For Matthias: a question about Big Lion/Taktix

Archive of the old Parsimony forum. Some messages couldn't be restored. Limitations: Search for authors does not work, Parsimony specific formats do not work, threaded view does not work properly. Posting is disabled.

For Matthias: a question about Big Lion/Taktix

Postby Telmo Escobar » 21 Jun 2003, 21:16

Geschrieben von: / Posted by: Telmo Escobar at 21 June 2003 22:16:34:

Among the files of Big Lion 1.80h (same for Taktix) are one named eval.prm, obviously for customizing the engine (and, I think, to create personalities). Some advice for using this feature?
In particular several lines of the file are for a personality. What's this?

Last question: there are three text files apparently related to the abovementioned. They are evalbnds.txt, evalbound.txt (this with 0 bytes) and evalprm.txt. What are these files for?
Thanks in advance!
Telmo
PS. the anthem (Cameroon.mid) sounds great!
Telmo Escobar
 

BigLion/Taktix personalities

Postby Matthias Gemuh » 22 Jun 2003, 08:46

Geschrieben von: / Posted by: Matthias Gemuh at 22 June 2003 09:46:06:
Als Antwort auf: / In reply to: For Matthias: a question about Big Lion/Taktix geschrieben von: / posted by: Telmo Escobar at 21 June 2003 22:16:34:
Among the files of Big Lion 1.80h (same for Taktix) are one named eval.prm, obviously for customizing the engine (and, I think, to create personalities). Some advice for using this feature?
In particular several lines of the file are for a personality. What's this?

Last question: there are three text files apparently related to the abovementioned. They are evalbnds.txt, evalbound.txt (this with 0 bytes) and evalprm.txt. What are these files for?
Thanks in advance!
Telmo
PS. the anthem (Cameroon.mid) sounds great!

Hi Telmo,
BigLion Personalities
===========================
How to create a BigLion/Taktix personality
You will have to edit Eval.prm and register the personality in ini file.
A file Evalprm.txt will be created by the engine, to indicate the
personality values it uses. Check for yourself, whether these are the
values you defined. EvalBnds.txt is for author (me) only.
If you create a personality that kills, please inform me.
It is clear that you can only fully judge the influence of personality values
if you have the source code of the evaluation function.
If you need the source code, contact me by e-mail.
Concerning Eval.prm,
;
; Don't change the numbers in first (left) column !! Each is an array index
; Don't change the order of the parameters !!
;
; GAMESTAGE => parameter depends also on stage of game (stage=0 at beginning, stage=80 when board empty)
; influence of game stage too complicated to be adjustable by user
; MATHS => parameter modified by (+*/-) and some unrevealable constant/variable apart from game stage
; BONUS => (modified) value added to score
; PENALTY => (modified) value subtracted from score
; ATB => valid only when using (slow) attack boards (Time per move in secs more than LongGameMoveTime)
; NOATB is always valid, even when ATB is valid.
; NOATB => valid when not using attack boards (Time per move in secs less than LongGameMoveTime)
; SYMM => symmetrical evaluation, both sides treated equally
; ASYMM => BigLion+Taktix side gets different treatment
; ROW => depends on row/rank on board
; COL => depends on column/file on board
; DENOM => a denominator (German: Nenner), should never be zero !!
;
; PERSONALITIES: NORMAL, AGGRESSIVE, SOLID
; You can define as many as 30 personalities, using [[]], capital letters, no spaces,
; and register one of them in INI file.
; Please do not edit the personality "NORMAL". Create your own personalities or edit "AGGRESSIVE", "SOLID".
;
; ******************************************************************************************

[BookLearning]
1 TimePerMove 6 ; Least time per move (in secs) for book learning to kick in (i.e. be triggered)
2 CheckPointMove1 30 ; At this number of game moves, learn first BookMovesPlys moves, if score better than CheckPointScore1
3 CheckPointScore1 500 ; Book learning threshold in centipawns
4 CheckPointMove2 50 ; At this number of game moves, learn first BookMovesPlys moves, if score better than CheckPointScore2
5 CheckPointScore2 600 ; Book learning threshold in centipawns
6 BookMovesPlys 28 ; Learn this number of moves of winning (own) side
;
[Pieces]
7 PawnValue 100 => BONUS ATB NOATB SYMM ;
8 KnightValue 331 => BONUS ATB NOATB SYMM ;
10 BishopValue 353 => BONUS ATB NOATB SYMM ;
11 RookValue 500 => BONUS ATB NOATB SYMM ;
12 QueenValue 909 => BONUS ATB NOATB SYMM GAMESTAGE ;
;
... etc
As you can see, the Queen is treated symmetrically and her value also depends
on the stage of the game and valid both when using or not using attack boards
and it is nice to have a Queen (BONUS, not PENALTY).

