Author: Eelco de Groot
Date: 08:41:26 09/28/05
Go up one level in this thread
On September 20, 2005 at 13:39:25, Ed Schröder wrote:
>On September 20, 2005 at 11:49:03, Eelco de Groot wrote:
>
>>Hello all,
>>
>>It is good to see Bruce Moreland back again! He posted something
>>computerchess-related in CTF once already before but it is good to see him back
>>here with Ferret!
>>
>>A few Pro Deo notes on the discussed position; Pro Deo has a bug somewhere with
>>KBPK handling so that must be a double disadvantage here for Pro Deo. I tried to
>>find relevant material about the Rebel KBPK bug with the CCC search engine but
>>could not find any.
>
>Something funny about the bug, try this position or other KBPK positions:
>
>[d]6k1/8/7P/8/8/8/8/4KB2 w - -
>
>And things go wrong. Now set [King Safety = 0] and everything is fine. Imagine
>my surprise.
>
>Ed
Hallo Ed,
This must be a really mysterious bug I think!
I know just changing just the settings for pro Deo is really never going to be
sufficient in tracking down the Pro Deo's KBPK bug. But sofar in a few tests, I
only made some more time for it yesterday, there is not even a discernible
pattern yet. All I can think now is that it is not really a King Safety issue.
At least not for most settings, because still very large scores for draw
positions keep coming up, -even for default Pro Deo?- or versions with
King_Safety instead of King Safety etc.
What I'm actually guessing is that with higher settings for things like
Attacking, or King Safety, the program is more likely to go search for an
attacking line and not a line that keeps the draw, or is more likely to change a
drawing line in search for something better. And because of the bug it often
finds a line that it thinks is much better than zero when King Safety is normal
or high, but is still really in a position that's a draw and the Black King
makes a fatal move somewhere in the line.
The tactical enginges therefore find more big scores in drawn positions than
settings with King Safety = 0. So it must be something in the search, or the
rules for this ending are just not powerful enough yet?
That would be my hypothesis.
Older Rebels by the way also seem to have the same behaviour? What I found
strange is that there is often a fail high, but this is sometimes immediately
recognized as a false fail high. So after a forty minute search for instance the
score goes up from zero to at least a piece up, and then in less than twenty
