The Moron and the Genius, Side by Side

[fern]

In a reply to a post by a pal here, I stumbled with a sentence that, later considered, reflex very well the unuseful feratures of the engines that pretend to play weak, friendly levels, even offering detailed curtailings of his strenght as 1900, 2000, 2100 and so on players.
Nothing could be more false.
If you want to play a 1900 player you must play or an engine with that strenght or a dedicated unit with that force. Stronger machines only give to us something like the tittle I used here suggest: an akward mixture of very bad and very good moves, these last the moves that the engine do at his best.
Playing a personality of Chessmaster rated 2300 or so, I met an opponent capable of 2600 Elo moves and then a patzer giving me a piece for almost nothing. In that conditions you never can submerge in the ilusion you are playing such a player, any player. And then as a result you tend to play badly, never matching your force with the program, sometimes expecting another imbecility only to lose to the GM, other times calculating for the GM and getting as an answer a patzer move.
In that conditions I prefer very much to play an honest, enterily 1500 machine than those monster where you have a bycephale criature, half moron, half genius.

My best
Fern

[Uri Blass]

In a reply to a post by a pal here, I stumbled with a sentence that, later considered, reflex very well the unuseful feratures of the engines that pretend to play weak, friendly levels, even offering detailed curtailings of his strenght as 1900, 2000, 2100 and so on players.
Nothing could be more false.
If you want to play a 1900 player you must play or an engine with that strenght or a dedicated unit with that force. Stronger machines only give to us something like the tittle I used here suggest: an akward mixture of very bad and very good moves, these last the moves that the engine do at his best.
Playing a personality of Chessmaster rated 2300 or so, I met an opponent capable of 2600 Elo moves and then a patzer giving me a piece for almost nothing. In that conditions you never can submerge in the ilusion you are playing such a player, any player. And then as a result you tend to play badly, never matching your force with the program, sometimes expecting another imbecility only to lose to the GM, other times calculating for the GM and getting as an answer a patzer move.
In that conditions I prefer very much to play an honest, enterily 1500 machine than those monster where you have a bycephale criature, half moron, half genius.

My best
Fern

You do not need a special engine with strength of 1900
if you want to play against engine with strength of 1900 you can use many strong engines with fixed number of nodes under arena.

I guess that some authors of chess engines tried to make engines more human like at specifice rating level so they made them unstable based on the idea that humans are unstable and even kramnik can miss mate in 1 so they decided that playing random move from time to time and playing like a GM in other moves is more human like then playing at fixed level.

Uri

[bob]

In a reply to a post by a pal here, I stumbled with a sentence that, later considered, reflex very well the unuseful feratures of the engines that pretend to play weak, friendly levels, even offering detailed curtailings of his strenght as 1900, 2000, 2100 and so on players.
Nothing could be more false.
If you want to play a 1900 player you must play or an engine with that strenght or a dedicated unit with that force. Stronger machines only give to us something like the tittle I used here suggest: an akward mixture of very bad and very good moves, these last the moves that the engine do at his best.
Playing a personality of Chessmaster rated 2300 or so, I met an opponent capable of 2600 Elo moves and then a patzer giving me a piece for almost nothing. In that conditions you never can submerge in the ilusion you are playing such a player, any player. And then as a result you tend to play badly, never matching your force with the program, sometimes expecting another imbecility only to lose to the GM, other times calculating for the GM and getting as an answer a patzer move.
In that conditions I prefer very much to play an honest, enterily 1500 machine than those monster where you have a bycephale criature, half moron, half genius.

My best
Fern

You do not need a special engine with strength of 1900
if you want to play against engine with strength of 1900 you can use many strong engines with fixed number of nodes under arena.

I guess that some authors of chess engines tried to make engines more human like at specifice rating level so they made them unstable based on the idea that humans are unstable and even kramnik can miss mate in 1 so they decided that playing random move from time to time and playing like a GM in other moves is more human like then playing at fixed level.

Uri

It is just not the same, and his point is well-taken. But I would suggest he try Crafty with the "skill" command as it is very good in how it "dumbs things down". As you lower the skill setting, the evaluation is toned down, the search is toned down, and some randomness is added in to the evaluation. By the time you get to skill 0, the evaluation is totally random so I would probably not go that far. But it is very easy to drop the elo by 200 and 400 using skill 70 and skill 50. Lower than 50 continues to drop the rating but I don't have a good way to calibrate it to know how far it drops at (say) skill 25 or something similar.

