Progress On Bricabrac

[Mike Sherwin]

I was finally able to do a little work on Bricabrac today. A few months ago I lost a few hundred elo in my own ability to play chess. I blame the jab. I just simply don't see what I used to be able to see. And my speed chess ability is in the crapper. Today, I replaced Bric's eval from scratch with this simple one that is after the style of RomiChess.

Code: Select all

for (sq = A1; sq <= H8; sq++) {

    col = sq & 7;
    wrow = sq >> 3;
    brow = 7 - wrow;
    cenCol = col < 4 ? col : 7 - col;
    cenRow = wrow < 4 ? wrow : 7 - wrow;

    // white pawns
    if (sq > H1 && sq < A8) {
      if (wrow == 1) wpPST[sq] = 12;
      else wpPST[sq] = cenCol * cenCol + wrow * wrow;
      wpPST[sq] += ply * wrow / 32;
    }


    // black pawns
    if (sq < A8 && sq > H1) {
      if (brow == 1) bpPST[sq] = 12;
      else bpPST[sq] = cenCol * cenCol + brow * brow;
      bpPST[sq] += ply * brow / 32;
    }

    // knights
    moves = knightMoves[sq] & ~allPieces;

    // white knights
    captures = knightMoves[sq] & (bRooks | bQueens | king[BLACK]);
    wnPST[sq] = __popcnt64(moves) * 4 + __popcnt64(captures) * 16;
    wnPST[sq] += (cenRow + cenRow) * (cenCol + cenCol);

    // black knights
    captures = knightMoves[sq] & (wRooks | wQueens | king[WHITE]);
    bnPST[sq] = __popcnt64(moves) + __popcnt64(captures) * 16;
    bnPST[sq] += (cenRow + cenRow) * (cenCol + cenCol);

    // bishops
    blocks.bb = allPieces & b7e[sq];
    blocks.bb = ((blocks.l32 >> 8) | blocks.h32);
    moves = bss[sq][blocks.r1][0]
          & bss[sq][blocks.r2][1]
          & bss[sq][blocks.r3][2];

    // white bishops
    captures = moves & (bRooks | bQueens | king[BLACK]);
    moves ^= captures;
    wbPST[sq] = __popcnt64(moves) * 4 + __popcnt64(captures) * 16;
    wbPST[sq] += (cenRow * cenCol) + wrow;

    // black bishops
    captures = moves & (wRooks | wQueens | king[WHITE]);
    moves ^= captures;
    bbPST[sq] = __popcnt64(moves) * 4 + __popcnt64(captures) * 16;
    bbPST[sq] += (cenRow * cenCol) + brow;

    // rooks
    blocks.bb = allPieces & rob[sq];
    moves = qss[sq][(blocks.bb >> (sq & 56)) & 127][0]
      & qss[sq][blocks.r2][1]
      & qss[sq][blocks.r3][2]
      & qss[sq][blocks.r4][3]
      & qss[sq][blocks.r5][4]
      & qss[sq][blocks.r6][5]
      & qss[sq][blocks.r7][6]
      & rob[sq];

    // white rooks
    captures = moves & (bQueens | king[BLACK]);
    moves ^= captures;
    wrPST[sq] = __popcnt64(moves) * 2 + __popcnt64(captures) * 16;
    wrPST[sq] += cenRow + wrow == 6 * 20;

    // black rooks
    captures = moves & (wQueens | king[WHITE]);
    moves ^= captures;
    brPST[sq] = __popcnt64(moves) * 2 + __popcnt64(captures) * 16;
    brPST[sq] += cenRow + brow == 6 * 20;

    // queens
    blocks.bb = allPieces & rob[sq];
    moves = qss[sq][(blocks.bb >> (sq & 56)) & 127][0]
      & qss[sq][blocks.r2][1]
      & qss[sq][blocks.r3][2]
      & qss[sq][blocks.r4][3]
      & qss[sq][blocks.r5][4]
      & qss[sq][blocks.r6][5]
      & qss[sq][blocks.r7][6];

    // white queens
    captures = moves & king[BLACK];
    moves ^= captures;
    wqPST[sq] = __popcnt64(moves) + __popcnt64(captures) * 16;
//    wqPST[sq] += (cenRow * cenCol) + wrow;

    //black queens
    captures = moves & king[WHITE];
    moves ^= captures;
    bqPST[sq] = __popcnt64(moves) + __popcnt64(captures) * 16;
//    bqPST[sq] += (cenRow * cenCol) + brow;