I thank all the people who are interested in such weak engines as
BigLion+Taktix. I am rewriting them now. Maybe that will increase
their strength by 500 Elo .
Regards,
Matthias.
Matthias Gemuh
 

Re: BigLion/Taktix personalities

Postby Uri Blass » 22 Jun 2003, 09:11

Geschrieben von: / Posted by: Uri Blass at 22 June 2003 10:11:59:
Als Antwort auf: / In reply to: BigLion/Taktix personalities geschrieben von: / posted by: Matthias Gemuh at 22 June 2003 09:46:06:
Among the files of Big Lion 1.80h (same for Taktix) are one named eval.prm, obviously for customizing the engine (and, I think, to create personalities). Some advice for using this feature?
In particular several lines of the file are for a personality. What's this?

Last question: there are three text files apparently related to the abovementioned. They are evalbnds.txt, evalbound.txt (this with 0 bytes) and evalprm.txt. What are these files for?
Thanks in advance!
Telmo
PS. the anthem (Cameroon.mid) sounds great!

Hi Telmo,
BigLion Personalities
===========================
How to create a BigLion/Taktix personality
You will have to edit Eval.prm and register the personality in ini file.
A file Evalprm.txt will be created by the engine, to indicate the
personality values it uses. Check for yourself, whether these are the
values you defined. EvalBnds.txt is for author (me) only.
If you create a personality that kills, please inform me.
It is clear that you can only fully judge the influence of personality values
if you have the source code of the evaluation function.
If you need the source code, contact me by e-mail.
Concerning Eval.prm,
;
; Don't change the numbers in first (left) column !! Each is an array index
; Don't change the order of the parameters !!
;
; GAMESTAGE => parameter depends also on stage of game (stage=0 at beginning, stage=80 when board empty)
; influence of game stage too complicated to be adjustable by user
; MATHS => parameter modified by (+*/-) and some unrevealable constant/variable apart from game stage
; BONUS => (modified) value added to score
; PENALTY => (modified) value subtracted from score
; ATB => valid only when using (slow) attack boards (Time per move in secs more than LongGameMoveTime)
; NOATB is always valid, even when ATB is valid.
; NOATB => valid when not using attack boards (Time per move in secs less than LongGameMoveTime)
; SYMM => symmetrical evaluation, both sides treated equally
; ASYMM => BigLion+Taktix side gets different treatment
; ROW => depends on row/rank on board
; COL => depends on column/file on board
; DENOM => a denominator (German: Nenner), should never be zero !!
;
; PERSONALITIES: NORMAL, AGGRESSIVE, SOLID
; You can define as many as 30 personalities, using [[]], capital letters, no spaces,
; and register one of them in INI file.
; Please do not edit the personality "NORMAL". Create your own personalities or edit "AGGRESSIVE", "SOLID".
;
; ******************************************************************************************

[BookLearning]
1 TimePerMove 6 ; Least time per move (in secs) for book learning to kick in (i.e. be triggered)
2 CheckPointMove1 30 ; At this number of game moves, learn first BookMovesPlys moves, if score better than CheckPointScore1
3 CheckPointScore1 500 ; Book learning threshold in centipawns
4 CheckPointMove2 50 ; At this number of game moves, learn first BookMovesPlys moves, if score better than CheckPointScore2
5 CheckPointScore2 600 ; Book learning threshold in centipawns
6 BookMovesPlys 28 ; Learn this number of moves of winning (own) side
;
[Pieces]
7 PawnValue 100 => BONUS ATB NOATB SYMM ;
8 KnightValue 331 => BONUS ATB NOATB SYMM ;
10 BishopValue 353 => BONUS ATB NOATB SYMM ;
11 RookValue 500 => BONUS ATB NOATB SYMM ;
12 QueenValue 909 => BONUS ATB NOATB SYMM GAMESTAGE ;
;
... etc
As you can see, the Queen is treated symmetrically and her value also depends
on the stage of the game and valid both when using or not using attack boards
and it is nice to have a Queen (BONUS, not PENALTY).

