Author: Ian Osgood
Date: 12:16:11 02/14/02
Go up one level in this thread
On February 13, 2002 at 17:13:25, Gian-Carlo Pascutto wrote: >On February 13, 2002 at 13:58:41, Uri Blass wrote: > >>Some questions: >>1)What is the rating difference from having hardware that is twice faster in >go? > >Approximately 0. > >Go programs are (in the general case) not time-bound. Making a program >that demonstrably plays better given more time is an open problem in >the Go world. > >-- >GCP As an example, a few years ago the winner of the World Go Championships was a program named HandTalk, which was clearly stronger than the competition. Its other distinguishing trait was it made its moves faster than any of the other programs! Programming Go is still in the realm of wide open research, whereas programming chess is an engineering exercise on how to best optimize an alpha-beta search. As a result, you see quite a wide range of techniques used in the Go Programming community, both between programs and even within the same program. This also makes Go a more interesting project for AI research. For instance, a typical program may use alpha-beta to search the main line of moves, but use Proof-Number search for tactical evaluation. Some of the early successful programs had no search at all! Most programs have a large database of patterns for generating likely moves, rather than sorting the exhaustive list of all possible moves. Many programs have a "joseki library", which is like an opening book applied to one corner of the board. Some programs use transposition tables, but since the search is shallower they are not as useful in Go as in Chess. Ian
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.