    // white king
    sum = (2200 - mat[BLACK]) >> 6;
    wkPST[sq] = (cenCol + cenRow) * sum;

    // black king
    sum = (2200 - mat[WHITE]) >> 6;
    bkPST[sq] = (cenCol + cenRow) * sum;
  }

  // white pawn specifics
  wpPST[E2] -= (board[E1] == WC) * 20;
  wpPST[E4] += (board[E1] == WC) * 20;

  // black pawn specifics
  bpPST[E7] -= (board[E8] == BC) * 20;
  bpPST[E5] += (board[E8] == BC) * 20;

  // white king specifics
  wkPST[G1] += (board[E1] == WC) * 40;

  // black king specifics
  bkPST[G8] += (board[E8] == BC) * 40;

Then I played a number of 2 min + 6 sec inc games against Bric and lost every game.

Here is my best game.

[pgn][Event "Computer chess game"]
[Site "DESKTOP-HFVHK2B"]
[Date "2021.10.23"]
[Round "?"]
[White "mjshe"]
[Black "Bricabrac"]
[Result "0-1"]
[BlackElo "2200"]
[ECO "D07"]
[Opening "QGD"]
[Time "23:48:18"]
[Variation "Chigorin, 3.Nf3"]
[WhiteElo "2400"]
[TimeControl "120+6"]
[Termination "normal"]
[PlyCount "96"]
[WhiteType "human"]
[BlackType "program"]

