Author: Robert Hyatt
Date: 07:14:57 07/30/04
Go up one level in this thread
On July 29, 2004 at 02:32:58, Martin Slowik wrote: >On July 28, 2004 at 23:59:41, Robert Hyatt wrote: >>>In some (most?) cases the user is only interested in a subset of the engines >>>abilities. If I use a chess program mainly for analysis and not for engine vs. >>>engine tournaments, I'm definitely not interested in its learning abilities. >>>Instead I want to know how it compares in fresh situations, in positions >>>probably never seen before. >> >>That's short-sighted. IE when you analyze, you often back up, try different >>moves, to see what is going on. Position learning is a great help there to use >>that information from earlier searches to help with later searches to make them >>more accurate. > >I don't think that's short-sighted, maybe we just speak about different things. >It's true, one goes back and forth in the various lines one sees and it's >sometimes surprising to watch some engines "learning" the evals during the >process. Very good at that task seems to be Yace in my experience, especially in >endgames. You can get very high mate announcements that way, for example. > >I thought this is not "position learning" as it is implemented into most engines >- if you start the program next day, it of course has already forgotten >everything and you have to start going through your analysis all over again. >Nothing has been stored on the hard disc. Position learning is simply carrying what you mention above across to a new game. Same idea. Same hash table idea. > >But I'm not a progammer and perhaps this is a somewhat skewed picture of the >truth. > >> The SSDF and most other rating lists unfortunately >>>don't answer that question. >> >>No, but then engines are not designed to answer that particular question very >>accurately. They are designed to play chess from one side of the board, not >>behave like a neutral observer and give zero-sum analysis from either side. >>They do OK at that, but not as well as they play real games. > >Yes, of course, no disagreement here. Twas just a thought about the usual use of >chess programs. There are certainly many people playing eng vs. eng tournaments >and watching the results. For those people it's a kind of sporting event and the >engine designs (including book learning etc.) are on par with it, no doubt. > >On the other hand there is certainly a large population of people who use their >program mostly for analysis of their own or of other people games. I was just >wondering if the programmers (particularly of the commercial programs) have >forgotten those. But it also may be that "doing OK at that task" is enough for >most of those users.
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.