Author: Brian Richardson
Date: 07:44:56 05/26/00
Go up one level in this thread
On May 26, 2000 at 08:38:08, Jari Huikari wrote: >There has been some discussion about usefulness of pondering under subject >"LG2000 vs Crafty more results". > >I'd like to know how people have implemented this feature. Is it common >way to have one guess for the opponents next move, and then calculate >further based on this one guess? Do some programs ponder with several >guesses for opponents move? Or even calculate all possible alternatives? > > Jari Tinker looks at all possible opponent moves as part of the search for its own move, but only picks one (the "best") to ponder. With no book, it always has a move to ponder, after the first move in a game (e.g. Tinker does not "puzzle" for a move to ponder if it does not have one already). Hash table aging and search time for the next move vary when the predicted (pondered) move is in fact made, or not. For example, if the prediced move was made, if the pondering time was greater than the new search time, the next move from the prior search is made immediately (subject to certain depth and other search safety constraints like two predicted moves in a row). If the predicted move was made but less pondering than new search time was used, partial credit for the ponder time is given, an so on. Brian
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.