Author: Jesper Antonsson
Date: 04:39:01 05/06/01
Go up one level in this thread
On May 05, 2001 at 22:09:31, Robert Hyatt wrote: >On May 05, 2001 at 21:36:07, Jesper Antonsson wrote: >>On May 05, 2001 at 21:11:55, Robert Hyatt wrote: >>>On May 05, 2001 at 20:04:06, Jesper Antonsson wrote: >>> >>>>True, but perhaps even more so for "bad" than "speculative". I would say that >>>>the better the eval, the more diminishing gains you will see. This is quite >>>>obvious when you think about it, if you get a correct best move early (which >>>>will happen with better eval) the best move won't likely change after that. >>This means that a steeper diminishing gains curve is a sign of quality. >>> >>>I think it is the opposite. The simpler the eval, the more pronounced the >>>diminishing returns, and vice-versa. Deeper search with a good eval not >>>only finds "tactical" things but it finds "positional tactics" as well... >> >>Then I ask you to think it through again. Deeper search with bad eval will often >>have to change it mind because the search depth will refute the choices made >>based on positional knowledge. Good eval will mean that there is on average less >>earlier choices to refute by deeper search, so it won't change it's mind as >>much. Better choices earlier must lead to fewer change-of-minds. > >I am thinking of a different way of evaluating. Two cases: (1) a very simple >eval. Deeper search only helps it find tactical ways of winning material or >tactical ways to achieve positional goals, but only those positional things its >eval can see. Uri came with an interesting extreme in eval simplicity: give everything 0.00 except when you find mate. This program will start on a pretty low best-change-rate, but on the other hand that rate will taper off *extremely slowly*, as it has essentially random move ordering and it will throw out one bad move at a time as it finds they lead to mate. It is the *steeper curve* that is a sign of quality, if I claimed it was the best-change-rate at a certain ply depth that signified quality, I apologize. >(2) a program with a complex/speculative evaluation. The eval >can offer speculative options that a deeper search will refine by eliminating >the speculative stuff that is tactically refuted. > >If you are thinking of "better" as in better eval requires less searching, >then I might agree. I am thinking about a different kind of "better eval" >that doesn't try to replace search with positional knowledge, rather it depends >on the search to help it out... > >This is how I would characterize my evaluation. Parts of it are designed to >replace search depth. Parts of it absolutely depend on search depth. If you do that right, you might start off with a slightly higher best-change-rate, but it will also taper off more quickly. If your speculative stuff is long range positionally unsound, however, the rate will taper off more slowly as the mistakes are slowly uncovered by search. IE, more pronounced diminishing returns are a sign of quality. Eventually, I will try to make experiments to prove my claims, but don't hold your breath. :-)
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.