Author: Robert Hyatt
Date: 08:34:05 02/17/03
Go up one level in this thread
On February 17, 2003 at 05:24:37, Steve Lim wrote: >I long time ago, I asked Bob about using Positional Learning to help solve wild7 >(Chess variant on ICC). > >I got the same response about local maxmia issues. I believe I understand the >example given... however I proposed a solution to help solve the >wild7 issue. > >Simple increase the threshold that triggers learning to some high number (eg. >MATE score). Then _only_ when it sees mate in the search will it learn. It >should back up the tree with time.. > >I don't really see any maxima issues at all with this workaround.. albeit it >will take a long long time to work. That is the key. The larger the learning threshold, the longer it will take to reach a position where learning is triggered. And then that score has to work its way back toward the root of the tree as more games are played. I won't venture a guess as to how long it would take for the scores to work back to the root and cause the first losing move to be discarded, but "long time" comes to mind. And now, think about how bushy that tree is from that position forward, and how many lines you are going to have to prove bad, before you find the one that leads to a win (only one winning move unless your opponent makes a mistake). This works pretty well in a "close neighborhood". But this is trans-continental, in effect, because the distances are huge. > >Note: I never did see crafty make progress despite playing thousands of wild7 >games against itself using the above mentioned settings. > > >Steve. > >On February 16, 2003 at 12:16:00, Robert Hyatt wrote: > >>>>This is a well-known issue, and I should add that position learning doesn't >>>>claim to solve everything. It is really a defense against repeating the same >>>>game multiple times which humans have been known to do. But as you notice, it >>>>will not solve every kind of position, because of local maxima issues. IE >>>>there are some holes you can not climb out because to do so means you first >>>>have to go for a _lower_ score, to reach the valley floor and start to climb >>>>up the other side... >>>
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.