one question about knowledge chess engine match one ply ?

[JVMerlino]

There is no "bug" in Chessmaster regarding search depth. You can create a personality identical to the default, with the only difference being a max search depth of 1. The King engine does handle this parameter correctly. I know this because several of the lower-rated personalities use limited search depth (typically anywhere from 1-5 plies).

But, Bob is also correct in that you can't compare the REPORTED depth from engines. The King, for example, uses a relatively odd system of X/Y for reporting its depth. The first number is the "brute force" search depth -- the number of plies that have been exhaustively searched. The second number is the MINIMUM extension depth for all reasonable candidate moves. It is impossible to compare The King's system against any other engine, except POSSIBLY in the case of "depth = 1".

And, for those who are not aware, it is the SECOND of these numbers that is affected by the "Max Search Depth" parameter.

jm

[bob]

ChessMaster won't be "instantaneous" on one-ply searches. That's the problem. A program that does checks in the q-search will see more with a 1 ply search than one that does not. A program that has a qsearch will see more with a 1 ply search than one that will not.

It's been known for years that you can't compare programs using "equal plies", because plies are not the same from one program to another.

ok. yes. chesmaster is bad example. this is a bug error program.

but.

rybka human, and rybka cpu, and rybka dinamic, have the same ply searcj.

and fritz 6,7,8,9 deep =

and chesstiger 14 and chess tiger 15 have indentical reference plys.

arasan OK

hiarcs OK

etc..

I see during the game, and a lot of engines are correct. chesmaster no. ok. don´t problem. I don´t test with chessmaster

but the conclusions evaluation are interesting.

I'm not going to argue further. I have told you "plies" != "plies" for different programs. If you don't want to find out why, then run whatever test you want. It will be flawed, but it won't be the first nor the last flawed test that produces flawed conclusions.

[bob]

There is no "bug" in Chessmaster regarding search depth. You can create a personality identical to the default, with the only difference being a max search depth of 1. The King engine does handle this parameter correctly. I know this because several of the lower-rated personalities use limited search depth (typically anywhere from 1-5 plies).

But, Bob is also correct in that you can't compare the REPORTED depth from engines. The King, for example, uses a relatively odd system of X/Y for reporting its depth. The first number is the "brute force" search depth -- the number of plies that have been exhaustively searched. The second number is the MINIMUM extension depth for all reasonable candidate moves. It is impossible to compare The King's system against any other engine, except POSSIBLY in the case of "depth = 1".

And, for those who are not aware, it is the SECOND of these numbers that is affected by the "Max Search Depth" parameter.

jm

It is particularly difficult in the case of Rybka, which reports false depths...

[JVMerlino]

It is particularly difficult in the case of Rybka, which reports false depths...

Surprising that none of the amazingly smart and severely dedicated people here (which wouldn't be me) haven't been able to determine how to glean the real search depths from the reported ones.

jm

[bob]

It is particularly difficult in the case of Rybka, which reports false depths...

Surprising that none of the amazingly smart and severely dedicated people here (which wouldn't be me) haven't been able to determine how to glean the real search depths from the reported ones.

jm

I think someone said depth+3 is the real depth, but I have not verified that. There is also some sort of calculation to correct nodes and NPS I think. Or it might be that someone "broke the code" and patched the binary to remove the obfuscation, that actually sounds more like what I remember.

[Karmazen & Oliver]

... It is impossible to compare The King's system against any other engine,... except POSSIBLY in the case of "depth = 1".
, except POSSIBLY in the case of "depth = 1".

jm

yes. I try it. if we use 1 ply depth. the diferents concept of "depth" for the engine alterate de experiment.

on depth 1 ... mabye be more correct.

and if we test the same engine... r human, r dinamic, fritz 6~11, or ct14~15~ct2007

in this case... is interesting see...

anyway it is important AS it selects the first plays the engine, because it will use that same function an and another time in the following search

for that reason it used single depth = 1, in that way the mistrust are avoided. and it is also certain that in some moment of each branch the evaluation system is based on depth 1.

it is the but important. then the speed and the optimization of the codelevels.

[Karmazen & Oliver]

It is particularly difficult in the case of Rybka, which reports false depths...

