Author: Andrew Williams
Date: 02:10:49 09/11/02
Go up one level in this thread
On September 10, 2002 at 18:03:02, Gian-Carlo Pascutto wrote: >On September 10, 2002 at 17:08:23, Dann Corbit wrote: > > >>Since it searches until upperbound and lowerbound converge, evidently, they can >>exchange roles. > >They do not within an MTD search. The search process goes like this: > >guess: 50 > >first search: fail low at 50, new guess 49 >second search: fail low at 49, new guess 45 >[...] >twentieth search: fail low at -20, new guess -19 You mean fail high? ^^^ >twentyfirst search: fail high at -19 > Admittedly I'm jumping into this discussion half-way through, but if in your second search you drop by 4 units, how can you be sure that you'll always straddle the score by 1? In other words, couldn't the score be -18 if you fail high at -20? In which case you do have an interval at one point? So you can approach the score from two directions (if you increment/decrement by > 1). Given this, you *can* find that you've got a gap between the upperbound and the lowerbound. However, I've never been able to improve my scheme with a binary chop. Andrew >At this point, we have converged at -19. > >It always goes in a single direction. This isn't a binary search >by any definition. > >-- >GCP
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.