Author: Robert Hyatt
Date: 20:49:17 09/01/02
Go up one level in this thread
On September 01, 2002 at 13:41:19, Gian-Carlo Pascutto wrote: >On September 01, 2002 at 13:28:20, Robert Hyatt wrote: > >>On September 01, 2002 at 03:20:20, Gian-Carlo Pascutto wrote: >> >>>On August 31, 2002 at 23:54:31, Robert Hyatt wrote: >>> >>>>Interesting question. Deep Blue essentially used it in the chess hardware, >>>>which means the last software ply was a sort of mtd(f) search. >>> >>>Except that it was missing the 'm' in mtd(f), which made it horribly >>>inefficient. >>> >>>-- >>>GCP >> >> >>I don't agree. They simply had a piece of hardware that could search a >>null-window tree, and nothing else. Which is all a single search in a single >>iteration of mtd(f) can do. The software provided the "m" at the point where >>the software handed things off to the hardware... > >Nonsense. The point of MTD is to use a hashtable to prevent wasted work when >researching the tree and trying to converge on a value. The Deep Blue chess >chips did *not* have hashtables. This makes them horribly inefficient, as anyone >that has actually used or uses MTD will tell you. > >The fact that the software part of their search had hashtables has *nothing* to >do with this. > >-- >GCP OK... Good point for the hardware part of the search, I wasn't thinking about M=memory at all... However, the hardware did the searches so quickly, and the searches were so shallow, that does offset part of the loss a normal program would face with mtd(f) with no hashing at all. They only did this the last N plies, where N was not very large (4-5-6-7 according to their logs).
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.