Glaurung 080420 evalaution

[Tony Thomas]

I know that I have seen the options before, and not being able to recall it is bothering me..

[Dr.Wael Deeb]

it would be an interesting psychological experiment to see how the names of your parameters influence how people create personalities. e.g., now you have a parameter called "cowardice." Who amongst us really wants a more cowardly engine?

Maybe you should call aggressiveness something like "bullying" and you should call "cowardace" something like "disciplined restraint"

Actually, the names "aggressiveness" and "cowardice" are taken from another well-known UCI chess engine. Can anyone guess which one?

If I had chosen names without external influence, I would probably have ended up with something dull like "King safety (engine)" and "King safety (opponent)". Perhaps this would have been less confusing to users.

Tord

SmarThink

[ozziejoe]

Glaurung is looking like a very nice improvement over epsilon 5


Toga II 1.4 beta5c versus glaurung 080420 --- +26/-15/=19

Toga II 1.4beta5c versus glaurung epsilon 5---- +33/-6/21

This result somewhat supports my contention that in addition to being better, the new glaurung is more aggressive than epsilon 5 (e.g., as indicated by the ratio of points gotten from wins versus draws)

Surely a few of you parameter-adjusters can give glaurung enough of a boost to catch toga? If you can improve cm by 50 points, why not glaurung?

Bright is next up


best
J

[Tony Thomas]

it would be an interesting psychological experiment to see how the names of your parameters influence how people create personalities. e.g., now you have a parameter called "cowardice." Who amongst us really wants a more cowardly engine?

Maybe you should call aggressiveness something like "bullying" and you should call "cowardace" something like "disciplined restraint"

Actually, the names "aggressiveness" and "cowardice" are taken from another well-known UCI chess engine. Can anyone guess which one?

If I had chosen names without external influence, I would probably have ended up with something dull like "King safety (engine)" and "King safety (opponent)". Perhaps this would have been less confusing to users.

Tord

SmarThink

That's the first engine I checked. Its not part of the UCI options, its part of the configuration file though..

[Dr.Wael Deeb]

it would be an interesting psychological experiment to see how the names of your parameters influence how people create personalities. e.g., now you have a parameter called "cowardice." Who amongst us really wants a more cowardly engine?

Maybe you should call aggressiveness something like "bullying" and you should call "cowardace" something like "disciplined restraint"

Actually, the names "aggressiveness" and "cowardice" are taken from another well-known UCI chess engine. Can anyone guess which one?

If I had chosen names without external influence, I would probably have ended up with something dull like "King safety (engine)" and "King safety (opponent)". Perhaps this would have been less confusing to users.

Tord

SmarThink

That's the first engine I checked. Its not part of the UCI options, its part of the configuration file though..

Yes,it's a part of SmarThink's ini file....

[Eelco de Groot]

Hello,

Thanks to Ryan, Shaun and Tony for their results!

It looks like the new version might be strong enough to merit a public release, even in its current raw and untuned stage. I can always release a more polished version later, if I or somebody else can come up with significantly better settings.

At any rate, I'll probably spend at least another week or two running experiments and cleaning up the code before I release Glaurung 2.1.

Here is +5 and -5 to the king attack values to check on how well they are in tune with the rest of eval. I think the default value is very good. The next natural step is to try new values in the SafetyTable[] array. Based on experience I think it would be good to try with values climbing more quickly and peaking sooner. I could easily be wrong though.

Quite possible. The current development version actually contains several new UCI parameters for configuring how the king safety values climb. The SafetyTable[] array, from which the king safety scores are taken, is no longer a constant array, but is computed based on the values of UCI parameters.

Tord

Hi Tord,

I only just checked what influence the King Safety X Intercept, King Safety Max value, King Safety Curve, King Safety Coefficient have in my Windows builds and I think something is still not working when reading the UCI parameters. When I change King Safety from 100 to 1000 I see some differences but very minimal, usually only in nodecounts. There is some influence I think from other changes because I am using

Code: Select all

      // Finally, extract the king safety score from the SafetyTable[] array.
      // Add the score to the evaluation, and also to ei.futilityMargin.  The
      // reason for adding the king safety score to the futility margin is
      // that the king safety scores can sometimes be very big, and that
      // capturing a single attacking piece can therefore result in a score
      // change far bigger than the value of the captured piece.
      assert(inc < 100);
      ei.mgValue -= ((1000 - ei.mgValue) * sign * Value(SafetyTable[inc])/256);
      if(c == p.side_to_move())
        ei.futilityMargin += Value(2 * SafetyTable[inc]);
    }
  }

