I do not use history information to select moves for reduction. Perhaps a few more words on what I do with LMR are appropriate:bob wrote:also the general consensus has been that the "history information" (I assume you are doing that) is not particularly effective, and is almost random noise thrown in to the equation.
There are other interesting aspects of "LMR" that can be tried as well, besides what is being done today by most everyone.
First: I sort the moves as usual. The first few mores (the number of which is specified by lmrmoves) will never be reduced. Nor will any move be reduced when this is not a null-window search (if lmrpv is off). When tactical patterns are recognized, LMR is not activated either. Checking moves will in principle not be reduced, unless it is a really bad move by SEE score. Queen and rook sacs are in this category.
LMR is not done during the endgame phase either. Finally: If the quiescent score of the position after the move is above beta, the move is not reduced either.
So I'm very restrictive about reducing moves, basically to try to reduce tactical blindness. The Baron is not a deep searcher, so it cannot afford to overlook shallow tactics.
There is another thing that may be unique for the Baron: The reduced search is done with a slightly offset window (lmrmargin specifies the offset). The default lmrmargin=5 means that alpha is lowered by 5 centipawns. The intention of this is to only reduce those moves that really are nonsense. If the score of a reduced search is close to alpha, it should probably not be reduced as it is a valid alternative. Experiments with settings for this parameter are probably the most interesting of the lmr parameters.
Richard.
