Author: Torstein Hall
Date: 04:11:55 06/12/99
Go up one level in this thread
On June 11, 1999 at 10:39:13, Dan Homan wrote: >On June 11, 1999 at 10:15:19, Torstein Hall wrote: > >>Its known that Fritz and some other fast programs reach their high nps because >>they do their evaluation mainly at the start of the search. >> >>Is it not possible to mix this up a bit. So that you do a "big eval" at the >>start, then at certain depths in the search. Or even better, if some condition >>sets in, like swap of queeen, endgame etc, then do a new thorough eval, else >>just the fast eval? >> >>But perhaps some programs do it that way already? Or perhaps its just not >>possible to implement? >> >> >>Torstein > >I think programs must do some evaluation at the leaf nodes.... if only >because it is very difficult to capture dynamic ideas in a piece-square >table. I tried piece-square only (filling the piece-square table using >a root evaluation routine) for EXchess when I first started it. >It was fast, but I found that evaluation at the leaf nodes gave much >better play. Perhaps you are right and some combination might be >the best approach. > >I've been thinking about this alot recently. One thing that occurred >to me is that this might be an effecient way to create a neural net based >program. > >Say the neural net (nn) is responsible for filling a piece-square table... >Then the nn could operate either once, at the root of the search, or >only selected times during the search. This would largely overcome >the major drawback of nn evaluations which is that they are slow. > >Then we could have a TD based learning program that also learns >evaluation features! > >I've thought alot about building my next chess program on this >idea.... The one major drawback that I see is that there will be >a relatively large number of parameters for learning to adjust. >The other drawback is that I really know nothing useful about >programming neural nets. :) > > - Dan I know next to nothing about neural net based programs ( and programming in general as I have not been programming for close to 10 years now! ), but my idea was simpler. Start with a very thorough eval to guide the search at first, then do some easy eval at the leaf. When you reach major points in the game, like endgame, queenswap, danger to king etc. turn on big eval(). I feel taht this would be fairly close ( or rather closer ) to how humans work at chess! ( ..and perhaps you can reach nice depths and still have mostly correct evals al the way.) Torstein
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.