fen where stockfish has trouble finding zh mate

[isaacl]

setoption name UCI_Variant value crazyhouse
position fen r2q1rk1/pp2bb1p/2pp2pK/7N/3pPR1P/1B1P2p1/PPPq4/R7[PPBNNN] b - - 49 25

At low depths and move times it evals the position as completely lost with f7b3, at higher depth it finds an #8 with Bxh4.

[ddugovic]

Good observation & I'll try that instead. (Also there's probably a more efficient way to check if the king attacks any not-friendly-occupied squares which are not attacked!)

Also I could probably define GenType FLIGHTS for the most readable possible solution:

#ifdef CRAZYHOUSE
template<>
ExtMove* generate<FLIGHTS>(const Position& pos, ExtMove* moveList) {

  Color us = pos.side_to_move();
  Square ksq = pos.square<KING>(us);

  // Generate flights for king
  ExtMove* cur = moveList;
  Bitboard b = pos.attacks_from<KING>(ksq) & ~pos.pieces(us);
  while (b)
      *moveList++ = make_move(ksq, pop_lsb(&b));

  if (!pos.checkers() && pos.can_castle(us))
  {
      if (us == WHITE)
      {
          moveList = generate_castling<MakeCastling<WHITE,  KING_SIDE>::right, false, false>(pos, moveList, WHITE);
          moveList = generate_castling<MakeCastling<WHITE, QUEEN_SIDE>::right, false, false>(pos, moveList, WHITE);
      }
      else
      {
          moveList = generate_castling<MakeCastling<BLACK,  KING_SIDE>::right, false, false>(pos, moveList, BLACK);
          moveList = generate_castling<MakeCastling<BLACK, QUEEN_SIDE>::right, false, false>(pos, moveList, BLACK);
      }
  }

  while (cur != moveList)
      if (!pos.legal(*cur))
          *cur = (--moveList)->move;
      else
          ++cur;

  return moveList;
}
#endif

[ddugovic]

In case this (admittedly delayed on my part as I resolve other issues) test fails, plan B is based on a @Vinvin20 suggestion:

May be do null moves only when there's less than 3 pieces in (sum of both) hands. May be add a condition to be sure kings are safe.

I think it would be more appropriate to check if the opponent has ~less than 3~ ~less than 2~ no pieces in hand, since the purpose of a null move verification heuristic is to efficiently prove that a null move does not sharply change the evaluation.

Preliminary tests show that the change ~doesn't excel at bullet~ performs OK once pieceCountInHand[WHITE][ALL_PIECES] and pieceCountInHand[BLACK][ALL_PIECES] are set.

[Vinvin20]

Nice to see there's progress here ! SF will be even stronger in the opening with null-moves !

[ddugovic]

Well... with #189 merged, this seems less necessary. I'm going to try changing the rule from "no pieces in hand" to "under (depth - 4) * 8 pieces in hand" (EDIT: that didn't work well).

Bullet preliminary results (more testing required) of "no pieces in hand":

Finished game 250 (Stockfish 311216 64 BMI2 vs PATCH): 1/2-1/2 {Draw by 3-fold repetition}
Score of PATCH vs Stockfish 311216 64 BMI2: 130 - 112 - 8  [0.536] 250