Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Pro Deo/Rebel parameter

Author: Eelco de Groot

Date: 04:30:16 10/21/05

Go up one level in this thread


On October 21, 2005 at 06:48:30, Kaj Soderberg wrote:

>Trying to tune Pro Deo 1.1, i ran into the possible use of [Pruning = PRUNE BAD
>MOVES_(0-8) where 8 would be default)] as a method to reduce the programs
>pruning, when studying Rebel 12 beta parameters. Could this still used, or is it
>maybe replaced by new parameters in Pro Deo? I remember Ed asking to be careful
>about using older parameters.
>
>Thanks in advance for answers.
>
>Greetings,
>
>Kaj

Helo Kaj,

Here is the complete list of parameters from Ed. But I'm not sure if this is
very readable because of the layout that I changed a bit for saving in Wordpad,
but there are still end of lines that throw things in disarray. I just try it

[Pruning = Prune Bad Moves_0]	     No reductions allowed		slows down search with
a factor of 2.
[Pruning = Prune Bad Moves_1]	    Allow only 1 reduction			slows down search
considerably (30-40%)
[Pruning = Prune Bad Moves_2]	    Allow only 2 reductions		slows down search
(10-15%), effects are little.
[Pruning = Prune Bad Moves_3]	    Allow only 3 reductions		slows down search a
bit, effects are negligible.

are still usable

 ===============================				===========
 REBEL [MISC] options    (01-80)				  REMARKS
 ===============================				===========

[MISC-01]       Don't touch	Is in use by [Eternal Checks]
[MISC-02]       Speedup		Small move ordering speedup (1.5%)       DEFAULT
SETTING.
[MISC-03]       Speedup		Small move ordering speedup (4-5%)	DEFAULT SETTING.
[MISC-04]       ALPHA/BETA	Set Window to 0.25 (instead of 0.50)	doesn't matter
much.
[MISC-05]       ALPHA/BETA	Fail-High or Fail-Low in 5 phases	            doen't
work currently.
[MISC-06]       ALPHA/BETA	Zero Width (speedup is 3%)		looks safe.
[MISC-07]       NULLMOVE	Detect Zugzwang (new style) 		untested and doubtful.
[MISC-08]       Reduction	History Reductions			DEFAULT SETTING.

[MISC-09]       Reduction	Reduce if bad HT score (max=1) (+12%)	unclear.
probably no improvement.
[MISC-10]       NULLMOVE	Block first extension (+10%)		untested. no high
expectations.
[MISC-11]       Reduction	TR70 only on RD-3			probably small improvement.
untested.
[MISC-12]       Lazy Eval	Double  Lazy Eval (safety)		            DEFAULT
SETTING.
[MISC-13]       Bishop Pair	Extra bonus (0.25) if opponent no N/B	untested.
[MISC-14]       Queen Ending	New passed pawn handling		untested. contains a bug.
[MISC-15]       Time Control	TR30 (quickly check next iteration)	unclear.
[MISC-16]       Sel. Search	Increase Brute Force if extension (1)	(2,0,0,0) see
also MISC-28

[MISC-17]       Search		Skip first extension (+30-35%)		unclear.
[MISC-18]       Bishop Pair	via no. of pawns instead of material	unclear.
[MISC-19]       Time Control	Play Fast against humans		mandatory against GM's.
Do NOT mix with MISC_68 (!)
[MISC-20]       Don't touch	In use for compatibility reasons with older
versions.
[MISC-21]       Lazy Eval	Tuning attempt (-4%)			            about equal.
[MISC-22]       Move Ordering	Sort root moves on nodes		no improvement.
[MISC-23]       Reduction	PLMAX-b reduction only once		could make a
difference.untested.
[MISC-24]       Search		Skip first recapture extension		unclear.

[MISC-25]       Reduction	PLW3 reduction (new style)		DEFAULT SETTING.
[MISC-26]       Extension	Skip endgame extension			untested. no expectations.
[MISC-27]       Eval		Eval double pawns (new style)		untested. interesting.
[MISC-28]       Sel. Search	Increase Brute Force if extension (2)	(2,2,1,1) see
also MISC-16
[MISC-29]       Search		Limit check extensions (2x SVDSG)	looks ok (+4%)
[MISC-30]       Extension	Allow singular check extensions (max=3)	great for
tactics, unclear in practical games
[MISC-31]       ALPHA/BETA	Fail_high_error = best move		unclear and risky.
[MISC-32]       Sel. Search	Increase Brute Force if extension (3)	untested.
Speed loss of 12% in endgame.