seconds the program recognizes this as a mistake. (I must confess that the
mechanism for fail-high and false fail high I don't understand so well yet)
Century 2.0 default in the following position, with a very small HT of 512 Kb:
Thinking....
+---+---+---+---+---+---+---+---+ -
8 | | | | | | | *K| |
+---+---+---+---+---+---+---+---+ 00:00:00 00:00:00
7 | | | | | | | | |
+---+---+---+---+---+---+---+---+
6 | | | | | | | | p | Time
+---+---+---+---+---+---+---+---+ Hash 512 Kb
5 | | | | | | | | | Engine Rebel Century 2.0
+---+---+---+---+---+---+---+---+ Personality Rebel.eng
4 | | | | | | | | |
+---+---+---+---+---+---+---+---+
3 | | | | | | | | |
+---+---+---+---+---+---+---+---+
2 | | | | | | | | |
+---+---+---+---+---+---+---+---+
1 | | | | | K | B | | |
+---+---+---+---+---+---+---+---+
a b c d e f g h
6k1/8/7P/8/8/8/8/4KB2 w - -
[Event ""]
[Site ""]
[Date "-9928.-54.-20"]
[Round ""]
[White ""]
[Black ""]
[Result "*"]
[BlackElo ""]
[WhiteElo ""]
[FEN "6k1/8/7P/8/8/8/8/4KB2 w - - 0 1"]
1.*
00:00 01.00 0.04 1.Kd2
00:00 01.03 0.04 1.Bc4+ Kh7
00:00 02.00 0.04 1.Bc4+ Kh7
00:00 03.00 0.00 1.Bc4+ Kh7 2.Kd2 Kxh6
00:00 03.06 0.00 1.Bd3
00:00 03.06 0.00 1.Bd3
00:00 03.06 0.04 1.Bd3 Kh8 2.h7 Kg7
00:00 04.00 0.04 1.Bd3 Kh8 2.h7 Kg7 3.Kd2
00:00 05.00 0.04 1.Bd3 Kh8 2.h7 Kg7 3.Kd2+
00:00 06.00 0.04 1.Bd3 Kh8 2.Kd2 Kg8 3.Bc4+ Kh7
00:00 07.00 0.04 1.Bd3 Kh8 2.Kd2 Kg8 3.Ke3 Kh8 4.h7
Kg7 5.h8=Q+ (0)
00:00 08.00 0.04 1.Bd3 Kh8 2.Kd2 Kg8 3.Ke3 Kh8 4.h7
Kg7 5.Kd4 (0)
00:00 09.00 0.04 1.Bd3 Kh8 2.Kd2
00:00 10.00 0.04 1.Bd3 Kh8 2.Kd2
00:00 11.00 0.04 1.Bd3 Kh8 2.Kd2
00:00 12.00 0.05 1.Bd3 Kh8 2.Kd2
00:00 13.00 0.05 1.Bd3 Kh8 2.Kd2
00:01 14.00 0.05 1.Bd3 Kh8 2.Kd2
00:10 15.00 0.05 1.Bd3 Kh8 2.Kd2 Kg8 3.Bf5 Kh8 4.Ke3
Kg8 5.Kd4 (1)
00:13 16.00 5.55 1.Bd3 Kh8 2.Kd1 Kg8 3.Ke2
00:17 17.00 0.05 1.Bd3 Kh8 2.Kd1
38:32 18.00 5.63 1.Bd3 Kh8 2.Kd1
38:44 19.00 0.05 1.Bd3 Kh8 2.Kd2 Kg8 3.Ke3+ Kh8 4.h7
Kg7 5.Kd4+ Kh8 6.Bf5 Kg7 7.Kd5 Kh8 (5917)
And here as another example with the "normal" Pro Deo Vulcan settings, nothing
extraordinary in the settings, but still an enormous number of false fail highs
this time:
After 1..Kh8 the following position is still a draw, only 1..Kf8 would have lost
(mate in twelve according to the Lokasoft tablebase server)
1..Kh8
[D]7k/8/7P/3K4/8/3B4/8/8 w - -
00:00:00.1 0,05 1 4 h7
00:00:00.1 0,07 1 6 Bf5
00:00:00.1 0,08 2 124 Bf5 Kg8
00:00:00.1 0,08 2 132 h7
00:00:00.1 0,08 2 135 h7
00:00:00.1 0,08 2 132 h7