But from playing a number of games against it personally, it doesn't "feel that bad". Just dropping the number of nodes only changes one component of the engine, yet its evaluation remains just as powerful as ever. Real 1900 players don't have sophisticated evaluation with dumb tactics, or vice-versa...

[meschle]

In a reply to a post by a pal here, I stumbled with a sentence that, later considered, reflex very well the unuseful feratures of the engines that pretend to play weak, friendly levels, even offering detailed curtailings of his strenght as 1900, 2000, 2100 and so on players.
Nothing could be more false.
If you want to play a 1900 player you must play or an engine with that strenght or a dedicated unit with that force. Stronger machines only give to us something like the tittle I used here suggest: an akward mixture of very bad and very good moves, these last the moves that the engine do at his best.
Playing a personality of Chessmaster rated 2300 or so, I met an opponent capable of 2600 Elo moves and then a patzer giving me a piece for almost nothing. In that conditions you never can submerge in the ilusion you are playing such a player, any player. And then as a result you tend to play badly, never matching your force with the program, sometimes expecting another imbecility only to lose to the GM, other times calculating for the GM and getting as an answer a patzer move.
In that conditions I prefer very much to play an honest, enterily 1500 machine than those monster where you have a bycephale criature, half moron, half genius.

My best
Fern

I totally agree fern - i have collected a series of engines (reliable) rated approx 1300 to 2000 and find them very good opponents - i have tried quite a few dumbed down engines - i am winning for sure and then the engine makes a gm style move

[geots]

In a reply to a post by a pal here, I stumbled with a sentence that, later considered, reflex very well the unuseful feratures of the engines that pretend to play weak, friendly levels, even offering detailed curtailings of his strenght as 1900, 2000, 2100 and so on players.
Nothing could be more false.
If you want to play a 1900 player you must play or an engine with that strenght or a dedicated unit with that force. Stronger machines only give to us something like the tittle I used here suggest: an akward mixture of very bad and very good moves, these last the moves that the engine do at his best.
Playing a personality of Chessmaster rated 2300 or so, I met an opponent capable of 2600 Elo moves and then a patzer giving me a piece for almost nothing. In that conditions you never can submerge in the ilusion you are playing such a player, any player. And then as a result you tend to play badly, never matching your force with the program, sometimes expecting another imbecility only to lose to the GM, other times calculating for the GM and getting as an answer a patzer move.
In that conditions I prefer very much to play an honest, enterily 1500 machine than those monster where you have a bycephale criature, half moron, half genius.

My best
Fern

You do not need a special engine with strength of 1900
if you want to play against engine with strength of 1900 you can use many strong engines with fixed number of nodes under arena.

I guess that some authors of chess engines tried to make engines more human like at specifice rating level so they made them unstable based on the idea that humans are unstable and even kramnik can miss mate in 1 so they decided that playing random move from time to time and playing like a GM in other moves is more human like then playing at fixed level.

Uri

It is just not the same, and his point is well-taken. But I would suggest he try Crafty with the "skill" command as it is very good in how it "dumbs things down". As you lower the skill setting, the evaluation is toned down, the search is toned down, and some randomness is added in to the evaluation. By the time you get to skill 0, the evaluation is totally random so I would probably not go that far. But it is very easy to drop the elo by 200 and 400 using skill 70 and skill 50. Lower than 50 continues to drop the rating but I don't have a good way to calibrate it to know how far it drops at (say) skill 25 or something similar.

But from playing a number of games against it personally, it doesn't "feel that bad". Just dropping the number of nodes only changes one component of the engine, yet its evaluation remains just as powerful as ever. Real 1900 players don't have sophisticated evaluation with dumb tactics, or vice-versa...