or in the present version from 080422 that has changed to the latest try

Code: Select all

      // Finally, extract the king safety score from the SafetyTable[] array.
      // Add the score to the evaluation, and also to ei.futilityMargin.  The
      // reason for adding the king safety score to the futility margin is
      // that the king safety scores can sometimes be very big, and that
      // capturing a single attacking piece can therefore result in a score
      // change far bigger than the value of the captured piece.
      Value v = apply_weight(((1000 - sign * ei.mgValue) * Value(SafetyTable[inc]))/256, WeightKingSafety[c]);
      ei.mgValue -= sign * v;
      if(c == p.side_to_move())
        ei.futilityMargin += v;
    }
  }

but that does not explain the minimal differences. I think I have been testing Glaurung 080422 until now with SafetyTable() initialized with very small values or zeroes, otherwise the changes for the weights should have had more effect? I tried changing a,b, maxslope and peak to type int but I don't think that makes a difference so far. What I have not yet tried is making SafetyTable() an array of ints. I was wondering if anybody else has encountered similar things under Windows Tord? There are no error warnings from the Microsoft Visual C++ 2005 compiler, only if I try to read floats into this piece of code

Code: Select all

    for(i = 0; i < 100; i++) {
      if(i < b) SafetyTable[i] = Value(0);
      else if(get_option_value_string("King Safety Curve") == "Quadratic")
        SafetyTable[i] = (Value)(a * (i - b) * (i - b));
      else if(get_option_value_string("King Safety Curve") == "Linear")
        SafetyTable[i] = (Value)(100 * a * (i - b));
    }

I am a bit confused what could be the actual problem at the moment But Glaurung was still playing good with the - minimal? - King Safety, so now I am very intrigued as to how I can get it to work again and what will happen then in Glaurung Mjolnir's testgames

Regards, Eelco

[Eelco de Groot]

Sorry, something must have gone wrong because I have to totally disable King safety to see any different result in the following position that was my testposition:

[d]3r1bk1/5p1p/p1p2np1/2P1p3/NP1nP3/4B1PP/5PB1/1R4K1 w - -

with

Code: Select all

  // Kings.  Kings are evaluated after all other pieces for both sides,
  // because we need complete attack information for all pieces when computing
  // the king safety evaluation. 
  // [EdG: Evaluation of kings moved to after passed pawns eval]
  //

  /*for(Color c = WHITE; c <= BLACK; c++) {
    s = pos.king_square(c);
    evaluate_king(pos, s, c, ei);
  }*/

I get


3r1bk1/5p1p/p1p2np1/2P1p3/NP1nP3/4B1PP/5PB1/1R4K1 w - -

Engine: Glaurung 080422 Mjolnir 005 (64 MB)
by Tord Romstad

2.00 0:00 +0.23 30.Nc3 h5 (323) 5

3.00 0:00 +0.35 30.Nc3 h5 31.Rd1 (683) 11

4.00 0:00 +0.45 30.Nc3 h5 31.Rd1 Be7 (2.251) 36

5.00 0:00 +0.47 30.Nc3 h5 31.Ra1 Rb8 32.Nd5 (5.036) 64

6.00 0:00 +0.37 30.Nc3 Kg7 31.Bf1 Nb5 32.Nxb5 axb5
33.f3 (10.639) 136

7.00 0:00 +0.35 30.Nc3 Kg7 31.Bf1 Nb5 32.Nxb5 axb5
33.f3 Be7 (17.850) 189

8.00 0:00 +0.52 30.Nc3 h6 31.Bf1 Nc2 32.Bc1 Rd4
33.Bxa6 Nxe4 34.Nxe4 Rxe4 (43.841) 310

9.00 0:00 +0.62 30.Nc3 h6 31.Bf1 Nb5 32.Nxb5 axb5
33.f3 Kg7 34.Ra1 Be7 (115.971) 435

10.00 0:00 +0.62 30.Nc3 h5 31.Ra1 Nc2 32.Rxa6 Nxb4
33.Rb6 Rd4 34.Rb8 Nd3 35.Bxd4 exd4 (407.356) 554

11.00 0:01 +0.58 30.Nc3 Nc2 31.Nd5 Nxe3 32.Nxf6+ Kg7
33.Nxh7 Bxc5 34.bxc5 Nxg2 35.Kxg2 Kxh7
36.Kf3 Rd3+ 37.Ke2 (962.304) 592

12.01 0:02 +0.29 30.Nc3 Nc2 31.Bf1 Nxe3 32.fxe3 Rd2
33.Bxa6 Rc2 34.Rb3 Bh6 35.Nd5 Kg7
36.Bd3 (1.683.862) 612

