Author: Robert Hyatt
Date: 07:52:16 06/09/02
Go up one level in this thread
On June 09, 2002 at 02:35:09, J. Wesley Cleveland wrote: >On June 08, 2002 at 01:02:23, Robert Hyatt wrote: > >>On June 07, 2002 at 19:44:41, J. Wesley Cleveland wrote: >> >>>On June 06, 2002 at 20:16:33, Robert Hyatt wrote: >>> >>>>On June 06, 2002 at 19:24:54, Roy Eassa wrote: >>>> >>>>>On June 06, 2002 at 19:18:16, Roy Eassa wrote: >>>>> >>>>>>On June 06, 2002 at 18:02:43, J. Wesley Cleveland wrote: >>>>>> >>>>>>>On June 06, 2002 at 16:30:19, Roy Eassa wrote: >>>>>>> >>>>>>>>On June 06, 2002 at 16:25:01, Michael Vox wrote: >>>>>>>> >>>>>>>>>On June 06, 2002 at 10:10:12, Robert Hyatt wrote: >>>>>>>>> >>>>>>>>>When Nalimov 32 piece tbs come out someday, it will be over. This will >>>>>>>>>eventually happen with stronger hardware. At least every worthy line will be >>>>>>>>>saved to dbases. It will no longer be Crafty vs Junior, it will be Crafty dbase >>>>>>>>>vs Junior dbase. >>>>>>>>> >>>>>>>>>No point in discussing computer chess anymore once this level of technology and >>>>>>>>>dbases is hit. >>>>>>>> >>>>>>>> >>>>>>>>I hope you're kidding. Even if every atom in the galaxy were used to store 1 >>>>>>>>bit of data, that still wouldn't be enough storage for 32-man TBs. (And 100 >>>>>>>>billion years wouldn't be enough time to compute them, even on a multi-terahertz >>>>>>>>computer.) >>>>>>> >>>>>>>You are off by a bit. All positions can be stored in ~160 bits, which means that >>>>>>>2^160 or 10^48 bits are enough for all TBs. There are more atoms than that in >>>>>>>the earth. As to calculation time, we should have fast enough computers in about >>>>>>>300 years, if Moore's law holds up. ;) >>>>>> >>>>>> >>>>>>First off, to store tablebases requires more data than just each position >>>>>>itself. Second, why did you raise 2 to the power of the number of bits? >>>>>> >>>>>>How many positions are possible in chess? It's a number with scores of digits, >>>>>>and *each* of these entries would require your 160 bits plus more for the other >>>>>>required fields (next move, etc.). >>>>>> >>>>>>And finally, I doubt Moore's Law will hold up for another 300 years! (If >>>>>>nothing else, it won't take nearly that long before the laws of physics prevent >>>>>>further speedups, at the rate of increase we've been experiencing.) >>>>> >>>>> >>>>>Upon further thought, I understand why you raised 2 to the power of the number >>>>>of bits. But does the 160 bits take into account the additional data required >>>>>in a tablebase? If not, you need several more bits, which should increase the >>>>>final number by orders of magnitude. Plus, I don't think anybody will ever turn >>>>>even every tenth atom in the Earth into storage for tablebases... >>>> >>>> >>>>2^160 represents something just below the total number of unique chess >>>>positions. >>>> >>>>Of course, that has _nothing_ to do with solving the game, because path >>>>information will be critical with the 50 move rule. >>>> >>>>In that context, 2^160 words will barely be a start. It might well bu >>>>2^160^160 for all I know at that kind of problem... >>> >>>If you could construct all tablebases up to 32 piece ( a VERY big if), chess >>>would be solved and it would fit in the 2^160 or so bits the tables would take. >>>It is *quicker* to create tablebases that obey the 50 move rule than tablebases >>>that do not. >> >> >>I don't know about the "quicker" part. But the 50-move rule is a serious >>problem. It would seem that we need to soon convert to a combination of >>DTM/DTC so that we can work around the 50 move rule with some sort of reasonable >>algorithm that will work in a running chess engine. >> >>But in any case, current DTM or DTC by themselves have difficulties that are >>serious problems.. > >If I understand correctly, TBs are generated iteratively one ply at a time >starting from mate or conversion until no positions are marked, and all other >positions are draws. If you stopped after 100 iterations, you would also have >all the positions that could not be won or converted after 50 moves as draws. >TBs with pawns would have to be calculated separately for all possible pawn >positions, but could be combined later. Right. But if you don't stop after 100 iterations, you will have DTM scores > 50 _and_ DTC scores > 50. You could propagate this info thru the TBs as larger ones are built, so that you would see "DTM=430" but "longest DTC=55" which would mean the 50 move rule would be violated if you choose to follow that path, while a 60 move rule would not. I haven't given a lot of thought to this, so I suspect there will be "issues" that are not obvious to me right now. But at first glance, this ought to be able to work.
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.