Author: KarinsDad
Date: 10:12:04 05/05/99
Go up one level in this thread
On May 05, 1999 at 12:45:28, Charles L. Williams wrote: >On May 05, 1999 at 11:32:15, KarinsDad wrote: > >>On May 04, 1999 at 23:40:34, Peter McKenzie wrote: >> >>[snip] >>> >>>How is your move ordering? Expect your nps to drop as your move ordering >>>improves. >> >>I keep hearing this type of statement, but I haven't figured out the reason for >>it. Could someone tell me the rational behind this? >> >>As to my move ordering, I have a sophisticated one in the design spec, but at >>the moment, we have a simple one of pieces most likely to be moved in chess >>first, other pieces later. The order is the same throughout the entire game (at >>this point of the implementation). >> >>KarinsDad :) > > >Move ordering requires a sort to be performed first. To do the sort, a list of >moves needs to be generated. This takes time. With improved move ordering >cutoffs happen sooner, which means less nodes evaluated. However, the moves are > still generated. Although the nps (board positions) evaluated decreases, so >does the overall time to do the search. > >If you change your alpha-beta to a negamax while keeping the eval the same, your >nps should increase for this reason. > >In short, with good move ordering, there is a slight inefficiency because a lot >of moves that will never become a node need to be generated up front for the >sort. I'm on my first cup of coffee and recovering from a cold, so this might >not make a lot of sense. :) > >Chuck Let's see if I can rephrase properly. Since more pruning due to cutoffs will occur, the move generator will generate just as many move (actually, it will have time to generate more moves), but the evaluator will not be called as often (due to cutoffs). Hence, more time will be spent generating moves as opposed to evaluating moves. Since most programs count nps via evaluated moves, the nps will drop. This makes total sense. Thanks. This also gives me a good idea on how to improve performance even more. KarinsDad :)
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.