1. d4 d5 {(d7d5 e2e3 c8d7 f1d3 e7e6 c1d2 g8f6 d1f3 b8c6) -0.35/9 29} 2. c4
Nc6 {(b8c6 e2e3 e7e5 d4e5 f8b4 c1d2 b4d2 d1d2) +0.06/8 23} 3. Nf3 e6 {(e7e6
e2e3 g8f6 f1d3 d5c4 d3c4 f8b4 e1f1) -0.01/8 19} 4. e3 Nf6 {(g8f6 c4c5 f6e4
f1b5 e8e7 b1d2 e7f6 d2e4) -0.16/8 16} 5. Nc3 Bb4 {(f8b4 f1d3 f6e4 c1d2 e4d2
e1d2 d5c4) -0.22/7 13} 6. Bd2 Bxc3 {(b4c3 d2c3 f6e4 d1d3 e4c3 d3c3 e8g8)
-0.26/7 11} 7. Bxc3 Ne4 {(f6e4 a1c1 e4c3 c1c3 c6b4 d1a4 b4c6) -0.46/7 10}
8. Qc2 Nxc3 {(e4c3 c2c3 g7g5 c4d5 e6d5 f1b5 d8f6) -0.32/7 9} 9. Qxc3 g5
{(g7g5 h2h3 d8f6 c4c5 e8g8 f1d3 g8g7) -0.53/7 8} 10. cxd5 exd5 {(e6d5 h2h3
c8f5 f1b5 f5d7 b5d3 e8g8) -0.55/7 7} 11. Bb5 Bd7 {(c8d7 c3c5 a7a6 b5d3 g5g4
f3d2 c6e7) -0.52/7 7} 12. Bxc6 Bxc6 {(d7c6 f3e5 c6b5 a2a4 b5d7 c3b4 b7b6)
-0.19/7 7} 13. Ne5 Bb5 {(c6b5 c3b4 c7c6 a2a4 a7a5 b4c3 b5a6) -0.24/7 6} 14.
a4 Ba6 {(b5a6 c3c5 d8d6 a1c1 d6c5 c1c5 c7c6) -0.36/7 6} 15. b4 c6 {(c7c6
e5c6 d8c8 b4b5 b7c6 b5a6 c8a6) -0.62/7 6} 16. Rb1 Rc8 {(a8c8 e5d3 a6c4 d3e5
b7b5 a4a5 e8g8) -0.16/7 6} 17. Qb3 f6 {(f7f6 e5d3 a6c4 b3c2 e8g8 e3e4 f8f7)
+0.13/7 6} 18. Nd3 Bc4 {(a6c4 b3d1 c8c7 d3b2 e8g8 b2c4 d5c4) -0.01/7 6} 19.
Qa3 Qe7 {(d8e7 e1g1 a7a5 f1d1 c8c7 g1f1 a5b4) +0.14/7 6} 20. O-O a5 {(a7a5
f1d1 f6f5 d3e5 c4e2 d1c1 a5b4) +0.08/7 6} 21. Rfd1 Rg8 {(h8g8 a3c3 a5b4
b1b4 g5g4 a4a5 f6f5) +0.05/7 6} 22. Ne1 Be2 {(c4e2 d1d2 e2g4 e1f3 g4f5 b1b2
a5b4) +0.48/7 6} 23. Rd2 Bg4 {(e2g4 e1d3 g4f5 a3c3 a5b4 b1b4 e8f7) +0.26/7
6} 24. Qc3 Bf5 {(g4f5 e1d3 a5b4 b1b4 e8f7 a4a5 f7e6) +0.13/7 6} 25. Rbb2
axb4 {(a5b4 c3b4 g8g7 e1d3 e7b4 b2b4 e8d8) +0.21/7 6} 26. Qxb4 Qxb4 {(e7b4
b2b4 g8g7 e1d3 c8a8 g1f1 g7e7 f1e2) +0.33/8 6} 27. Rxb4 Rg7 {(g8g7 e1d3
c8c7 a4a5 g5g4 g1f1 c7e7 f1e2) +0.31/8 6} 28. Rdb2 Ra8 {(c8a8 e1c2 a8a5
g2g4 f5e4 c2e1 a5a7 e1c2) +0.45/8 6} 29. Rxb7 Rxb7 {(g7b7 b2b7 a8a4 b7b3
g5g4 g1f1 a4a1 f1e2) +0.63/8 6} 30. Rxb7 Rxa4 {(a8a4 b7b8 e8d7 b8b7 d7d6
g1f1 g5g4 f1g1) +0.59/8 6} 31. f3 h5 {(h7h5 b7b2 h5h4 b2b7 e8d8 g1f1 d8e8
f1g1) +0.64/8 5} 32. Rb2 Ra3 {(a4a3 g1f2 h5h4 f2e2 a3c3 b2d2 h4h3 g2h3)
+0.59/8 6} 33. Kf2 Kd8 {(e8d8 b2b8 d8d7 f3f4 g5g4 f2e2 d7c7 b8g8) +0.63/8
5} 34. h3 h4 {(h5h4 f3f4 g5f4 e3f4 a3a7 e1f3 a7h7 f2e1) +0.46/8 5} 35. Nc2
Ra5 {(a3a5 f2e2 a5a7 c2b4 d8c7 b4d3 c7c8 d3c5 c8c7) +0.34/9 5} 36. Ke2 Ke8
{(d8e8 b2b8 e8d7 c2b4 a5a1 e2d2 d7c7 b8e8) +0.50/8 6} 37. Kd2 Kd7 {(e8d7
d2e2 d7c8 c2e1 f5g6 e2d2 a5a3 d2e2) +0.49/8 5} 38. Ne1 Ra3 {(a5a3 f3f4 g5f4
e3f4 d7d6 e1f3 f5h3 f3h4 h3e6) +0.37/9 5} 39. Ke2 Kd6 {(d7d6 e1c2 a3d3 b2b7
d6e6 e2f1 d3d1 c2e1) +0.54/8 6} 40. Kd2 Kd7 {(d6d7 e1c2 a3d3 d2e2 d7d6 c2b4
d3a3 b4c2) +0.58/8 5} 41. Ke2 Kd6 {(d7d6 e1c2 a3d3 b2b7 d6e6 e2f1 d3d1
c2e1) +0.54/8 5} 42. Kd2 Ke6 {(d6e6 d2e2 e6e7 e1c2 a3d3 b2b8 e7d6 c2b4
d6c7) +0.40/9 6} 43. Ke2 Ke7 {(e6e7 e1c2 a3c3 e2d1 c6c5 d4c5 c3c5 c2d4
f5d3) +0.30/9 5} 44. Rd2 Bb1 {(f5b1 e1d3 b1a2 d3c5 a2c4 e2f2 e7d6 d2c2
d6c7) +0.35/9 5} 45. Nd3 Ba2 {(b1a2 d2b2 a2c4 b2d2 f6f5 f3f4 g5f4 e3f4
c4d3) +0.16/9 6} 46. Nc5 Bc4+ {(a2c4 e2f2 f6f5 e3e4 f5f4 e4e5 e7f7 e5e6
f7f6) +0.26/9 5} 47. Kf2 Ra1 {(a3a1 e3e4 a1f1 f2e3 f1e1 e3f2 e1f1 f2e3
d5e4) +0.30/9 5} 48. Rb2 Rf1# {(a1f1 e3e4) +M1/9 6} 0-1
[/pgn]
At least I did not loose on time!

I hope I can now start making progress again on Bricabrac.

[tcusr]

hi mike, do you host your code somewhere?