12.02 0:03 +0.58 30.Nb2 Nc2 31.Bg5 Kg7 32.Nc4 Rd4
33.Nxe5 Rxb4 34.Rd1 Nd4 35.Nd3 Rb8 (2.120.943) 619

13.01 0:05 +0.70 30.Nb2 Nd7 31.Rd1 Rb8 32.Nd3 f6
33.f3 Kg7 34.Kf2 Be7 35.h4 h6 36.Bh3 (3.538.526) 625

13.03 0:07 +0.82 30.Nb6 Nc2 31.Bg5 Kg7 32.Nc4 Rd4
33.Nxe5 Rxb4 34.Rd1 Nd4 35.Nd3 Rb8
36.e5 (4.551.109) 633

14.01 0:09 +0.84 30.Nb6 Nc2 31.Bg5 Kg7 32.Nc4 Rd4
33.Nxe5 Rxb4 34.Rd1 Nxe4 35.Nxc6 Nxg5
36.Nxb4 Nxb4 (5.852.824) 638

15.01 0:15 +0.92 30.Nb6 Nc2 31.Bg5 Kg7 32.Nc4 h6
33.Nxe5 hxg5 34.Nxc6 Rd2 35.e5 Nd7
36.Be4 Na3 37.Ra1 (9.608.393) 640

16.01 0:24 +0.60 30.Nb6 Nc2 31.Bg5 Kg7 32.Nc4 Rb8
33.Nxe5 Rxb4 34.Rd1 Nxe4 35.Nxc6 Nxg5
36.Nxb4 Nxb4 37.c6 Nxc6 38.Bxc6 Nxh3+
39.Kg2 (15.575.452) 644

16.05 1:07 +0.96 30.Rb2 Nb5 31.Kh2 Bg7 32.Ra2 Ne8
33.Bf1 Nf6 34.f3 Rd1 35.Kg2 Rb1
36.Nb6 Rxb4 37.Nc4 Nc3 38.Rxa6 (42.305.314) 629

17.01 1:31 +0.94 30.Rb2 Bg7 31.Nb6 Ne8 32.Ra2 Nc7
33.Bf1 Nf3+ 34.Kg2 Ne1+ 35.Kh1 Rd1
36.Bxa6 Nxa6 37.Rxa6 Nd3+ 38.Kg2 Nxb4
39.Ra7 Bf6 40.f3 (58.085.274) 638

18.01 2:39 +0.86 30.Rb2 Ne8 31.Nb6 Nc7 32.Bg5 Rb8
33.Ra2 h6 34.Be3 Rd8 35.Bf1 Nf3+
36.Kg2 Ne1+ 37.Kh1 Nf3 38.Bxa6 Rd1+
39.Kg2 Ne1+ 40.Kf1 Nc2+ 41.Ke2 Re1+
42.Kd2 Nxa6 43.Rxa6 (102.047.436) 641

best move: Rb1-b2 time: 2:50.859 min n/s: 641.111 nodes: 109.530.000

In all other cases I get, with only minor differences in nodecounts:

Engine: Glaurung 080422 Mjolnir 005 (64 MB)
by Tord Romstad

2.00 0:00 +0.21 30.Nc3 Be7 (332) 5

3.00 0:00 +0.33 30.Nc3 Be7 31.Rd1 (658) 10

4.00 0:00 +0.47 30.Nc3 Be7 31.Bf1 Nf3+ 32.Kg2 (2.846) 45

5.00 0:00 +0.37 30.Nc3 Be7 31.Bf1 Nb5 32.Nxb5 axb5 (4.442) 70

6.00 0:00 +0.47 30.Nc3 Be7 31.Bf1 Nb5 32.Nxb5 axb5
33.f3 (8.563) 109

7.00 0:00 +0.37 30.Nc3 Be7 31.Bf1 Nb5 32.Nxb5 axb5
33.f3 Kg7 (15.551) 165

8.00 0:00 +0.52 30.Nc3 h6 31.Bf1 Nc2 32.Bc1 Rd4
33.Bxa6 Nxe4 34.Nxe4 Rxe4 (48.791) 346

9.00 0:00 +0.60 30.Nc3 h6 31.Bf1 Nb5 32.Nxb5 axb5
33.f3 Kg7 34.Ra1 Be7 (101.273) 432

10.00 0:00 +0.72 30.Nc3 Be7 31.Bf1 Nc2 32.Bg5 Rd4
33.Bxa6 Nxb4 34.Be2 Nd3 (303.965) 511

11.00 0:01 +0.58 30.Nc3 Nc2 31.Nd5 Nxe3 32.Nxf6+ Kg7
33.Nxh7 Bxc5 34.bxc5 Nxg2 35.Kxg2 Kxh7
36.Kf3 Rd3+ 37.Ke2 (628.845) 543

