Author: Stephen A. Boak
Date: 19:43:19 03/15/03
Go up one level in this thread
On March 10, 2003 at 14:48:06, Uri Blass wrote: >On March 10, 2003 at 13:55:52, Matthew Barnett wrote: > >>On March 10, 2003 at 05:34:07, emerson tan wrote: >> >>>what program is best to play correspondece chess against humans? >> >>When analysising middlegame positions with either the Deep Position Analysis >>feature in the Fritz GUI or just Infinite Analysis I came to the conclusion (a >>long time ago) that often the longer you leave the engine (any engine) to >>analyse a position the more inaccurate it gets. It's easy to demonstrate that >>the evaluation of a position, say 18-30 plies deep might be +0.10 but when the >>position at the end of its analysis is in turn pasted and analysed that >>evaluation may well drop to -1.10. > >It is known that you never can trust the move in the end of the pv and the time >control is not important but things are different for first moves of the pv. > > > I'm fairly sure that, in most middlegame >>scenarios, engines get weaker the longer you leave them. It's a different >>matter in the endgame! > >You are sure about wrong things. > >Engines get stronger in middle game positions if you give them more time >unless the engine has bugs. > >looking at the full pv is irrelvant because I talk only about the first >move of the pv. > >I often give the engine to analyze again for hours after the first 2 or 3 moves >of the pv. > >Uri The point is (I think): The eval given for the PV is based on the score at the leaf points (end of the analyzed best line). If that score is actually erroneous, then the best move recommendation given by the program may be incorrect. I have also noticed this behavior before (that evaluating the end of a given line shows a much different score than the line itself was given). This is not surprising, in itself. Deeper positions in the analyzed line will not themselves be analyzed as deeply as the initial game position. But it suggests complete reliance on program eval scores is done at one's peril.
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.