Author: Eugene Nalimov
Date: 15:37:24 07/14/99
Go up one level in this thread
On July 14, 1999 at 10:29:56, Robert Hyatt wrote: >On July 14, 1999 at 01:30:22, Dann Corbit wrote: > >>On July 14, 1999 at 01:20:26, Dann Corbit wrote: >> >>>It might be interesting to make a pawn race tablebase for large numbers of >>>pieces. Suppose that you have 2 king and x/y pawns left. It might be >>>worthwhile to compute "how many plies till I convert? How many plies till enemy >>>converts?" If I have a two ply advantage, it seems significant. >>> >>>If you have a king and three pawns on both sides, we could never count all >>>possibilities. But we could find out how fast we can convert. >>Where is such a thing useful? >>I think it unlikely to find two kings and 12 pawns on the board. And 5 piece >>tablebase files are common. So for small numbers of pawns, yet too many pieces >>for the tablebase files, it might be very useful information. >> >>Also, since we know we are computing a pawn race, it may simplify calculations. >>Because we stop once we convert (or perhaps a couple moves after) it would not >>be nearly as difficult as a complete endgame tablebase. > > >Tablebases are not nearly so hard to compute as they are to store, already. >Just remember that you need roughly 64^N bytes, where N is the number of >pieces on the board counting kings. for N > 5, this is _huge_. KP...Pkp...p is a "good" TB because (1) there are a lot of pawns, so replace "64" by "48" in the formula, (2) there are a lot of identical pieces, and with 2 identical pieces TB is 2 times smaller, with 3 - 6 times smaller, etc. (2) If there is equivalent set of white and black pieces, you can store only one TB from the pair. For example, KPPkpp will contain "only" 3,197,807,670 positions using my indexing schema (slightly more if you want to store positions where en passant is possible). Nevertheless, KPPkpp is near the limit for current desktop computers. And I don't see principal difference between "full" TB and "pawn race" one. Yes, "pawn race" can be done faster - you don't need to generate minor TBs, but it'll contain too much inaccurate positions. Eugene
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.