[MISC-33]       Eval		Cancel exchange when up/down algorithm. Unclear.
[MISC-34]       ............	...................................
[MISC-35]       ............	...................................
[MISC-36]       Extension	Extend double attacks (W6) at the leafs	looks
promising (-2%)
[MISC-37]       Extension	Hash Table extension (-18%)		DEFAULT SETTING.
[MISC-38]       .............	....................................
[MISC-39]       NULLMOVE	Move ordering (+4%)			DEFAULT SETTING.
[MISC-40]       Sel. Search	PLMAX-Z3 reduction (+15%)		DEFAULT SETTING.

[MISC-41]       .............	....................................
[MISC-42]       REDUCE		no reduction if prev plies extended	interesting. good
first results. needs more testing.
[MISC-43]       Search		No TWS extensions in the last x plies	untested. better
don't touch.
[MISC-44]       .............	....................................
[MISC-45]       .............	....................................
[MISC-46]       .............	....................................
[MISC-47]       Extension	King Safety extension (-5%)		equal. matter of taste.
[MISC-48]       Q-Search	No FHR pruning in QS (-6%)			unclear.

[MISC-49]       .............	....................................
[MISC-50]       Search		Maximize reductions iteration driven	unclear.
[MISC-51]       Time Control	Minimum search depth easy-moves	untested. could
matter a lot in blitz games.
[MISC-52]       Reduction	FH reduction (new style) (max=1)	DEFAULT SETTING.
[MISC-53]       Reduction	BBB reduction (new style) (max=1)	untested. could
matter.
[MISC-54]       .............	....................................
[MISC-55]       .............	....................................
[MISC-56]       NULLMOVE	Limit Nulmove Mate Extension (max=1)	untested. probably
no big deal.

[MISC-57]       Extension	Extend withdrawing moves at the leafs	looks promising
(-2%)
[MISC-58]       Eval		Recalculate passed pawns		the [M58_Table = midg]
sub-setting looks promising.
[MISC-59]       Eval		Recalculate Bad Bishop (endgame)	untested.
[MISC-60]       Eval		Recalculate Bishop financhetto		untested.
[MISC-61]       Eval		Rook behind passed pawn tuning		Sub-setting [M61_Perc =
100]
[MISC-62]       .............	....................................
[MISC-63]       Lazy Eval	Include pawns on 7th (2th) rank		            unclear.
[MISC-64]       Search		No 7th (2th) rank pawn exte in MIDG	untested.

[MISC-65]       .............	....................................
[MISC-66]       Hash Table	Bugfix					DEFAULT SETTING.
[MISC-67]       .............	....................................
[MISC-68]       Time Control	Play fast against computers		promising. Do NOT mix
with MISC_19 (!)
[MISC-69]       Eval		King tropism opponent passed pawn	DEFAULT SETTING.
[MISC-70]       .............	....................................
[MISC-71]       .............	....................................
[MISC-72]       Search		  Research FH error directly		unclear.

[MISC-73]       .............	....................................
[MISC-74]       Reduction	  HT singular reduction (2.00) (+11%)	unclear.
[MISC-75]       Eval		  King Safety (no pawnram queen endings)untested.
[MISC-76]       Endgame		  Rook 7th rank (new style)		untested. note the 2
sub-settings.
[MISC-77]       Search		  Avoid double reductions		untested. probably no big
deal.
[MISC-78]       NULLMOVE	  Adaptive nullmove.			DEFAULT SETTING.
[MISC-79]       EVAL		  Adjust King Safety (insuf. material)	attempt to fix the
KBPK bad bishop bug.
[MISC-80]       Draw		  Bugfix 50-move rule			don't touch. probably even worse.