Bob, i wonder if you could do me a favor when you have time. I would like to try your idea with Crafty. Just tell me which version. And i would like to start off at about 1300 and work up 100 points at a time till i was say around 2300. If there is any way possible you could tell me a way that might work approximately well by using which skill level for 1300 and then 1400, etc. to be close- on up to 2300 if possible. If you have the time and it can be done, please email me at geotsp8@aol.com


Thanks,

George

[bob]

In a reply to a post by a pal here, I stumbled with a sentence that, later considered, reflex very well the unuseful feratures of the engines that pretend to play weak, friendly levels, even offering detailed curtailings of his strenght as 1900, 2000, 2100 and so on players.
Nothing could be more false.
If you want to play a 1900 player you must play or an engine with that strenght or a dedicated unit with that force. Stronger machines only give to us something like the tittle I used here suggest: an akward mixture of very bad and very good moves, these last the moves that the engine do at his best.
Playing a personality of Chessmaster rated 2300 or so, I met an opponent capable of 2600 Elo moves and then a patzer giving me a piece for almost nothing. In that conditions you never can submerge in the ilusion you are playing such a player, any player. And then as a result you tend to play badly, never matching your force with the program, sometimes expecting another imbecility only to lose to the GM, other times calculating for the GM and getting as an answer a patzer move.
In that conditions I prefer very much to play an honest, enterily 1500 machine than those monster where you have a bycephale criature, half moron, half genius.

My best
Fern

You do not need a special engine with strength of 1900
if you want to play against engine with strength of 1900 you can use many strong engines with fixed number of nodes under arena.

I guess that some authors of chess engines tried to make engines more human like at specifice rating level so they made them unstable based on the idea that humans are unstable and even kramnik can miss mate in 1 so they decided that playing random move from time to time and playing like a GM in other moves is more human like then playing at fixed level.

Uri

It is just not the same, and his point is well-taken. But I would suggest he try Crafty with the "skill" command as it is very good in how it "dumbs things down". As you lower the skill setting, the evaluation is toned down, the search is toned down, and some randomness is added in to the evaluation. By the time you get to skill 0, the evaluation is totally random so I would probably not go that far. But it is very easy to drop the elo by 200 and 400 using skill 70 and skill 50. Lower than 50 continues to drop the rating but I don't have a good way to calibrate it to know how far it drops at (say) skill 25 or something similar.

But from playing a number of games against it personally, it doesn't "feel that bad". Just dropping the number of nodes only changes one component of the engine, yet its evaluation remains just as powerful as ever. Real 1900 players don't have sophisticated evaluation with dumb tactics, or vice-versa...

Bob, i wonder if you could do me a favor when you have time. I would like to try your idea with Crafty. Just tell me which version. And i would like to start off at about 1300 and work up 100 points at a time till i was say around 2300. If there is any way possible you could tell me a way that might work approximately well by using which skill level for 1300 and then 1400, etc. to be close- on up to 2300 if possible. If you have the time and it can be done, please email me at geotsp8@aol.com


Thanks,

George

Any recent 22.x version has this, but it has to be compiled in with -DSKILL. Several compile it that way. It is fractionally slower and is generally left out for max performance.

I can't tell you how to get it to 1300. I ran a test on the cluster when I first did this, and found the following:

skill 70. Reduces elo by roughly 200.

skill 50. reduces elo by roughly 400.

I could not measure further, reliably, because the opponents I test with are too strong to compare with programs rated more than 400 Elo below them. I can't even guarantee you that skill 50 is really 400 lower, it might be worse as when you go much lower than the opponents you test against, the ratings become unreliable. By the time you get to the 5-10 level, most players should be able to beat it although it still plays what looks like real chess, just with bad moves. What it does is this:

eval = skill * real_eval / 100 + (100 - skill) * random() / 100

Which is simple-speak for use "skill%" of the real eval and use (100-skill%) of a random number... at skill=0, it is a pure random evaluation, no material or anything...

[geots]

Thanks Bob, for the reply and the help.

Best,

[Father]

Hello Friends,

........... I would like:

1- to have a chess machine, where I can play against it, in the elo that i like to play. For example. 900 elo, 100 elo, 1100 elo, 1200 elo ... 2350 elo until over 300 elo.
1.1- I have seen that these machines, have strong opening books database..... then, this is not the chess machine I like to play against( absolutly agree with Fernando sentence)

This special tools or kind of chess machine will be very usefull, becouse of, many time, our friends, when we are fighting in - holidays, or freetime, are lower or stronger that we are...... then, it is necesary for them to have the oportunity to play against that kind of chess machine, becouse, they are gpoing to be happy, and then, many other fans will be buying new softwares.

