I am not sure I quite understand what you mean here, but to me a side to move bonus is one of the most unquestionably "correct" evaluation terms in existence, apart from material. In almost all chess positions, having the move is an advantage.bob wrote:I've never had a side to move bonus, except for pawn endings where the side to move determines whether a pawn is catchable, or whether a pawn is unstoppable due to opposition. But for normal positions, I don't do it because I know of no reliable way to do so. And with the depths we see today, and the extensions/reductions in use, it ends up being apples-to-oranges anyway...
How big the side to move bonus should be is a difficult question, however. Should it be a constant, or should it depend on other components of the evaluation? I currently use a constant value of 0.2 in the middle game and 0.08 in the endgame, but these numbers are completely untuned (the only thing I have verified is that these numbers perform better than no side to move bonus at all). Interesting ideas to play with would be to consider the king safety and/or passed pawn scores (if both sides have a strong king attack, having the move is usually more important, and similarly for positions where both sides have high passed pawn scores), or to let the side to move bonus depend on how open the position is.
Perhaps it would also be possible to determine good values of the side to move bonus experimentally: Start with a large set of quiescent positions, and search each position to depth N, first with white to move, then with black to move. Half of the median of the differences between the scores when white moves and when black moves would be a natural value to use for the side to move bonus.
It's trivial in all programs. The only non-trivial thing is to find the time to test it, when there are always countless other things waiting to be tried.In Crafty it is trivial.True: For time managment reasons, one ply is probably a better increment than two plies. But I don't see any a priori reason to believe that an increment of exactly one ply is optimal - it is possible that a value of slightly above or below one ply would work better. It might also depend on the phase of the game, and on other considerations. For instance, at the end of an iteration, if you have some time left, but probably not quite enough to finish another iteration with a full ply more, it makes some sense to only increase the depth by only half a ply for the next iteration.
I have personally never tried anything else than exactly one ply per iteration, but I think it might be worth some experiments.
But did you try to use half a ply only at the last iteration, and only when there was not much left of the allocated time? This seems to me like an interesting thing to try. Another idea is to use more than one ply per iteration in simple endgames, where the branching factor is low. As has been pointed out elsewhere in the thread, an increment of two plies works well in checkers, so why not in simple chess endgames as well.Several years ago I tried several options, including 1/2 and 3/4 of a ply. In some cases it works better (those positions where iteration N+1 somehow blows up and takes way longer than expected). But in more cases, it was worse...
Tord