Author: Bruce Moreland
Date: 14:30:03 10/26/00
Go up one level in this thread
On October 26, 2000 at 09:58:53, Sven Reichard wrote: >I have a question about the implementation. Right now I call SEE before every >move in the qsearch. Suppose QxB loses 6 points, and I discard the move. Two >plies down the tree I call SEE again, and QxB still loses 6 points. Does anybody >have experience of updating the SEE information incrementally? If it works, it works. I don't think it would make sense. A large part of the time you hit a quiescent node you return without generating any captures, since the static evaluation score is higher than beta. In many remaining cases you find a capturing move that looks good, you recurse once, find no suitable replies, return alpha, then immediately return beta because you failed high after finding a good capture of dropped piece. My point is that if you figure out how many times you get any sort of depth in a quiescent search, it should be very small. So if you are going to save data structures up front, which you can use at a deeper depth, it had better be a massive savings deeper, at essentially no cost. That sounds like a rough set of requirements. This is, of course, testable. bruce
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.