Author: Jay Scott
Date: 13:06:57 01/06/99
Go up one level in this thread
On January 05, 1999 at 16:15:47, Dann Corbit wrote: >How might this work in a database? Suppose I have a few million analyzed >points. I can look at some point, and suppose it wants to move g4. Then I can >move to g4. Hopefully, g4 is already there and analyzed. If not, I can >generate and analyze it. Now, I can take the new information and update our >first position. Maybe from here, the analysis suggests bxh6 for the opponent. >So at the same time, we extend the opponents view from that position, using the >same methods. We continue this process until we reach termination >(win/loss/draw). OK, you play out the game to the end, using database info while you have it. Do you back up the win/loss/draw score all the way back to the original position where you played g4? Or what? >Now that we have done this, we repeat the process for every >point in the database (not yet analyzed). After we have done so, it turns out >that our first move, g4 was wrong! So instead we take the new suggestion and >analyze that. This process can be repeated as many times as desired. Each >time, we make a linear path, choosing only what appears to be the best choice. This sounds like a database-oriented variant of what I've called playout analysis: http://forum.swarthmore.edu/~jay/learn-game/inspire/playout.html It can also be seen as a selective search with a very large persistent tree. In fact, if you always started the analysis from the root then it would be a classical best-first search. You sound more optimistic than I am about how well this is going to work. I'd say you're pretty much guaranteed to discover some interesting surprises, but I think you'll have a hard time convincing me you've achieved the "exponentially improved" analysis that you're expecting. >The reason I call this process simulated annealing is because of the similarity >to the physical process. We start out with a large number of small crystals, >and then melt them together until they become very large. This looks completely different from what we artificial intelligence people call simulated annealing. How confusing! This page introduces simulated annealing in the traditional sense: http://chem1.nrl.navy.mil/~shaffer/optsa.html Jay
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.