Author: KarinsDad
Date: 20:54:11 04/17/99
Go up one level in this thread
On April 17, 1999 at 03:44:55, Jeremiah Penery wrote: [snip] >It sounds like a good idea (what you propose), but I think your information on >TBs isn't quite correct. >Tablebases do not store any moves. They are a data-stream of all possible >positions along with the result of that position (Moves to mate [or conversion, >depending on the type of TB's], or draw). When the search encounters a >position, it sees that there are (say) 10 moves to mate. What it does then is >look up all adjoining positions for the one that says '9 moves to mate'. It >then makes the move that will cause that position to happen. Ah, this explains a lot. Wish I had a spec for this, I'm sure there is one out there somewhere. > >There has been talk of result-only TBs [Storing only win, loss, or draw, not the >distance to these things] to make the 6-man files (Then possibly probing a >full-sized table only at the root) - They would be much smaller than the full >size TBs would be (I'm not sure how much smaller...). If I understand you correctly, this means that the 5, 4, and 3 man files would be removed from the table since you must get to 6 man before the others. This would seem to make it smaller, but I doubt it would be significantly (maybe 10% to 20%, if that?). The reason I say this is that there must be MANY takes on a board to go from 6 man to 5 man (and into multiple cases). If someone knows the sizes of the 3 man, 4 man, and 5 man, they may be able to extrapolate a rough guess on the 6 man size (or maybe someone has already figured it out, but it would take some real work). The other issue here is that if you do not have a mechanism to make the "next move", then instead of just win/draw/loss, it may be important to keep some type of "score" (I am not yet convinced of this however). My thinking is that one may not be needed if the program had some smarts in it, however, I could see a position where you would ramble about the board making moves that all bounce between 9 and 10 moves from mate (since they were found in the "new type of tablebase") and you would never get there (or you would get there in 25 moves since it took you that long to accidentally pick a different winning move that did not lead to draw by repetition or somesuch). >I think it would be quite difficult to write a good algorithm for a 6-piece >ending (or even some 5-piece), however. If the algorithm can't be done >sufficiently well, the result-table may not help all that much by itself. Agreed. Thanks for the info, KarinsDad :)
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.