Author: Eugene Nalimov
Date: 10:01:19 02/29/00
Go up one level in this thread
On February 29, 2000 at 08:52:22, Robert Hyatt wrote: >On February 28, 2000 at 16:50:55, Eugene Nalimov wrote: > >>On February 28, 2000 at 16:48:38, Robert Hyatt wrote: >> >>>On February 28, 2000 at 15:51:58, Andrew Dados wrote: >>> >>>>On February 28, 2000 at 13:49:59, Robert Hyatt wrote: >>>> >>>>>On February 28, 2000 at 10:35:29, David Eppstein wrote: >>>>> >>>>>>On February 28, 2000 at 08:45:32, Robert Hyatt wrote: >>>>>>>On February 28, 2000 at 03:00:24, Gregor Overney wrote: >>>>>>>>How large is KQQKQP? How about KPPKPP? >>>>>>> >>>>>>>First, you can forget kppkpp for a long while. The file will be huge and there >>>>>>>are a _bunch_ of other 6-piece files that have to be completed first. >>>>>> >>>>>>I would think that all of the KxxKyy should be about the same size as each >>>>>>other, and all of the KxxKyz should be about the same size as each other. >>>>> >>>>> >>>>>It depends. IE if the max mate is <= 127, then 1 byte per entry is enough. >>>>>If we get some that have a max mate in 400, they will take forever to compute, >>>>>and they will be bigger. And then there is compression. If a particular >>>>>ending is mostly drawn, the database compresses like crazy. If it is not >>>>>mostly drawn, it doesn't compress as well. >>>> >>>>How important is it to store exact mate in x and not, say, mate in x div 4, >>>>assuming whole database woul be flagged as 'div4'? Progress at root can then be >>>>determined by shallow search till depth 8... (or mate div 3 and search till >>>>d=6). Or am I missing something? >>>>-Andrew- >>>>[snip] >>> >>> >>>How would you force mate? IE you get to choose between mate in N scores, where >>>for one move n=131/4 and in the other case it is 130/4. If you take the 131/4, >>>you will again have to choose between 131/4 and 130/4 the next time. And you >>>might run afoul of the 50-move rule as a result. >> >>Do a full 4-moves search when root position is TB hit. >> >>Eugene > > > >This doesn't work with current code. I could do a 20 ply search, but I get the >TB hit at ply=2 and go no further. Yes, and current TB code also would not produce those N/4 values -- but it always can be modified, as well as Crafty. IMHO, the saving is not worth the trouble. Much better idea is to produce 2 set of TBs - first W/D/L to be used in search, and second one the "normal" one but with much larger block size to be used at root. Maybe when I'll have some free time... 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.