Author: rasjid chan
Date: 11:38:50 01/24/04
Go up one level in this thread
On January 24, 2004 at 12:38:07, milix wrote:
Lets say you implemented Ed's idea of long-check correctly.
It does not mean it will work if the other part of your program
is not well done. A little extra good idea done well don't
give a huge boost to the program, or better in solving test positions.
if (score >= beta) return score / beta mean different
thing to different implementations. If you have a hashtable
and hash upper-bound with value < alpha or lower-bound
with value > beta, it may back-fire very badly if done wrongly.
This is outside your question on long-checks but if this is your program
bug, no long-check can save it.
Rasjid
>Ok, I tried the ideas posted here but no success.
>My new code goes like this:
>
>quiescent(alpha, beta, mate_score, checks_depth)
>{
> score = full_evaluate()
> if (score >= beta) return score // or beta, same effect
> if (score > alpha) alpha = score
>
> if (side_to_move_is_in_check)
> generate_legal_moves()
> if (no_legal_moves) return -mate_score
> else if (only_one_legal_move) checks_depth = checks_depth + 3
> else if (only_two_legal_moves) checks_depth = checks_depth + 2
> else if (checks_depth > 0)
> generate_good_equal_captures_queenpromotions_and_checks()
> else generate_good_equal_captures_queenpromotions()
> sort_the_moves()
>
> if (checks_depth < 0) checks_depth = 0
> if (checks_depth > 3) checks_depth = 3
>
> for each move in the list {
> make_the_move()
> score = -quiescent(-beta, -alpha, mate_score-1, checks_depth-1)
> undo_the_move()
> if (score >= beta) return score // or beta, same effect
> if (score > alpha) alpha = score;
> }
>
> return alpha
>}
>
>Tried to
> if (checks_depth <= 0 and score >= beta) return score
>but QS exploded.
>
>Tried Ed's suggestion to call QS with alpha=LOWVALUE (=MATE_SCORE-SOME_MARGINE)
>but failed also.
>
>My main search is a PVS Alpha-Beta and for this test i search only in depth 1. I
>use no extensions, no TT (only a repetion-check wich works ok). My move-list
>sorting algorithm is a SEE which puts checks on the top (for this test).
>I call my main search from root search as
>
>alpha_beta(1, -INFINITY, +INFINITY, MATE_SCORE)
>
>For the test position
>
>FEN "2k4r/ppp5/2p1b1Q1/4q3/3bP3/3B2P1/PPP1N2P/5R1K b - - 0 1"
>
>I got (briefly):
>
>.......
> After QS Score for Bd4xb2 is -1.05
>Trying move Rxh2
>GOING TO QS WITH Rxh2: SIDE TO MOVE IS IN CHECK
>-----------------------------------------------
> QS called with args (-3000000, 1046, 299999, 3)
> Generated moves for White: Kxh2
> QS move played Kh1xh2
> QS called with args (-1046, -1034, 299998, 2)
> Generated moves for Black: Qh8 Bg1 Bxa2 Qxg3 Qh5
> QS move played Qe5-h8
> QS called with args (1034, 1046, 299997, 1)
> Generated moves for White: Qh5 Qh7 Qh6 Kg2
> QS move played Qg6-h5
> QS called with args (-1046, -6263, 299996, 0)
> score=6263
> score=-1046
> QS move played Bd4-g1
>......
>and missed the mate.
>
>When I icreased checks_depth by 4 (or 3 points) it again missed the mate.
>
>Finally, it suggests Bd4xb2 in all cases....
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.