Author: Danniel Corbit
Date: 22:26:22 07/02/98
Go up one level in this thread
On July 02, 1998 at 19:10:43, Robert Henry Durrett wrote: [snip] >I suspect that the word "massive" would be the achille's heel of the proposed >project. > >As an alternative, this might be easy to program: > >(1) Start with a large database of serious games [not blitz, etc.]. >(2) Throw out all games in which neither player was titled. >(3) Create an ordered set of chess positions, with the positions recurring most >often listed at the top. >(4) Cull out all positions not generally regarded as being the "critical" >positions of an opening. [A panel of strong chessplayers could do the culling.] >(5) Find other ways to cull out positions of low significance, value, etc. >(6) Then do what you were going to do with the resulting ordered set of >positions. This is, to a degree what I had in mind. Just analyzing the heck out of the "best possible" position from a given FEN location. I don't see a big need to waste time analyzing the openings. Just go from there. There are obviously many criteria for "excellence of move." For instance, which move, taken by at least 30 GM's, resulted in the highest winning percentage from a given FEN? We could heuristically try several different measurements and find out which choice of goals "follows wins to the end" most closely. After we have analyzed the "#1" best possible move all the way to the end of the tree, try the "#2" etc. Naturally, we will uncover some blunders and the like. What I believe is this [and I feel it will be confirmed by sensible analysis]: While there are 28 or so "possible" choices from any given starting point, "usually" only two or three of them are any good. Now, if we store the information in a database, and tag the positions that won, but have hidden blunders, we can discover the best possible paths, and have analysis prestored. Let's try another idea: Choose the very best openings available, that remain good out to 30 half ply, on the basis of (win/loss/draw value) * probablility. Suppose that there are a thousand of them. We could try to find the best position for each of those games, calculating out 15 more moves. That's only 300,000 FEN positions. Now, let's look at a ludicrous number: 3^30=205,891,132,094,649*1000 This [206 million CD's worth] will allow every conceivable [supposedly] first, second, and third best move to be investigated out 15 full moves. I suspect that 99.999 percent of that would prune out as useless. I also realize that very few systems on earth have that kind of storage capacity [though I did work at a place with 120-Terabytes in on line storage]. Further, I fully realize that no computer system in the world has the capacity to tackle a job like this. Nonetheless, I feel that it is worthy of serious thought. Computer systems continue to advance exponentially, as does computer power. No sense waiting until the solution is in hand to get started looking at it. Despite all that enormous effort, some games would [of course] slip through the cracks. But eventually, they would be discovered. So what is all this number crunching crap good for? Well, for instance, if Crafty or Fritz plays a nearly identical game twice, it still redoes all the calculations. But if we saved the best possible track, and the computer power used to analyze it, we could know if we have been there before [and if the result was still good]. Now, imagine one hundred Crafty/Ferret/Comet servers running 24 hours a day, 7 days a week, 365.25 days per year, taking all rated opponents within 300 points and storing the calculations to disk [BEST TRACK ONLY for starters]. Then, when investigating a FEN, they could do a database probe and ask "have I been here before?". For a year or so, it would probably be like endgame database probes early in a regular chess game. Mostly misses. But eventually, it would become smarter and smarter. How much smarter? The equivalent of all those machines doing all that work and never doing anything redundant. They could have the time saved by not analyzing a position "saved in the bank" so that the first moves go like wildfire, and then they have extra time. Right now, there is the "endgame tablebase" exploration which is reeling in the far end of chess. Opening theory reels in the start. I'm suggesting a way to [partially anyway] reel in the middle.
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.