[mvanthoor]

I was finally able to do a little work on Bricabrac today.
...
I hope I can now start making progress again on Bricabrac.

Congratulations on the progress I look forward to seeing the engine in one of the rating lists

[Mike Sherwin]

hi mike, do you host your code somewhere?

cpp file
https://www.mediafire.com/file/5b00xvnx ... c.cpp/file

MSVS 2019 full folder including the exe
https://www.mediafire.com/file/02gda4iu ... ac.7z/file

[Mike Sherwin]

I was finally able to do a little work on Bricabrac today.
...
I hope I can now start making progress again on Bricabrac.

Congratulations on the progress I look forward to seeing the engine in one of the rating lists

Thanks! I seem to have found a little gem of an engine, Sam, that only has static PST's except the kings PSTs are adjusted by game phase. And it is performing at TSCP 1.81 + 252 elo. And it's PST's are created algorithmically like I did above. Except mine is TSCP 1.81 - 330 elo. Now I have to find out why Sam's PSTs are as good or better than PESTO?
? ?

Code: Select all

// A small chess program by Jonathan Kreuzer
#include <stdio.h>
#include <math.h>
#include <stdlib.h>
#include <time.h>

const int HASH_SIZE = 400000;
const int INVALID=32, IV=32, EMPTY=0, WHITE=8, BLACK=16, WH=8, BL=16, PAWN=0, KNIG=1, BISH=2, ROOK=3, QUEE=4, KING=5;
const int TIME = 300000;
const int B_QS = 4, B_KS = 8, W_QS = 1, W_KS = 2;
const int N_dirs[8] = {-21, -19, -12, -8, 8, 12, 19, 21};
const int K_dirs[8] = {1, 9, 10, 11, -1, -9, -10, -11};
const int Q_dirs[8] = {1, -1, 9, -9, -10, 10, -11, 11};
const int R_dirs[4] = {1, -1,  -10, 10};
const int B_dirs[4] = {9, -9, -11, 11};
const int P_dirs[8] = {-10, -20, -9, -11, 10, 20, 9, 11};
int inline SRC   ( int Move ) {return (Move&127);}
int inline DST   ( int Move ) {return ((Move>>7)&127);}
int inline PROMO ( int Move ) {return ((Move>>14)&3);}
int inline VALUE ( int Move ) {return ((Move>>16)&16383);}
int inline SWITCH ( int Color ){return Color^(WHITE|BLACK);}
int inline File ( int Sq ) {return (Sq-20)%10;}
int inline Rank ( int Sq ) {return (Sq-10)/10;}
int inline SQ ( int file, int rank) {return 21+file+rank*10;}
int EvalSq [ 26 * 128 ];
int SetCastle[120];
const unsigned char StartBoard[120] = { IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV,
        IV, ROOK|BL, KNIG|BL, BISH|BL, QUEE|BL, KING|BL, BISH|BL, KNIG|BL, ROOK|BL, IV,
        IV, PAWN|BL, PAWN|BL, PAWN|BL, PAWN|BL, PAWN|BL, PAWN|BL, PAWN|BL, PAWN|BL, IV,
        IV, 0, 0, 0, 0, 0, 0, 0, 0, IV, IV, 0, 0, 0, 0, 0, 0, 0, 0, IV, IV, 0, 0, 0, 0, 0, 0, 0, 0, IV, IV, 0, 0, 0, 0, 0, 0, 0, 0, IV,
        IV, PAWN|WH, PAWN|WH, PAWN|WH, PAWN|WH, PAWN|WH, PAWN|WH, PAWN|WH, PAWN|WH, IV,
        IV, ROOK|WH, KNIG|WH, BISH|WH, QUEE|WH, KING|WH, BISH|WH, KNIG|WH, ROOK|WH, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV, IV };
const int PieceValues[8]    = { 100, 328, 330, 522, 952, 5000 };
const int MPieceValues[8]   = {  00, 328, 330, 522, 952,   00 };
const int KingEval[10] = { 0, 8, 12, 5, 0, 0, 5, 14, 9, 0};
const int CentEval[10] = { 0,-6, -3, -1, 0, 0, -1, -3, -6, 0};
const int Cent[10]         = { 0, 1, 2, 2, 3, 3, 2, 1, 1, 0};
void SendThinking ( int move, int Depth, int Eval, __int64 Nodes, int bFinished );

struct SBoard 
        {
        unsigned char m_Sqs[120];
        int stm, Eval, EPsq, Castling, WKsq, BKsq, WMat, BMat, nMoves, nLastCapOrPawn;