00:00:00.1 0,07 3 329 Bf5 Kg8 Be4 Kh8 Kc6 Kg8 h7+ Kg7 Kd6 Kh8 Ke5 Kg7 Bf5 Kh8
Be4 Kg7 Bf5 Kh8
00:00:00.1 0,07 3 346 h7
00:00:00.1 0,07 3 349 h7
00:00:00.2 0,07 3 346 h7
00:00:00.2 0,08 3 554 Kd6 Kg8 Be4
00:00:00.2 0,08 4 1444 Kd6 Kg8 Be4 Kh8
00:00:00.2 0,10 4 2419 Ke5 Kg8 Kf5 Kh7 Kg5+ Kh8
00:00:00.2 0,10 5 3452 Ke5 Kg8 Kf5 Kh7 Kg5+ Kh8
00:00:00.2 0,10 6 5917 Ke5 Kg8 Kf6 Kh8 Kg6 Kg8 h7+ Kh8 Kg5 Kg7 Bf5 Kh8 Bd3 Kg7
Bf5 Kh8
00:00:00.2 0,10 6 6362 Bf5
00:00:00.2 0,10 6 6365 Bf5
00:00:00.2 6,36 6 6560 Bf5 Kg8 Be4 Kf7 Bd3
00:00:00.2 0,08 7 18918 Bf5 Kg8 Be4 Kh8 Kc6 Kg8 Kd6 Kh8 Ke5 Kg8 Kd6 Kh8 Ke5 Kg8
00:00:00.2 0,12 7 21807 Ke5 Kg8 Kf6 Kh8 Kg6 Kg8 h7+ Kh8 Kh6
00:00:00.2 0,13 8 42116 Ke5 Kg8 Kf6 Kh8 Bh7 Kxh7 Kg5 Kg8 Kg6 Kh8 Kf5 Kh7 Kg5 Kg8
Kg6 Kh8 Kf5 Kh7
00:00:00.2 0,13 9 55295 Ke5 Kg8 Bf5 Kh8 Kf6 Kg8 Bd3 Kh8 Bh7 Kxh7 Kg5 Kg8 Kg6 Kh8
Kf5 Kh7 Kg5 Kg8 Kg6 Kh8 Kf5 Kh7
00:00:00.2 0,13 9 55505 Bf5
00:00:00.2 0,13 9 55508 Bf5
00:00:00.2 6,26 9 64737 Bf5 Kg8 Be4 Kf7 Bh7 Kf6 Bc2 Kf7 Ba4 Kf6 Be8 Kg5 h7 Kf4
00:00:00.2 6,26 9 75369 Be4
00:00:00.2 6,26 9 64737 Bf5 Kg8 Be4 Kf7 Bh7 Kf6 Bc2 Kf7 Ba4 Kf6 Be8 Kg5 h7 Kf4
00:00:00.2 6,26 9 81108 h7
00:00:00.2 6,26 9 81111 h7
00:00:00.2 9,73 9 83171 h7 Kg7 Bb1 Kh8 Ke5 Kg7 Bg6 Kxg6
00:00:00.2 9,73 10 84687 h7 Kg7 Ke5 Kh8 Bb1 Kg7
00:00:00.3 9,73 11 90766 h7 Kg7 Ke5 Kh8 Bb1 Kg7 Bg6 Kxg6
00:00:00.3 9,73 12 108649 h7 Kg7 Ke5 Kh8 Bb1 Kg7 Bg6 Kxg6
00:00:00.3 0,12 13 259887 h7 Kg7 Ke5 Kh8 Be4 Kg7 Kf4 Kh8 Kg5 Kg7 Kh5 Kh8 Kh6
00:00:00.5 0,14 13 507090 Bf5 Kg8 Ke5 Kh8 Kf6 Kg8 Bh7+ Kh8 Bg8 Kxg8 Kg6 Kh8 Kf5
Kh7 Kg5 Kg8 Kg6 Kh8 Kf5 Kh7 Kg5 Kg8
00:00:01.7 0,14 14 2619785 Bf5 Kg8 Ke5 Kh8 Bd3 Kg8 Kf6 Kh8 Kg6 Kg8 Bc4+ Kh8 Bg8
Kxg8 Kf5 Kh7 Kg5 Kg8 Kg6 Kh8 Kg5 Kh7 Kh5 Kg8
00:00:01.7 0,14 14 2622951 h7
00:00:01.7 0,14 14 2622954 h7
00:00:01.7 0,14 14 2622951 h7
00:00:01.7 0,14 14 2641959 Kc5
00:00:02.0 0,14 14 2622951 h7
00:00:02.0 0,14 14 3153174 Kc6
00:00:02.0 0,14 14 2622951 h7
00:00:02.0 0,14 14 3287622 Kc4
00:00:02.3 0,14 14 2622951 h7
00:00:02.3 0,14 14 3780422 Ke5
00:00:02.5 0,14 14 2622951 h7
00:00:02.5 0,14 14 4155843 Kd4
00:00:02.7 0,14 14 2622951 h7
00:00:02.7 0,14 14 4576568 Be4
00:00:03.0 0,14 14 2622951 h7
00:00:03.0 0,14 14 5011535 Kd6
00:00:03.3 0,14 14 2622951 h7
00:00:03.3 0,14 14 5525781 Ke6
00:00:03.5 0,14 14 2622951 h7
00:00:03.5 0,14 14 5797233 Bc2
00:00:03.7 0,14 14 2622951 h7
00:00:03.7 0,14 14 6242746 Bb1
00:00:03.8 0,14 14 2622951 h7
00:00:03.8 0,14 14 6363514 Bg6
00:00:03.9 0,14 14 2622951 h7