12.01 0:01 +0.19 30.Nc3 Nc2 31.Bg5 Be7 32.Rb2 Nd4
33.Ra2 Rb8 34.Ra4 Kg7 35.Rxa6 Rxb4 (970.888) 559

12.04 0:03 +0.43 30.Bg5 Bg7 31.Nb6 h6 32.Bxf6 Bxf6
33.Nd5 Kg7 34.Ra1 Bg5 35.Rxa6 Ne2+
36.Kf1 Nd4 (1.967.590) 577

12.06 0:05 +0.68 30.Nb2 Nc2 31.Bg5 Kg7 32.Nc4 Rd4
33.Nxe5 Rxb4 34.Rd1 Nd4 35.Nd3 Rb8 (3.352.771) 586

13.01 0:07 +0.84 30.Nb2 Nc2 31.Bg5 Kg7 32.Nc4 Rd4
33.Nxe5 Rxb4 34.Rd1 Nxe4 35.Nxc6 Nxg5
36.Nxb4 Nxb4 (4.586.599) 593

14.01 0:10 +0.84 30.Nb2 Nc2 31.Bg5 Kg7 32.Nc4 Rd4
33.Nxe5 Rxb4 34.Rd1 Nxe4 35.Nxc6 Nxg5
36.Nxb4 Nxb4 (6.168.620) 599

15.01 0:15 +0.72 30.Nb2 Nc2 31.Bg5 Kg7 32.Nc4 h6
33.Bxh6+ Kxh6 34.Nxe5 Rd2 35.Nxf7+ Kg7
36.Ne5 Nd4 37.Ra1 Nc2 38.Nc4 (9.550.944) 605

15.06 0:32 +1.09 30.Rb2 Bg7 31.Nb6 Ne8 32.Nd5 f5
33.Bg5 Rd7 34.Nb6 Rb7 35.Ra2 fxe4
36.Bxe4 Nf6 37.Bxf6 Bxf6 38.Rxa6 (19.598.481) 604

16.01 0:40 +0.84 30.Rb2 Bg7 31.Nb6 Ne8 32.Ra2 Nc7
33.Bf1 Nf3+ 34.Kg2 Ne1+ 35.Kh1 Rd1
36.Bxa6 Nxa6 37.Rxa6 Nc2+ 38.Kg2 Nxb4
39.Ra8+ Bf8 40.Ra7 (24.792.084) 606

best move: Rb1-b2 time: 0:44.078 min n/s: 606.170 nodes: 26.700.000

Maybe I am skipping some part of the code so nothing gets counted...

The SafetyTable[] is probably fine, I replaced it wth the constans from the earlier version, there was still no change in output

[Tord Romstad]

SmarThink

That's the first engine I checked. Its not part of the UCI options, its part of the configuration file though..

SmarThink is indeed the right answer! You both win a free copy of Glaurung 2.1, when it is finished. Congratulations!

I remembered wrong about the SmarThink UCI options. As you point out, these are not UCI options in SmarThink, but configuration file options.

Tord

[Tord Romstad]

Eelco,

Glaurung does not evaluate king safety in positions without queens.

Tord

[Werner]

Hi,
in Smarthink Moscow you will find these uci parameters:

SmarThink v1.1 [r1119/27.02.2008] (SmarThink-Moscow)
104704 nodes in openings book
Transposition size is 8388608 entries (201326592 bytes)
Pawn hash size is 524288 entries (16777216 bytes)
EGTBHashSize=1048576 bytes
Aggressivity=100 (50)%
Cowardice=100 (50)%
Tropism=100 (200)%
ResignValue=999
ResignTimes=9
UseEGTB=on
EGTB path="J:\Tbs"
UseBookWeights=on
UseBookLearning=on
BookLearningMode=KILL
VarietyOfPlay=3
BookLearningBound=260
UseLearning=on
UseLearningData=on
LogLearningIntoFile=on
Pondering=on

I found it in my test-version of Smarthink 1.0 too )

SmarThink v1.0 [r426/20.12.2005]
Error opening "Book.smt"!
Transposition size is 16777216 entries (402653184 bytes)
Pawn hash size is 8388608 entries (268435456 bytes)
EGTBHashSize=1048576 bytes
Aggressivity=100 (36)%
Cowardice=100 (36)%
Tropism=100 (144)%
ResignValue=999
ResignTimes=9
UseEGTB=on
EGTB path="F:\Tbs"
UseBookWeights=on
UseBookLearning=on
BookLearningMode=KILL
VarietyOfPlay=3
BookLearningBound=260
UseLearning=on
UseLearningData=on
LogLearningIntoFile=on
Pondering=on