        void Init()     {
                int sq, pc;
                for (sq = 0; sq < 120; sq++) m_Sqs[sq] = StartBoard [ sq ];
                Eval = 0; nMoves = 0;
                Castling = 0;
                stm = WHITE;
                WKsq = 95; BKsq = 25;
                BMat = WMat = 2*PieceValues[BISH]+2*PieceValues[KNIG]+2*PieceValues[ROOK]+PieceValues[QUEE];
                for (sq = 0; sq < 120; sq++) SetCastle[sq] = 0;
                SetCastle[21] = B_QS; SetCastle[28] = B_KS; SetCastle[25] = B_QS|B_KS; SetCastle[91] = W_QS; SetCastle[98] = W_KS; SetCastle[95] = W_QS|W_KS;
                // Evaluation is incremental and uses the piece square table in EvalSq arrays.
                for (pc = 0; pc < 8; pc++)
                        for (sq = 0; sq < 120; sq++) {
                                EvalSq[ ((pc|WHITE)<<7)+sq ] = PieceValues[pc];
                                EvalSq[ ((pc|BLACK)<<7)+sq ] = -PieceValues[pc];
                                if (pc == PAWN) 
                                        {EvalSq[ ((pc|WHITE)<<7)+sq ] += (9-Rank(sq)) * Cent[File(sq)];
                                         EvalSq[ ((pc|BLACK)<<7)+sq ] -=   Rank(sq)   * Cent[File(sq)];}
                                else if (pc == KING) {}
                                else{if (pc!=ROOK && Rank(sq)==8) EvalSq[ ((pc|WHITE)<<7)+sq ] -= 8;
                                         if (pc!=ROOK && Rank(sq)==1) EvalSq[ ((pc|BLACK)<<7)+sq ] += 8;
                                         EvalSq[ ((pc|WHITE)<<7)+sq ] += CentEval[ File(sq) ];
                                         EvalSq[ ((pc|BLACK)<<7)+sq ] -= CentEval[ File(sq) ];
                                        }
                        EvalSq[ (0<<7)+sq ] = (Rank(sq)-9)*2 + KingEval[File(sq)]; // King Values depend on stage
                        EvalSq[ (1<<7)+sq ] = (Rank(sq))*2 - KingEval[File(sq)];
                        EvalSq[ (2<<7)+sq ] = 2*CentEval[ File(sq) ];
                        EvalSq[ (3<<7)+sq ] = -2*CentEval[ File(sq) ];
                        }
                }

[Mike Sherwin]

Bric got all the perft test, at the CPW, position 2 to position 5 correct to the maximum depths given.

[mvanthoor]

Bric got all the perft test, at the CPW, position 2 to position 5 correct to the maximum depths given.

Congratulations Maybe I should write an update as well, after about 5 months...

[Mike Sherwin]

After my last post I've been under the weather. About two weeks ago I came down with shingles. That was a painful two weeks. Still is but at least I was able to start on the hash tables today. For starters it is only 1 million entries and the only thing added so far is beta cutoffs. No move ordering yet is done. Anyway in the original position to finish 15 ply before the HT it took 26.54 seconds. After the HT was added it took only 14.52 seconds. Also I'm experimenting with 128 bits of signature. Still it managed 10,167,093 n/s on a R9 3950x @ 4.2 GHz on a single thread. A short day but a productive day!

[Mike Sherwin]

Of course using PeSTO Bric was able to defeat TSCP easily even when Bric had many bugs. Today however Bric finally won a match against TSCP with all original code.

Bricabrac - Tscp181 : 52.5/100 49-44-7 (111011101111001010011001101100=01101111010=1110011010=1110=1010010010001101=1=0001100001101001000=10) 53% +21

It won because of search improvements. The evaluation is lucky to be 1400 elo.

[Mike Sherwin]

I was able to do a little bit of work yesterday. My immediate goal since PeSTO pst was replaced by algorithmically created pst was to surpass PeSTO. I'm almost there!

BricPESTO1 - Tscp181 : 59.0/100 49-31-20 (101=011=10011000=011=0001=01=110=111=1111100011101011==110=1==111001100100==0==111100011=01=110111=1) 59% +63

Bricabrac - Tscp181 : 57.5/100 47-32-21 (11=110100=10101011=1=1===011000=11=11000101110101=00101=01001111100=0=11101=00=0111=1=0==110101=11=1) 58% +56

Just to be clear, all Bric is doing now for an eval is algorithmically filling the pst before each search. While PeSTO is static and does not change.