Author: Robert Hyatt
Date: 11:02:32 08/28/98
Go up one level in this thread
On August 28, 1998 at 13:00:15, Jeff Anderson wrote: >On August 28, 1998 at 09:59:15, Robert Hyatt wrote: > > >>I hate to throw a damper on this, but it won't work, period. The first reason >>is that if the tactical engine rejects the positional engine's best move, the >>positional engine won't have a "second-best" move to try, because alpha/beta >>doesn't produce that. To obtain it, you have to do a full re-search, but with >>the rejected best move eliminated from consideration. > >As you can probably tell my understanding of computerchess is not nearly as >great as my interest, but when I have Fritz5 'watch' a position it has a list >move moves it analyses, I can have it analyse 1 move, or all the legal moves in >the position. Now if I had another computer right next to me doing the same >thing, but with a different engine all I have to do is follow a few simple rules >and I can produce my move. With a history in computer chess like yours you >ought be able to think of some even better way to to combine to chess engines >with very different styles. > > >>To make this work, you take a *huge* performance hit, and will play far worse >>than if you do what we are doing now... > >Yes, each engine would only have half of the hardware use than if it were just >that engine alone, but in this day and age technology is progressing very >rapidly. I look at some of the hardware being used by chess computers on ICC >and am amazed, quite a bit of the programs are running on very fast computers >with multiple CPU's. On computers like these I think that two very unique chess >engines, probably a tactical engine and a positional engine, working together >efficiently would be able to play a very good game of chess. >It is so simple, the tactical engine checks for blunders in the positional >engines analysis, and it also checks for shots and such, and the positional >engine will be slow but have a huge wealth of chess knowledge. I think that the >engines would compliment eachother and perform better than programs that do it >all in one today. > >Perhaps someone could try it at home with Fritz 5 and Hiarcs in 'watch' in Fritz >5. Though these engines are far from ideal because they were designed to work >on their own and (Hiarcs must have comprimised some of it's knowledge to be able >to see blunders, and Fritz 5 must have comprimised some of its speed to not look >like an ass in closed positions), this would still prove that the idea is >workable. > >Jeff This has already been done once. Check out the papers on "Sun Phoenix" which used a smart program and a dumb (minix) program to do just this. But as I said, the problem comes in when the tactical program says the positional program's best move is no good. Because alpha/beta doesn't give a "second-best" move, all you know is move X is best, and all the rest are "worse" but you have no idea how much worse. Unless you search again without the "lemon" move in the move list. And the computational cost for that is huge. And it plays worse than if you just used the positional engine by itself...
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.