I thank all the people who are interested in such weak engines as
BigLion+Taktix. I am rewriting them now. Maybe that will increase
their strength by 500 Elo .
Regards,
Matthias.
I also decided to modify things in movei
and I added new arrays to the last tested non public version.
First tests suggest that adding the arrays and calculate them after every move made movei 20% slower but I can use the arrays to do significant part of my program faster and smaller so even only from speed point of view it is not clear if the arrays are counterproductive.
I plan first to see how much speed I can gain by the new arrays and only later to use the arrays for better evaluation and better search decisions.
Uri
Uri Blass
 

Re: BigLion/Taktix personalities

Postby Matthias Gemuh » 22 Jun 2003, 09:29

Geschrieben von: / Posted by: Matthias Gemuh at 22 June 2003 10:29:11:
Als Antwort auf: / In reply to: Re: BigLion/Taktix personalities geschrieben von: / posted by: Uri Blass at 22 June 2003 10:11:59:

I also decided to modify things in movei
and I added new arrays to the last tested non public version.
First tests suggest that adding the arrays and calculate them after every move made movei 20% slower but I can use the arrays to do significant part of my program faster and smaller so even only from speed point of view it is not clear if the arrays are counterproductive.
I plan first to see how much speed I can gain by the new arrays and only later to use the arrays for better evaluation and better search decisions.
Uri


Hi Uri,
while rewritng, I will proceed as you recommend, that is
1) test everything well before continuing
2) if anything seems counterproductive, drop it
3) strength depends mostly on how bug-free the engine is.
4) nps speed is not very important.
My personal aim will be to score 10% or more against mighty Movei.
/Matthias.
Matthias Gemuh
 

Re: BigLion/Taktix personalities

Postby Uri Blass » 22 Jun 2003, 09:52

Geschrieben von: / Posted by: Uri Blass at 22 June 2003 10:52:59:
Als Antwort auf: / In reply to: Re: BigLion/Taktix personalities geschrieben von: / posted by: Matthias Gemuh at 22 June 2003 10:29:11:
I also decided to modify things in movei
and I added new arrays to the last tested non public version.
First tests suggest that adding the arrays and calculate them after every move made movei 20% slower but I can use the arrays to do significant part of my program faster and smaller so even only from speed point of view it is not clear if the arrays are counterproductive.
I plan first to see how much speed I can gain by the new arrays and only later to use the arrays for better evaluation and better search decisions.
Uri

Hi Uri,
while rewritng, I will proceed as you recommend, that is
1) test everything well before continuing
2) if anything seems counterproductive, drop it
3) strength depends mostly on how bug-free the engine is.
4) nps speed is not very important.
My personal aim will be to score 10% or more against mighty Movei.
/Matthias.
After almost every small change that should influence only speed I tell movei to analyze the opening position for a minute to see if I have not a new bug.
It helps to detect bugs(if the number of nodes is not the same)
I will do more tests after I complete the changes that are supposed to make the program faster.
I also believe that knowledge is more important but there are cases when there is no contradiction and knowledge helps to do the program faster because some expensive things that you do may be done faster thanks to arrays.
If you do it 500 elo better than you are going to be near crafty level and you are going to score clearly more than 10% even in the optimistic assumption that movei will get Shredder7.04's level(biglion already beated shredder7.04)
Uri
Uri Blass
 

Re: BigLion/Taktix personalities

Postby Matthias Gemuh » 22 Jun 2003, 11:14

Geschrieben von: / Posted by: Matthias Gemuh at 22 June 2003 12:14:46:
Als Antwort auf: / In reply to: BigLion/Taktix personalities geschrieben von: / posted by: Matthias Gemuh at 22 June 2003 09:46:06:


BTW, the NIMZO personality has been created by Eike Hinner.
Looks like he is still tuning it.
Everybody should feel free to create a personality that I can include in Eval.prm
or even replace the default INTERNAL personality with.
Just to clearify again, a personality is a set of more than 100 parameters,
which the user can modify to influence evaluation and style of play.
/Matthias.
Matthias Gemuh
 


Return to Archive (Old Parsimony Forum)

Who is online

Users browsing this forum: No registered users and 16 guests