Author: Peter Fendrich
Date: 10:42:04 10/02/98
Go up one level in this thread
On October 01, 1998 at 21:25:11, Christophe Theron wrote: >On October 01, 1998 at 17:54:48, Jim Phillips wrote: > >>On October 01, 1998 at 08:25:15, Roberto Waldteufel wrote: >> >>> >>>On September 28, 1998 at 11:28:42, Peter Fendrich wrote: >>> >>(snip) >>> >>>>Sure, winning games is better than losing them... :) >>>>Search does in fact have some good effects by itself... >>>>There was an intresesting article in ICCA Journal a few years ago. I don't >>>>remember any details about the authors and such but here is what I remember from >>>>the article. >>>>They played two programs against each other with completely random evaluation. >>>>One program searched the tree, gave the leafs a random evaluation and and backed >>>>up the values in an alfa/beta manner. The other program just gave each move from >>>>the root a random evaluation. The first thoughts about a match like this is that >>>>the result will be as random as the evaluation code, but it wasn't! >>>>The results showed that the tree version was better because of a tendency to get >>>>more space just because of the tree search itself. Well, my memory is fading >>>>here... >>>> >>>>//Peter >>> >>>Maybe the tree version would find shallow mates, and the root version would not? >>>I think this might well account for the difference. >>> >>>Best wishes, >>>Roberto >> >>I'd like to throw in my 2 cents.... If you choose a move at random from >>the "leaf" nodes, you are going to tend to return to a root move that >>generates a "bushier branch". Let's say you have three possible root >>moves, and you search two plies deep and choose a move at random. If >>there are 10 "leaves" corresponding to the 1st root move, 20 leaves >>corresponding to the 2nd root move, and 70 leaves for the 3rd root move, >>there is a 70% chance that your random function will choose the 3rd root, >>which is the one that will also lead you to a more open position with a >>lot of possible moves to make. The program that simply chooses a move at >>the root at random has only a 33 1/3 % chance of picking the 3rd root.... >>Does this seem reasonable, or am I off around Jupiter? >>-Jim Phillips > >You are absolutely right. The random evaluation in fact turned out to be a way >to find the moves that lead to more mobility. > >Both programs were able to find shallow mates. This was not the issue here. > > > Christophe Correct! And this is the article: Levene, M. and Fenner, T. (1995). A Partial Analysis of Minimaxing Game Trees with Random Leaf Values. ICCA Journal, Vol. 18, No. 1, pp. 20-33. (A) The article is a continuation from: Beal, D.F. and Smith, M.C. (1994). Random Evaluations in Chess. ICCA Journal, Vol. 17, No. 1, pp. 3-9. (A) //Peter
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.