Author: Dann Corbit
Date: 14:43:09 02/06/02
Go up one level in this thread
On February 06, 2002 at 15:35:28, Sune Fischer wrote: [snip] >Well I don't disagree with that, but suppose in any position we knew what the >best move(s) was. Now how often would a n-ply search find the correct move(s)? >This will give some kind of probability distribution on the plies. I'm sure this >is directly related to the diminishing returns at the higher plies. Diminishing returns at higher plies has not been demonstrated. The research by both Heinz and Hyatt is statistically inconclusive. If you make a program with a material only eval, it will swim through plies like a salmon on steroids. And it will lose all of its chess games to programs searching half as deep. IOW -- plies don't tell the whole story. And what happens when we gain another ply is still somewhat mysterious as far as what the value should be. Here is a puzzler... When we search another ply, we typically expend more effort than all the previous plies combined. Why is it (then) that instead of being at least twice as good in the evaluation, we are only fractionally better? After all, we are looking at exponentially more board positions. It's a bit odd that our strength does not increase at least linearly (or does it, at some point?) The strength increase looks a bit like a decaying exponential (considering the graphs from available research papers and ignoring the enormous error bars).
This page took 0.01 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.