Update Best Score and Best Move

[tomitank]

Hi all!
I write from Hungary, therefore sorry for my english.
I'd like to know, what's the best way to update in PVS the best score and best move?

Sample 1.

Code: Select all

			if (Score > BestScore) {
				if (Score > alpha) {
					if (Score >= beta) {
						return Score;
					}
					alpha = Score;
				}
				BestScore = Score;
				BestMove = move;
			}

Sample 2.

Code: Select all

			if (Score > BestScore) {
				BestScore = Score;
				BestMove = move;
				if (Score > alpha) {
					if (Score >= beta) {
						return Score;
					}
					alpha = Score;
				}
			}

Sample 3.

Code: Select all

			if (Score > BestScore) {
				BestScore = Score;
				BestMove = move;
				if (Score > alpha) {
					alpha = Score;
					if (Score >= beta) {
						return Score;
					}
				}
			}

Or another idea?

Thanks, Tamas

[hgm]

Sample 1 seems to do the least superfluous work. The difference could be immeasurably small, though.

For the benefit of multi-PV I often use this:

Code: Select all

if(score > bestScore) bestScore = score, bestMove = move;
if(score > alpha) {
  if(score >= beta) return score;
  alpha = bestScore;
  UpdatePV(move);
  if(ply == ROOT) {
    PrintPV();
    alpha -= multiPvMargin;
  }
}

This tests for score > alpha on every move, while normally (i.e. in every node except in the root when running multi-PV) this should not be possible without increasing bestScore. (Which only very few moves do.)

[Sven]

Or another idea?

Sample 4

Code: Select all

if (score >= beta) {
    return score;
}
if (score > bestScore) {
    bestScore = score;
    if (score > alpha) {
        UpdatePV(move);
        // bestMove is first entry of current PV
        // updating alpha is not really necessary, simply pass something like Max(alpha, bestScore) to the recursive call
        if (ply == 0) {
            // root
            PrintPV();
        }
    }
}

[mar]

All fail soft variants are functionally equivalent, so I'm not sure what you're asking about.

Naturally you may want to exit as soon as possible, however you'll most likely do some extra work later, like

- extracting PV in PV nodes if score > alpha
- updating killers/history/TT on beta cutoff

[tomitank]

Hi!
Currently i extract PV (and the UCI best move) from the Hash.
I tried many more variation, but [Sample 1.] work for me better.

I tried this avoiding the hash bug, but it's not so good for me:

Code: Select all

         if (Score > BestScore) {
            if (Score > alpha) {
               if (Score >= beta) {
                  return Score;
               }
               alpha = Score;
               if (ply == 0) UCIBestMove = move; // best root move
            }
            BestScore = Score;
            BestMove = move;
         } 

I do not know why?
I do something wrong?

With Hash Update (currently version):

Code: Select all

		moves loop {
            .......
			if (Score > BestScore) {
				if (Score > alpha) {
					if (Score >= beta) {
						Update Killer, History

						StoreHashMove(currentMove, Score, FLAG_BETA, Depth);

						return Score;
					}
					alpha = Score;
				}
				BestScore = Score;
				currBestMove = currentMove;
			}
		}

		if (alpha != OldAlpha) {
			StoreHashMove(currBestMove, BestScore, FLAG_EXACT, Depth);
		} else {
			StoreHashMove(currBestMove, BestScore, FLAG_ALPHA, Depth);
		}

Thanks, Tamas

[hgm]

I usually assign a hash slot and store the entry at the end of the node (just before Search() returns). Nothing should be probing for it in the mean time anyway. Only in the root I store after every iteration. Because in analysis mode the root never returns normally; it is always aborted, returning without hash store.

[AlvaroBegue]

What's the point of keeping both BestScore and alpha? In my code I only have alpha and this doesn't seem to be a problem.

Code: Select all

    if (score > alpha) {
      best_move = move;
      alpha = score;
      if (alpha >= beta)
        break;
    }

[hgm]

It is the difference between fail hard and fail soft. In fail soft you return bestScore, which can stay below alpha.

[AlvaroBegue]

It is the difference between fail hard and fail soft. In fail soft you return bestScore, which can stay below alpha.

Oh, that makes sense. Thanks.

[Sven]

Hi!
Currently i extract PV (and the UCI best move) from the Hash.
I tried many more variation, but [Sample 1.] work for me better.

I tried this avoiding the hash bug, but it's not so good for me:

Code: Select all

         if (Score > BestScore) {
            if (Score > alpha) {
               if (Score >= beta) {
                  return Score;
               }
               alpha = Score;
               if (ply == 0) UCIBestMove = move; // best root move
            }
            BestScore = Score;
            BestMove = move;
         }

I do not know why?
I do something wrong?

Updating the best move should only happen if the new score is inside the window, i.e. alpha < Score < beta. When failing low or high there is no best move.