00:00:04.7 0,14 15 8178980 Bf5 Kg8 Ke5 Kh8 Bd3 Kg8 Kf6 Kh8 Kg6 Kg8 Bc4+ Kh8 Kf6
Kh7 Kg5 Kh8 Bd5 Kh7 Bg8+ Kxg8 Kg4 Kh8 Kf4 Kg8
00:00:04.7 0,14 15 8214580 Bg6
00:00:04.7 0,14 15 8214583 Bg6
00:00:04.9 0,14 15 8214580 Bg6
00:00:04.9 0,14 15 8540268 Kc5
00:00:05.6 0,14 15 8214580 Bg6
00:00:05.6 0,14 15 9786453 Kc4
00:00:05.9 0,14 15 8214580 Bg6
00:00:05.9 0,14 15 10372556 Kd6
00:00:06.5 0,14 15 8214580 Bg6
00:00:06.5 0,14 15 11600831 Be4
00:00:07.0 0,14 15 8214580 Bg6
00:00:07.0 0,14 15 12517481 Kd4
00:00:07.3 0,14 15 8214580 Bg6
00:00:07.3 0,14 15 12923645 Ke5
00:00:07.7 0,14 15 8214580 Bg6
00:00:07.7 0,14 15 13710951 Bc2
00:00:08.3 0,14 15 8214580 Bg6
00:00:08.3 0,14 15 14945605 Kc6
00:00:09.0 0,14 15 8214580 Bg6
00:00:09.0 0,14 15 16266520 Ke6
00:00:09.3 0,14 15 8214580 Bg6
00:00:09.3 0,14 15 16724614 h7
00:00:09.3 0,14 15 8214580 Bg6
00:00:09.3 0,14 15 16769461 Bb1
00:00:09.4 0,14 15 8214580 Bg6
00:00:09.7 7,08 16 17507285 Bf5 Kg8 Ke5 Kh8 Bg6 Kg8 Be4 Kf7 Bf5 Kg8 Kd4 Kh8 Kc4
Kg8 Kd4 Kh8 Kc4 Kg8
00:00:09.8 7,08 16 17774843 Kd6
00:00:09.8 7,08 16 17774846 Kd6
00:00:10.3 7,08 16 17774843 Kd6
00:00:13.3 6,37 17 24821302 Bf5 Kg8 Ke5 Kh8 Be4 Kg8 Bd3 Kh8 Bf5 Kg8 Bd3 Kh8
00:00:39.3 10,01 18 75760204 Bf5 Kg8 Ke5 Kh8 h7 Kg7 Bd3 Kh8 Be4 Kg7 Bd3 Kh8 Be4
Kg7
00:01:15.2 0,14 19 143858237 Bf5 Kg8 Ke4 Kh8 Kd4 Kg8 Bd3 Kh8 Bf5 Kg8 Bd3 Kh8
00:05:14.5 0,14 20 593884122 Bf5 Kg8 Ke5 Kh8 Bd3 Kg8 Be4 Kh8 Kf5 Kh7 Kg5+ Kg8
Kf5 Kh7 Kg5+ Kg8 Kf5
00:05:14.5 0,14 20 593905880 Kc5
00:05:14.5 0,14 20 593905883 Kc5
00:05:29.3 0,14 20 593905880 Kc5
00:05:29.3 0,14 20 621380157 Kd6
00:06:01.1 0,14 20 593905880 Kc5
00:06:01.1 0,14 20 680950620 Ke5
00:07:18.3 0,14 20 593905880 Kc5
00:07:18.3 0,14 20 826761504 Kc4
00:08:36.6 0,14 20 593905880 Kc5
00:08:36.6 0,14 20 975077943 Kc6
00:10:14.1 0,14 20 593905880 Kc5
00:10:14.1 0,14 20 1158916821 Kd4
00:12:01.6 0,14 20 593905880 Kc5
00:12:01.6 0,14 20 1361275558 Ke6
00:13:04.5 0,14 20 593905880 Kc5
00:13:04.5 0,14 20 1478886630 Bg6
00:13:41.8 0,14 20 593905880 Kc5
00:13:41.8 0,14 20 1548489856 Be4
00:14:07.6 0,14 20 593905880 Kc5
00:14:07.7 0,14 20 1596776024 Bb1
00:15:55.1 0,14 20 593905880 Kc5
00:15:55.1 0,14 20 1799157178 Bc2
00:16:29.3 0,14 20 593905880 Kc5
00:16:29.4 0,14 20 1863241351 h7
00:16:36.0 0,14 20 593905880 Kc5
00:18:46.8 6,34 21 2131911239 Bf5 Kg8 Kd6 Kf7 Bd3 Kg8 Ke5 Kh8 Bf5 Kg8 Bd3 Kh8
Bf5 Kg8
I can't imagine King Safety alone could really be responsible for generating all
these false Fail Highs Ed?