2- to have a chess machine that has in its very hight chess tools, mizelaneous puzzles, like Polgar books; Fritz 11 has this chance, and this is very great, but in my humbble opinion, the tryining puzzles, looks too fast, just 5 minutes per sesion- not enough time for a good ponderation and recognizion of a new and strong knowdlodge in our brain(in my poor amateur brain)

3- a machine where could i be talking in English lenguage, and about history of chess too, all time thinking in how rto solve a puzzle is very boring, becouse of, in a same time, we will learning how to listening, writing, speaking, and reading, well the English lenguage, and learn about chess history, chess computers history etc, and how tro play chess of course....... I would like to have conversations in English with my chess machine.....

4- a machine that play in many diferent stiles...... super agressive, solid, dinamic, defensive, antiwall techiniques (like Rybka 3), etc....

5- a machine where could I learn with explanation, talking in English, or other lenguage, about the top chess human beings games in all History..........

So,

Where could I buy this machine ? I like Rybka 3 now, becouse it is really an strong machine. bUt does it have all about I was talking ?

Thanks in advance,


Pablo

[bob]

Hello Friends,

........... I would like:

1- to have a chess machine, where I can play against it, in the elo that i like to play. For example. 900 elo, 100 elo, 1100 elo, 1200 elo ... 2350 elo until over 300 elo.
1.1- I have seen that these machines, have strong opening books database..... then, this is not the chess machine I like to play against( absolutly agree with Fernando sentence)

This special tools or kind of chess machine will be very usefull, becouse of, many time, our friends, when we are fighting in - holidays, or freetime, are lower or stronger that we are...... then, it is necesary for them to have the oportunity to play against that kind of chess machine, becouse, they are gpoing to be happy, and then, many other fans will be buying new softwares.

2- to have a chess machine that has in its very hight chess tools, mizelaneous puzzles, like Polgar books; Fritz 11 has this chance, and this is very great, but in my humbble opinion, the tryining puzzles, looks too fast, just 5 minutes per sesion- not enough time for a good ponderation and recognizion of a new and strong knowdlodge in our brain(in my poor amateur brain)

3- a machine where could i be talking in English lenguage, and about history of chess too, all time thinking in how rto solve a puzzle is very boring, becouse of, in a same time, we will learning how to listening, writing, speaking, and reading, well the English lenguage, and learn about chess history, chess computers history etc, and how tro play chess of course....... I would like to have conversations in English with my chess machine.....

4- a machine that play in many diferent stiles...... super agressive, solid, dinamic, defensive, antiwall techiniques (like Rybka 3), etc....

5- a machine where could I learn with explanation, talking in English, or other lenguage, about the top chess human beings games in all History..........

So,

Where could I buy this machine ? I like Rybka 3 now, becouse it is really an strong machine. bUt does it have all about I was talking ?

Thanks in advance,


Pablo

A couple of points.

1. I don't modify book behavior based on the skill setting, but your point is well-taken. I could either (a) randomize selection more so that it will play even ugly openings (1. f3 and 2. Kf2 and such nonsense) or I could trim the depth down as skill goes down to simulate a "weaker book memory". That's an interesting topic I have not considered, but will.

2. Specific elos are _very_ difficult to emulate. First, different hardware speeds would require different "skill level settings" inside Crafty to produce the same Elo, which is a problem. I know that skill 70 drops the Elo by about 200 rating points, but that is a drop from the "optimal setting" given your hardware. Slow hardware starts off with a lower Elo automatically, which makes the skill 70 200 points weaker than the already weaker playing level produced by that slow hardware.

This is not an easy things. The main problem is, who can explain exactly what a 1500 Elo player is capable of, both positional judgement-wise, as well as tactical-wise? I don't believe we know. Given enough humans with stable ratings, I could certainly work out "skill" settings to hit every 200-Elo category, but it would take some time. And changing the hardware would raise _all_ the Elos up by some unknown amount...

What you probably want is to be able to say "elo 1950" and have the program suddenly play like the average 1950 Elo player plays. It is extremely unlikely that will happen...

[Father]

Doctor Robert.
Thank you very much for your explanation. It is very interesting.