Author: Tim Foden
Date: 06:52:01 02/22/04
Go up one level in this thread
On February 22, 2004 at 08:25:13, Bob Durrett wrote: >On February 21, 2004 at 22:05:52, Jaime Benito de Valle Ruiz wrote: > >>Hi, >> >>I'm sure we can all improve lots of things here. >>However, I'd like to point out something: Most engines give a score between >>-2.00 and -4.00 more or less, in a position where the side to move (black) is >>hopelessfuly lost. Your engine gives a high positive score for white (if I >>understood, the side to move). This, in my opinion, is the reason for we started >>this experiment: To find flaws. >>My engines is still too weak to be compared to most engines here, and my scores >>would only polute the whole thing, rather than help others. But still, I've >>already found uncountable ideas to improve it. I find all this unvaluable. >>Keep giving suggestions, and we'll all find a way to make all this more fruitful >>for everyone: Soon enough, we'll split the eval function into themes (or parts), >>so we can refine our own work. Until then, we still have a long way to go. >>Regards, >> >> Jaime > >If the "eval score" outputted to the user is what you are talking about, then >the user perspective surely must be the desire for accuracy. Ideally, all >strong engines should give the same eval score for a given position if the >engine is allowed to think long enough. No Bob, what we are doing here is asking the engines what value they give to a position that that (in your terms) they haven't thought about at all. I.E. They take a quick look at the position and total up things like the material, and bonusses for piece centralisation, bonusses for good and bad pawns structure and king safety, etc. They get a total value for this. As you can see, all the engines so far give different assessments of the positions. When the engines are thinking (as programmers we call this searching) they are looking at lots of possible variations, and at the ends of the variations they make exactly the assessment of the position that was explained above. Then, with the scores they have computed, they use a technique called min-max (or neg-max it's the same thing really) to decide what is the best move to make from the current position. Thus these position evaluations can be very important, and the value shown when thinking will be ONE of the thousands or millions of scores from positions that we searched though while thinking. Cheers, Tim.
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.