Surprising that none of the amazingly smart and severely dedicated people here (which wouldn't be me) haven't been able to determine how to glean the real search depths from the reported ones.

jm

it should not be very difficult, only putting to evaluate chess problems that put an end in victory to checkmate... in depth 1, 2 and 3... this way we can put to rybka to think of positions with traps of material gain, where the engine can win a lady or a tower and with another play to give checkmate...

if the engine doesn't find a checkmate in 2, it is for that this calculating in 1 ply, of depth, and for her him but interesting serious to eat up the lady, because you don't see the exact victory.

let us say that similar as "advancing" the horizon of final to 1 ply, test 2 plys... 3... 4.

in that way tests the engines is been able to, to see if they say the truth, although I believe that in sand chess gui, the one calculates of 1 ply, it is correct, no engine thinks of 3 plays, and some of them are let pieces hung in the board besides that you/they repeat the exact departures, a clear symptom that they are not deepening in any branch that implies a certain random election.

all the programs have certain dose of calculate random

I believe that tests the engines with examples of problems with checkmate you could conclude if they calculate to that depth.

for example, no engine should be able to see a victory forced in 3 plays if this limited one to 1 mov of depth.

[tano-urayoan]

Without permission of the forum I will answer in spanish as there seems to be communication problems.

Sr. Oliver, lo que se le esta tratando de explicar es que cada programa reporta 'plys' de manera diferente y que no se pueden comparar. Un ply en Fritz no es lo mismo que un ply en Rybka. Por ejemplo se ha reportado que cuando el programa Rybka reporta un 1 realmente esta haciendo una busqueda de 4 plies. Por lo tanto si tomamos que 1 ply de Fritz es cierto y jugamos contra Rybka esta ultima le lleva ventaja en la busqueda de 3 plys por lo tanto la ventaja sera abrumadora. De hecho se ha mencionado que en su version mas reciente se le puede ajustar a Rybka para jugar hasta con - 3 plies. Como entonces ud, explica que 1 es igual para todos los programas, como defines un ply negativo. Por lo tanto para que tal vez Rybka juegue a 1 ply tienes que ajustar los parametros hasta -3. Realmente su prueba es totalmente invalida cientificamente y no tiene mas que un caracter de entrenimiento propio.
Saludos Tano-Urayoan

[JohnS]

Bob wrote:

"I'm not going to argue further. I have told you "plies" != "plies" for different programs."

This reminds me of the "oils aint oils" ad from one of the oil companies many years ago.

[Karmazen & Oliver]

Without permission of the forum I will answer in spanish as there seems to be communication problems.

Sr. Oliver, lo que se le esta tratando de explicar es que cada programa reporta 'plys' de manera diferente y que no se pueden comparar. Un ply en Fritz no es lo mismo que un ply en Rybka. Por ejemplo se ha reportado que cuando el programa Rybka reporta un 1 realmente esta haciendo una busqueda de 4 plies. Por lo tanto si tomamos que 1 ply de Fritz es cierto y jugamos contra Rybka esta ultima le lleva ventaja en la busqueda de 3 plys por lo tanto la ventaja sera abrumadora. De hecho se ha mencionado que en su version mas reciente se le puede ajustar a Rybka para jugar hasta con - 3 plies. Como entonces ud, explica que 1 es igual para todos los programas, como defines un ply negativo. Por lo tanto para que tal vez Rybka juegue a 1 ply tienes que ajustar los parametros hasta -3. Realmente su prueba es totalmente invalida cientificamente y no tiene mas que un caracter de entrenimiento propio.
Saludos Tano-Urayoan

1 - al bloquear analisis a 1 ply, nos aseguramos de entender las seleciones del programa aunque piense 3 plys, nunca pensara mas que eso

2 - esto sigue siendo util para evaluar distintas versiones de la misma engine

3 - otras engines tambien hacen trampas, luego estamos en igualdad de condiciones.

The experiment has been useful and amusing.

1 - this evaluation is very useful to compare the evolution of different versions of the same engine, to compare among ryka 1, rybka 2, rybka 3 stiles, rybka mp. and to compare among fritz 6 to fritz 11, and versions deep... and to compare chesstiger versions... and to compare hiarsc 9 to 12...

2 - when we block analysis to 1 ply, at least we make sure of understanding the first movements candidates of the program, without influencing the quick cpus and code optimizations, perhaps be not exactly 1 ply, but it will always be the same analysis depth, and that is the important thing..

3 - this will reveal concept shortcomings in some positions...

4 - it can improve the level of an engine improving the selections that he/she makes in the first play... (first movements candidates) because starting from here the whole tree of variants is born.

5 - other engines calculate more +1 ply too... = conditions...

a greeting.

bye. from spain. oliver.

postcript:

http://www.talkchess.com/forum/viewtopic.php?t=23549