moves they actually played will be strongly favored.
b. If position was not played historically, blend moves other GMs played with this player's evaluation profile to arrive at a decision.
c. Once a position is attained that is not in book all players will rely entirely on their evaluation profile.
3. Evaluation profiles are a set of adjustments to Rybka's engine that modify evaluation weights, search depth, contempt, etc. I think it would be kind of interesting for some of these ratings to be within a "range", with the actual numbers used determined before the game starts through some randomization (so the personalities won't play exactly the same every time).
4. Playing strength should not be affected by hardware. Personalities should be capped at some set processing speed.
Anyway...I know this is a diversion from your core mission, so I can imagine your reflexive avoidance of going down this time-consuming path. But personally, I think it would be a hit--especially if the GUI allowed tournaments.
Clearly you'd need to lay down some basic assumptions and parameters first. For instance, something like...
1. Personality strength is assumed to be at approximately their career peak; personalities should play in a 2650-2850 range
2. Openings played follow a three step sequence...
a. If position was played historically by this GM, moves they actually played will be strongly favored.
b. If position was not played historically, blend moves other GMs played with this player's evaluation profile to arrive at a decision.
c. Once a position is attained that is not in book all players will rely entirely on their evaluation profile.
3. Evaluation profiles are a set of adjustments to Rybka's engine that modify evaluation weights, search depth, contempt, etc. I think it would be kind of interesting for some of these ratings to be within a "range", with the actual numbers used determined before the game starts through some randomization (so the personalities won't play exactly the same every time).
4. Playing strength should not be affected by hardware. Personalities should be capped at some set processing speed.
Anyway...I know this is a diversion from your core mission, so I can imagine your reflexive avoidance of going down this time-consuming path. But personally, I think it would be a hit--especially if the GUI allowed tournaments.
![[hu]](/mwf/v12/flags/hu.png "Hungary")
Nelson, not everything is a matter of the opening repertoire :)
Exposing many eval parameters for customers to tinker with would be easy. However, KISS is working against this. Who really wants an engine pane full of things nobody has any clue about?
Having special Rybka versions which play in the style of various players should be a doable task. Actually, I had the idea of developing some automated statistical procedure for this, which could then be applied to any human player (or more precisely to any collection of games). So, you could for example create a Rybka version which plays in the style of Nelson Hernandez. This version would for example give extra priority to doing weird things with her rook pawns, while neglecting other pawns & pieces :)
It's all on the long list of cool ideas for later.
Vas
Exposing many eval parameters for customers to tinker with would be easy. However, KISS is working against this. Who really wants an engine pane full of things nobody has any clue about?
Having special Rybka versions which play in the style of various players should be a doable task. Actually, I had the idea of developing some automated statistical procedure for this, which could then be applied to any human player (or more precisely to any collection of games). So, you could for example create a Rybka version which plays in the style of Nelson Hernandez. This version would for example give extra priority to doing weird things with her rook pawns, while neglecting other pawns & pieces :)
It's all on the long list of cool ideas for later.
Vas
I understand why it is on the list for later, Rybka being a primarily analytical tool, but I have to tell you, I bet you would increase your market share much more than you would think by making options such as:
1) Play in the style of Tal, Karpov, Fischer, etc.
2) Play with various odds (pawn, pawn and move, and knight, rook, queen(!))
3) Play to win even when down material, that is, a more flexible and sophisticated contempt factor
The reason I say this is that none of these things have ever really been done effectively elsewhere. Yes, you can make programs more or less aggressive, but they aren't consistent and don't have much of a style. You can take away a knight, but they don't "realize" they are playing an odds game, and don't really try for the win or understand contempt well enough.
I understand that in each case Rybka would have to play at a little less than its best to accomodate the feature, but the pleasure it would give chess players would be immense. Imagine playing against Tal with him giving you knight odds! It would be a completely new set of features that would appeal to a wider range of chessplayers.
I have a friend who is a candidate master, and who has played other top programs over the years. He said that when he lost to them, he never felt he learned anything much about chess-- he just felt that he was outcalculated, and that he felt he played better positionally but couldn't keep up with the machine on tactics. With Rybka, he feels he is actually learning how to play better-- that its static evaluation is in a different league. So, to tie this back to the previous point, the "feel" of Rybka is so much better than other programs that coupling it with these other features would make for a totally new experience, different in kind to other "similar" features in other programs.
Thanks for writing such a great program! I guess you have enough requests to keep you busy for the next few years.