        ==============================
	MISC sub-settings (parameters)
	==============================

[MISC-09]       Reduction	  Reduce if bad HT score (max=1) (+12%) (default is
2.00)
[M09_val = 2.00]		  Margin is 2.00

[MISC-27]       Eval		  Eval double pawns (new style)		untested. interesting.
[M27_midg = 100]		  Double pawn factor middlegame
[M27_end1 = 100]		  Double pawn factor endgame (normal)
[M27_end2 = 100]		  Double pawn factor endgame (simple)

[MISC-43]       Search		   No TWS extensions in the last x plies	untested.
better don't touch.
[M43_TWS = 3]			   RD=3

[MISC-52]       Reduction	   FH reduction (new style) (max=1)	DEFAULT SETTING.
[Fail High Margin = 9.00] 	   Default is 9.00

[MISC-58]       Eval		   Recalculate passed pawns		 the [M58_Table = midg]
sub-setting looks promising.
[M58_Table = midg]		   normal|strong|midg|endg                       (default is
normal)

[MISC-61]       Eval		   Rook behind passed pawn tuning	 untested.
[M61_Perc = 100]		   Default is 100.

[MISC-69]       Eval		    King tropism opponent passed pawn	 DEFAULT SETTING.
[M69_Perc = 100]		    Default is 100. Tuning is needed

[MISC-74]       Reduction	    HT singular reduction
(+11%) (default is 2.00)
[M09_val = 2.00]		    Margin is 2.00

[MISC-76]         			     Endgame		                          Rook 7th rank (new
style)
[Rook 7th Rank (low) = 50]              50   (default)
            (when opponent king is on the 8th rank)
[Rook 7th Rank (high) = 50]             50   (default)
            (when opponent king can't escape from 8th rank)

        ================
	Other parameters
	================

[Your Queen = 100]		    Increase (decrease) value opponent queen
[My Queen = 101]		    Own Queen +1% (0.09)		DEFAULT SETTING.
[Own Queen = 1.00]                       Increase own queen when down	unclear.
[Queen Endgame Bonus = 0.50]     Increase value of Queen in endgame	unclear.

[Endgame Mobility = 100]	    Self explanatory			untested.
[Knight Mobility = 100]		    Self explanatory			A lower value is probably
better.

[Minimum Knight Mobility = 100]     Self explanatory			DEFAULT SETTING. Tuning
is needed nevertheless.
[Minimum Bishop Mobility = 100]     Self explanatory			DEFAULT SETTING. Tuning
is needed nevertheless.

[Keep Main Line = 10]                   Minimum margin for each new
mainline	Untested.Risky! (speedup 5%)

[Time Control = 125]                      125
               clear improvement using blitz levels, unclear on all other
levels.

[Time Control Status = ALWAYS]   ALWAYS|PB ONLY

[Eternal Checks = Normal]             off|normal|risky|safe			default=off.
Algorithm to detect eternal checks, its quality is doubtful.

[Recaptures = Normal]
1.00|1.50|2.00|2.50|3.00|4.00|5.00|normal|aggressive
				    I am afraid none of these will give a clear improvement, but I only
tried [2.00]

[Chaos = 100]			    Substitute for [Attacking = 100]	[Attacking] [Chaos] and
[Confusion] are algorithms to use against humans. It's practical use against
computers is doubtful, their status is untested.



[Confusion = 100]		    Substitute for [Attacking = 100]
[Lazyeval MIDG = moderate]          * normal|moderate|strong		probably a small
improvement.
[Lazyeval ENDG = moderate]         * normal|moderate|strong		probably a small
improvement.

[Checks Depth = 1]                        limit QS checks to 1 ply
(+9%)	probably an improvement.
[Clear Hash Table = Always]           Always|Never|PB only			probably a small
improvement.

[Pruning = Prune Bad Moves_0]	     No reductions allowed		slows down search with
a factor of 2.
[Pruning = Prune Bad Moves_1]	    Allow only 1 reduction			slows down search
considerably (30-40%)
[Pruning = Prune Bad Moves_2]	    Allow only 2 reductions		slows down search
(10-15%), effects are little.
[Pruning = Prune Bad Moves_3]	    Allow only 3 reductions		slows down search a
bit, effects are negligible.

[Nullmove Margin = 0]		    Search speedup depending on the height of the
parameter.
                                                    Reasonable values in the
range of 0-16. Above that chances are high for big search gaps.
Default setting is 0 (off)

[Transpositions = ON]		    on|off (for testing purposes only)

[Extensions (remaining)= 3]       Maximum recapture extensions          (default
is 3)  (maximum is 15)

[Extensions (checks)= NORMAL]   low|moderate|normal|strong|high|all

        Low         (+25%)
									Moderate (  +8%)
									Normal     Default setting
									Strong      (-16%)

        High	    (-21%)

        All	    (-26%)





This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.