This was with Pro Deo 1.1 Vulcan {9k16ip} settings
-------------------------------------------------------------------------------
[Personality = {9k16ip}] * Pro Deo 1.1 (February 2005)
[Pawn Value = 90]
[Knight Value = 90]
[Bishop Value = 90]
[Rook Value = 90]
[Queen Value = 95]
[My Queen = 96]
[King Safety = 170]
[Mobility = 125]
[Pawn Structure = 100]
[Passed Pawns = 110]
[Pins = 100]
[Bad Bishop = 100]
[Bishop Pair = 100]
[Chess Knowledge = 300]
[Attractiveness = 110]
[Attacking = 110]
[Strength of Play = 100]
[Draw Contempt Factor = 0.00]
[Selective Search = 100]
[Search Technique = NULLMOVE]
[Engine Master = off] off
[Engine Learner = off] off|on
[Book Learner = off] off|passive|moderate|strong|aggressive
[Position Learner = off] off|on
[Extended Book Learner = off] off|read|write|read&write
*
[Pruning = MISC_25] *
[Pruning = MISC_37] * Hash Table extension (-18%) DEFAULT
[M37_Val = 3.00] * Default unknown
[Pruning = MISC_40] *
[Pruning = MISC_52] *
[QS Margin = 10.00] * Default is 3.00
[Fail High Margin = 10.00] * Default is 9.00
[Checks Depth = 4] * Limit QS checks to 4 ply, default is 2
[Pruning = MISC_66] *
[Pruning = MISC_69] *
[Pruning = MISC_78] *
*
[Center Control = 35] *
[Bishop Mobility = 25] *
[Right to Move = 85] *
[Strong Squares = 125] *
*
[Pruning = MISC_02] *
[Pruning = MISC_03] *
[Pruning = MISC_39] *
[Pruning = MISC_08] *
[Pruning = MISC_12] *
[Pruning = MISC_76] * Rook 7th rank (new style)
[Rook 7th Rank (low) = 60] *
[Rook 7th Rank (high) = 60] *
[Pruning ? MISC_48] * No FHR pruning in QS
[Pruning = MISC_17] * Skip first extension (+30-35%)
[Pruning = MISC_10] * NULLMOVE Block first extension
[Pruning = MISC_22] * Sort root moves on nodes
[Pruning ? MISC_30] * Extend singular checks
[Pruning ? MISC_36] * Extend double attacks (W6) at the leafs
[Pruning ? MISC_28] * (2,2,1,1) >> in test <<
[Pruning = MISC_13] * Extra bonus Bishop Pair
[Pruning = MISC_27] * Eval double pawns (new style)
[Futility = 100] * Futility Pruning.
[Blitz Time Control = 120] * use more time at blitz levels
[Eternal Checks = Normal] * Algorithm to detect eternal checks
*
[Pawn Pressure midg = 150] * weight pawn pressure middlegame
[Pawn Pressure end0 = 130] * weight pawn pressure early endgame
[Pawn Pressure end1 = 120] * weight pawn pressure endgame
[Pawn Pressure end2 = 110] * weight pawn pressure simple endgame
*
[Knight Mobility = 88] * 88
[Double Pawns = 110] * 112
[Pawn Formation = 110] * 108
[Weak Pawns = 110] * 112
[Minimum Knight Mobility = 100] * 100
[Minimum Bishop Mobility = 100] * 100
[Passed Pawn King Tropism = 100] * king tropism to own passed pawns
[Progressive Isolated Pawns = 100] * weight isolated pawns (new style)
*
*
[ANTI-GM = OFF]
[EVALUATION = NORMAL]
[Extensions (remaining)= 3]
[Extensions (checks)= normal] * low|moderate|normal|strong|all
[Extensions (captures)= 0]
-------------------------------------------------------------------------------
Hartelijke groeten ,
Eelco
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.