LMR - [for starters] - [Advance] and [Expert]

[Rebel]

Two LMR versions (for starters) and one Advanced, my current system. I am interested in feedback naturally but most in similar LMR pseudo code from others. Which with permission I can put on my web site. Could be a good learning experience.

http://rebel13.nl/rebel13/blog/lmr%20advanced.html

[Ferdy]

Two LMR versions (for starters) and one Advanced, my current system. I am interested in feedback naturally but most in similar LMR pseudo code from others. Which with permission I can put on my web site. Could be a good learning experience.

http://rebel13.nl/rebel13/blog/lmr%20advanced.html

Interesting.

Code: Select all

a=depth; b=movecount; if (b>63) b=63;

Is the movecount the moves that were already made or the sequence of the sorted move list?

In

Code: Select all

 if (score > ALPHA+margin) R--; 

What composes the score?

[Rebel]

Two LMR versions (for starters) and one Advanced, my current system. I am interested in feedback naturally but most in similar LMR pseudo code from others. Which with permission I can put on my web site. Could be a good learning experience.

http://rebel13.nl/rebel13/blog/lmr%20advanced.html

Interesting.

Code: Select all

a=depth; b=movecount; if (b>63) b=63;

Is the movecount the moves that were already made or the sequence of the sorted move list?

The first.

In

Code: Select all

 if (score > ALPHA+margin) R--; 

What composes the score?

Evaluation score, changed the page, that was indeed a bit unclear.

[PK]

have You tried dividing history by eight at the beginning of a new search and zeroing it only when the game starts?

[cdani]

Thanks! I can add for example that late moves in endgames can be reduced a bit more, at least for Andscacs.

[Rebel]

have You tried dividing history by eight at the beginning of a new search and zeroing it only when the game starts?

Makes sense, will try.

[Rebel]

Thanks! I can add for example that late moves in endgames can be reduced a bit more, at least for Andscacs.

I know what you mean, create a second more tight "lmr_tab" for the endgame due to the decreasing number of legal moves. Or something like that.

[zenpawn]

In your pseudocode, does "move produces a better score" mean a beta cutoff, i.e, score >= beta (that's where I currently increment my history table) or score > alpha? Thanks.

[brtzsnr]

Thank you for the ideas, Ed. By comparison my LMR logic is very primitive. All my attempts to improve further have failed, but I see a lot of good ideas in your post.

Code: Select all

                // Late move reduction: search best moves with full depth, reduce remaining moves.
                lmr := int32(0)
                if !sideIsChecked && depth > lmrDepthLimit && !critical {
                        // Reduce quiet moves and bad captures more at high depths and after many quiet moves.
                        // Large numMoves means it's likely not a CUT node.  Large depth means reductions are less risky.
                        if move.IsQuiet() {
                                lmr = 2 + min(depth, numMoves)/6
                        } else if seeSign(pos, move) {
                                lmr = 1 + min(depth, numMoves)/6
                        }
                }

[Rebel]

In your pseudocode, does "move produces a better score" mean a beta cutoff, i.e, score >= beta (that's where I currently increment my history table) or score > alpha? Thanks.

Score > alpha works a little bit better for me.