Joker in my Division Junior now on 3rd

[hirdelgird]

Motor Punkte Ro Cy Jo Re Bu Co Ai Ex To An Gr Sc Ol An Ne Na Bl Mo S-B
01: Romi P3k 23,0/29 ·· =1 0 10 11 1 == 1 11 11 =0 11 1 11 1= 1 =1 11 318,75
02: Cyrano 0.30 22,5/30 =0 ·· 1= 1 =1 1= 0= 1= 1 1= 11 11 1 1 1= 11 0= 11 314,75
03: Joker 1.1.14 20,0/30 1 0= ·· == 01 01 == 11 10 01 11 0 01 11 1 11 1= 1 278,75
04: Resp 0.19ja 19,5/30 01 0 == ·· 0 00 11 1= =1 1= 1 11 10 11 10 01 11 1 262,75
05: Butcher 1.58 18,5/30 00 =0 10 1 ·· 1= = == 0 =0 11 11 11 0 1= =1 =1 11 241,50
06: Counter 0.80 18,0/30 0 0= 10 11 0= ·· 0= 10 1= 0 10 1 11 == 11 11 1 == 240,50
07: Aice 0.99.2 17,0/30 == 1= == 00 = 1= ·· 10 00 1 0= 11 10 0 =1 11 1 =1 240,25
08: Ex 5.01b 16,5/30 0 0= 00 0= == 01 01 ·· =1 0 == 1 =1 11 = 11 11 10 208,50
09: Tornado 1.00 16,0/30 00 0 01 =0 1 0= 11 =0 ·· 11 0 =1 0= = 11 00 11 11 209,75
10: Anechka 0.08 16,0/30 00 0= 10 0= =1 1 0 1 00 ·· 10 01 01 01 11 =1 1 11 205,75
11: GreKo 5.50 14,5/30 =1 00 00 0 00 01 1= == 1 01 ·· 10 1 10 == 0 11 =1 196,25
12: Scid 3.61ja 13,0/30 00 00 1 00 00 0 00 0 =0 10 01 ·· 10 11 11 11 = 11 144,00
13: OliThink 5.09 11,5/29 0 0 10 01 00 00 01 =0 1= 10 0 01 ·· 1= 1 10 00 1 157,75
14: Ant 2006f 10,0/30 00 0 00 00 1 == 1 00 = 10 01 00 0= ·· 01 == 01 10 130,75
15: Neurosis 2.40 9,5/30 0= 0= 0 01 0= 00 =0 = 00 00 == 00 0 10 ·· 0 11 11 121,75
16: NagaSkaki 4.00 9,0/30 0 00 00 10 =0 00 00 00 11 =0 1 00 01 == 1 ·· 10 0 122,25
17: BlackBishop 0.47 8,0/30 =0 1= 0= 00 =0 0 0 00 00 0 00 = 11 10 00 01 ·· 0= 116,25
18: Monarch 1.70 6,5/30 00 00 0 0 00 == =0 01 00 00 =0 00 0 01 00 1 1= ·· 81,25

[Michael Sherwin]

Joker is gaining ground fast. Probably another engine that will soon overtake Romi. But, right now I am enjoying the moment! Tomorrow I will continue to plug along. Thanks for the surprise!

[hgm]

And I haven't really worked on it for half a year now! (Since Open War 3 started, and that version was a bust, which I now retracted.)

Is this blitz? I noticed that Joker is much stronger (compared to other engines) at blitz than at long time controls. The most likely explanation is that it has a comparatively large branching ratio, probably due to poor move ordering. Although it could be that the fact that Joker only uses R=2 null-move pruning also contributes to this. I understand that most people use R=3, or at least adaptive (R=3 at larger depthLeft, R=2 close to the leaves). I never really tested what was better at large time control, as I lack the facilities to do such time-consuming testing.

[hirdelgird]

sorry but as mostly i am in hurry so only copy and paste from my webside:
================================================
Conditions:
blitz-time-control: 40 moves in 5 minutes repeating
ponder: off
used Gui: topical stable ARENA-GUI
used CPU:
Hardware: Intel Pentium M processor 1.733 GHz
Betriebssystem: MS Windows XP Service Pack 2
Hash tables: all Engines up to 64 MB (Ed1to Ed6 there was 64 MB for Division A to C, other 32 MB)
Pawn Hash: 4 MB (pawn hash settings is changed by request of author)
EGTB: no
in the german chess magazin "CSS" (paper edition - year 2004/2005) the EGTB were discused controverse.
test-results said better to do without (among others speed of engine is reduced down to 30 per cent when use it)
furthermore not all engines are able to handle it
OpeningBook
all Engines use own book if available (Ed1to Ed6 there was own book only for Division A to C, others use book from Harry Schnpp)
for those engines whithout book "Arena little mainbook 1.00" from Harry Schnapp is used
Download PGN
download of the pgn-files is possible, files larger > 1 MB are packed with WinRar ......

[Uri Blass]

And I haven't really worked on it for half a year now! (Since Open War 3 started, and that version was a bust, which I now retracted.)

Is this blitz? I noticed that Joker is much stronger (compared to other engines) at blitz than at long time controls. The most likely explanation is that it has a comparatively large branching ratio, probably due to poor move ordering. Although it could be that the fact that Joker only uses R=2 null-move pruning also contributes to this. I understand that most people use R=3, or at least adaptive (R=3 at larger depthLeft, R=2 close to the leaves). I never really tested what was better at large time control, as I lack the facilities to do such time-consuming testing.

I use R=3 always and
I suspect that R=3 works better if you do checks in the first plies of the qsearch.

I never tested R=3 without checks in the qsearch but based on my memory for movei R=3 was better than R=2 at blitz.

I tested it some years ago when hardware was slower than today.
Note that I believe that Movei should have poor move ordering because it does not do iid but inspite of that movei's rating is higher at longer time control based on CCRL and I have no idea what is the reason for it.

I also did not work on Movei in the last months and my programming time in the last months is devoted to modifying strelka and understanding it(I think that the code of strelka is too complex because it has specific code both for white and for black and I try to make it simpler and I already got rid of color dependent code in the move generator with no significant change in speed).

I think that it may be interesting to see if strelka with some simpler evaluation function like micromax's evaluation function is better or worse than Joker and I may try to change strelka's evaluation to test it(I still did not decide if I am going to do it).

Uri

[Uri Blass]

Unfortunately I find micromax's evaluation as too complex to implement in strelka because it has too many scores that are dependent on the path and not on the final position.

I found that simple evaluation of strelka that include no material imbalance table,no mobility,no pawn structure and no king safety is clearly enough to bear joker at 1+1 so it seems that joker has no strong search even at 1+1(because I cannot imagine that the evaluation of joker is worse than it).

The evaluation that I talk about include only average of piece square table scores between opening and endgame based on stage of the game and static small bonus for the side to move.

Uri

[hgm]

Yes, this is an unfortunate property of uMax. I sacrificed consistency and accuracy of the evaluation for reducing the size. This leads to inconsistent pathe-dependent scores. But as long as the magnitude of the inconsistency is not larger than the typical evaluation error, other was almost no adverse effect of this.

It is possible to make a consistent, path-independent evaluation that is very micro-Max-like in character, though. It would use the center-points table of uMax as piece-square table for N+B, and for P is would use one that added 9 cP per rank to these center points, plus 64 and 128 cP bonuses for 6th and 7th rank, respectively. But you would have to give a 9 cP penalty for each square on the board that can no longer be attacked by a pawn (without capture) because the pawns on the neighboring files all passed it. And the 9pts per rank would actually have to be scaled with total Piece material on the board, so that it would go up to 19 pts per rank in a Pawn ending.

The most difficult thing to implement is King safety: uMax gives a 20 cP penalty for each King move (except castling). The best way to approximate this is to have piece-square tables for King that give 20 cP penalty per move along the shortest part to reach each square from the original King position (for an uncastled King), or from g1/g8 or c1/c8 for a castled King in the middle game. And a 9 cP bonus for having a Pawn directly in front of the King.

[hirdelgird]

hello

the games are finished, sorry Joker missed at 2197... may be next time with more luck

blitz 40/5

1 Cyrano 0.30 52.0 / 68 2332
2 Romi P3k 51.0 / 68 2319
3 Aice 0.99.2 44.5 / 68 2216
4 Resp 0.19ja 44.0 / 68 2238

5 Joker 1.1.14 40.0 / 68 2197
6-18 please visit my webside

download of pgn is available with comments

[hgm]

Yes, a narrow escape. I asked because I saw many engines in division E of your previous edition that Joker would easily rank above. And indeed it did in this tournament. It is just Joker's bad luck that you boot them all...

[hirdelgird]

... ok you are right,

I have added same more infos (ratings) to div junior and div e. The difference in strenght is small also when look in div d in second half of table

in last editions i do more for the higher divisions, but i myself was not pleased with that. So from beginning of these edition all engines have 4 RR, even div junior. Also to help the lower divisions to get a better "design in strenght" there will be 6 promoters and 6 descenders for each division

the idea was to create a div senior and junior with elo-barrier 2.200 so i can better handle these amounts of free available chess engines

when you have any suggest to do please tell me, so i will think it over