1) Play in the style of Tal, Karpov, Fischer, etc.
2) Play with various odds (pawn, pawn and move, and knight, rook, queen(!))
3) Play to win even when down material, that is, a more flexible and sophisticated contempt factor
The reason I say this is that none of these things have ever really been done effectively elsewhere. Yes, you can make programs more or less aggressive, but they aren't consistent and don't have much of a style. You can take away a knight, but they don't "realize" they are playing an odds game, and don't really try for the win or understand contempt well enough.
I understand that in each case Rybka would have to play at a little less than its best to accomodate the feature, but the pleasure it would give chess players would be immense. Imagine playing against Tal with him giving you knight odds! It would be a completely new set of features that would appeal to a wider range of chessplayers.
I have a friend who is a candidate master, and who has played other top programs over the years. He said that when he lost to them, he never felt he learned anything much about chess-- he just felt that he was outcalculated, and that he felt he played better positionally but couldn't keep up with the machine on tactics. With Rybka, he feels he is actually learning how to play better-- that its static evaluation is in a different league. So, to tie this back to the previous point, the "feel" of Rybka is so much better than other programs that coupling it with these other features would make for a totally new experience, different in kind to other "similar" features in other programs.
Thanks for writing such a great program! I guess you have enough requests to keep you busy for the next few years.
Of course you can play with odds now, just remove the piece with setup mode. We have already done a great deal to make Rybka play better when down material, with the incentive of the pawn odds matches with Ehlvest and Benjamin. The one thing that is missing is a small handicap opening book (which I've already made, perhaps it could go into Rybka 3). The more sophisticated contempt factor has already been done (though it will surely be further improved) and should be in Rybka 3. This works especially well in handicap play, as high contempt is needed when down material so Rybka won't just grovel for a draw.
I can tell you these things all make a difference. A year ago I could normally beat Rybka in fast play at knight odds. Now, with a quad computer, using our latest (non-public) version, setting contempt to a pawn, and using my knight-odds opening book, if I play straight blitz (5' no increment), I'm a clear underdog. I get plenty of draws, but winning is very difficult for me now.
I can tell you these things all make a difference. A year ago I could normally beat Rybka in fast play at knight odds. Now, with a quad computer, using our latest (non-public) version, setting contempt to a pawn, and using my knight-odds opening book, if I play straight blitz (5' no increment), I'm a clear underdog. I get plenty of draws, but winning is very difficult for me now.
Thanks for the response. All I was really saying is that it would be great if these things were included in Rybka 3-- often you and Vas say such things are "for a future version", and being good at odds, sophisticated contempt, and if possible, different playing styles, are all in my opinion huge improvements for the typical person who buys a high-end engine and should be available in the next major version if that is at all workable for the Rybka team. My feeling is that there are a whole lot of Chessbase and even many Chessmaster(!) customers who would appreciate a sophisticated approach to these three features. Many of them may not appreciate many of the subtleties that people on this forum do, but I think many club players would love to have a machine attack when playing with rook odds and play the kind of chess where you understand you have not just been beaten by a mechnical tactician but actually outclassed. You could learn from that kind of experience. Thanks again for all your work.
Hi,
Larry's answer was specifically about handicap play. As it turns out, this is very closely linked to trade-down incentives, which are important for playing strength (and which Larry couldn't really ignore as part of his work). That's why this issue managed to get to the top of the priority list already earlier this year.
Re. the other features, I completely agree that these would be extremely cool, that many people would really like them, and that the existing attempts at doing these things are not yet very convincing. We will get to all of this. At the moment, we just can't lose the one main thing we have, which is engine strength. We will definitely try to expand when the right moment comes.
Vas
Larry's answer was specifically about handicap play. As it turns out, this is very closely linked to trade-down incentives, which are important for playing strength (and which Larry couldn't really ignore as part of his work). That's why this issue managed to get to the top of the priority list already earlier this year.
Re. the other features, I completely agree that these would be extremely cool, that many people would really like them, and that the existing attempts at doing these things are not yet very convincing. We will get to all of this. At the moment, we just can't lose the one main thing we have, which is engine strength. We will definitely try to expand when the right moment comes.
Vas
Got it-- that makes a lot of sense. Thanks for elaborating.
Tony
Tony
For the record, Anson is the team captain and even though I continually pressure him to use strange openings, he seldom follows my advice. However with the very best players--you know their names--I think he is more inclined to be creative as well. Anyway we will see tomorrow, won't we?
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill