NAUM 4.2

[mcostalba]

LOL. Frustrating isn't it. It would be so much easier if you could just copy the algorithm
For the last 2 weeks I am trying various combinations of the BSR pruning (reduce later moves more). Engine goes so much deeper, but it sucks in games

LMR is very dependant on quality of your move sorting: if you have a good sorting policy with accurate data you can push on the LMR pedal, otherwise you should not.

We have failed for months with aggressive LMR and we started even before SF 1.6 was out (actually in 1.6 LMR is unchanged from earlier release because we failed all our attempts to increase it), but very recentely we may have found the right receipe

[naum]

My LMR is extremely aggressive. I just want to make it super-extremely aggressive and add your futility pruning on top of it.
Just that. Is it too much to ask

Actually yesterday I finally found formula that seems to be working, but I want more.

[mcostalba]

My LMR is extremely aggressive. I just want to make it super-extremely aggressive and add your futility pruning on top of it.
Just that. Is it too much to ask

Nice to know we are working in the same direction

Actually yesterday I finally found formula that seems to be working, but I want more.

Of course !

[ernst]

In general, how long does it take to receive the engine after a paypal payment?

Edit: Never mind, Alex already answered me by mail.

[Uri Blass]

My LMR is extremely aggressive. I just want to make it super-extremely aggressive and add your futility pruning on top of it.
Just that. Is it too much to ask

Actually yesterday I finally found formula that seems to be working, but I want more.

I wonder if you plan to improve Naum's evaluation

Here is a drawn position when Naum gives evaluation of more than +3 pawns for white.

see my subject about test position

[d]R7/P4p2/5k2/8/6P1/r7/7K/8 w - - 0 50

I do not expect top programs to see the draw by evaluation but expect them at least to be smart enough not to give +3 to drawn positions when white is only one pawn up.

You can also modify the position and still get a draw.

[d]1R6/1P3pk1/8/6P1/8/1r6/7K/8 w - -

I think that top programs should know that pawn in the 7th that is blocked by a friendly rook is a bad structure in rook endgames.

Uri

[Werner]

Hi Uri,
this is a must (look what Rybka does):

R7/P4p2/5k2/8/6P1/r7/7K/8 w - -

Engine: Rybka 3 x64 1CPU (64 MB)
von Vasik Rajlich, Larry Kaufman

22.01 0:23 +3.38 1.g5+ Kg7 2.Kg2 Ta6 3.Kg3 Kh7 4.Kh4 Ta4+
5.Kh3 Ta6 6.Kg3 Kg7 7.Kg4 (1.003.921) 44
TB:11.442

[ernst]

DF10 gives it a draw score

New game
R7/P4p2/5k2/8/6P1/r7/7K/8 w - - 0 1

Analysis by Deep Fritz 10:

1.Kh2-g2
+/- (1.32) Depth: 18/50 00:00:05 23910kN, tb=52420
1.Kh2-g2 Kf6-g7
+/- (1.04) Depth: 19/31 00:00:05 26010kN, tb=61416
1.Kh2-g2 Kf6-g7 2.g4-g5
+/- (1.04) Depth: 19/55 00:00:06 29316kN, tb=70775
1.g4-g5+
+/- (1.04) Depth: 19/63 00:00:07 36909kN, tb=99206
1.g4-g5+ Kf6-g7 2.Kh2-g2 Ra3-a4 3.Kg2-f3 Kg7-h7 4.Kf3-e2 Ra4-a2+ 5.Ke2-d3 Ra2-a3+ 6.Kd3-d4 Kh7-g7 7.Kd4-c5 Ra3-a1 8.Kc5-d5 Ra1-a4 9.Kd5-d6
+/- (1.27) Depth: 20/61 00:00:12 63447kN, tb=210316
1.g4-g5+ Kf6-g7
+/- (0.99) Depth: 21/61 00:00:13 71569kN, tb=242898
1.g4-g5+ Kf6-g7 2.Kh2-g2 Ra3-a1 3.Ra8-d8 Ra1xa7 4.Kg2-f3 Kg7-g6 5.Kf3-f4 Ra7-a4+ 6.Kf4-g3 Kg6xg5
= (0.00) Depth: 21/95 00:00:20 108mN, tb=363351
1.Kh2-g2
= (0.01) Depth: 21/95 00:00:25 141mN, tb=436630
1.Kh2-g2
= (0.16) Depth: 21/95 00:00:30 174mN, tb=510632
1.Kh2-g2
+/= (0.47) Depth: 21/95 00:00:35 212mN, tb=584616
1.Kh2-g2 Ra3-a2+
= (0.19) Depth: 22/97 00:00:50 287mN, tb=729409
1.Kh2-g2 Ra3-a2+ 2.Kg2-g3 Kf6-g7 3.Ra8-d8 Ra2-a3+ 4.Kg3-g2 Ra3-a2+ 5.Kg2-g1 Ra2-a1+ 6.Kg1-f2
= (0.00) Depth: 22/97 00:01:13 426mN, tb=1073599
1.Kh2-g2 Ra3-a2+ 2.Kg2-g3 Kf6-g7 3.Ra8-d8 Ra2-a3+ 4.Kg3-g2 Ra3-a2+ 5.Kg2-g1 Ra2-a1+ 6.Kg1-f2 Ra1-a2+ 7.Kf2-e1 Ra2-a1+ 8.Ke1-e2 Ra1-a2+ 9.Rd8-d2 Ra2xa7 10.g4-g5 f7-f5 11.Ke2-f3
= (0.00) Depth: 23/97 00:01:42 591mN, tb=1527986

[naum]

Hi Werner,

You have unproportionally large number of unusual openings in this match.
Number of Nf3 openings is almost two times larger then the number of Spanish and Sicilian openings put together.
Only 4 Spanish games (Naum won 2 Rybka 1).

Hence the bad score for Naum. These openings really favour Rybka.
I suspect that book was most likely tuned using Rybka.

Regards,
Alex

[naum]

This is my excuse and I am sticking to it

[Werner]

This is my excuse and I am sticking to it

this book has:
e4 25%
d4 25%
Sf3 25%
c4 25%

and after 1. e4
c5 25%
e5 25%
e6 25%
c6 25 %

...and it is from Salvo Spitaleri not from Rybka Team
I can send it to you for tuning Naum
Next match is against Deep Fritz 12 using Silver-Suite!
Werner