Author: Tony Werten
Date: 16:08:58 09/06/02
Go up one level in this thread
On September 06, 2002 at 18:50:23, Dieter Buerssner wrote: >On September 06, 2002 at 17:17:55, Tony Werten wrote: > >>On September 06, 2002 at 16:58:50, Dieter Buerssner wrote: >>No problem. Point is that kppkpp itself will not help you so much. It might say >>you are winning, but chances are very small you'll checkmate with the 2 pawns so >>you need the conversions as well. > >I am not sure. I have no experience with 6-men TBs (besides KNNKNN, which is the >smallest download, and which I used for some tests. But this has obviously no >practical relevance). I am well aware of the TB problems with 5-men TBs and >missing conversion cases. In all such 5-men positions I checked, my simple code >did work well, and never spoiled a win into a draw. (Various things, how to >avoid those probs have been repeated in CCC). Perhaps, to stay with this >concrete example of kppkpp, it will be much more difficult. I could imagine (but >I cannot check - who can?) that there may be very difficult to win lines that >transform to kqpkqp. Perhaps here, the methods helpful in 5-men TB positions >will fail. Still, I would think, that in the big majority of cases, the missing >conversion TBs, should not hurt. In most cases of kppkpp an engine would manage to find the win. That would be called ~kppkpp wich would be a lot smaller. (Bittables would suffice) But egtb's are meant to give the distance to mate, wich wouldn't be true with ~kppkpp. And it's the exception that would upset people. Tony > >>>BTW with these 6 man tables you probably need a 16 bit counter for distance to >>mate, making them twice as big. (You could do with less extra bits but that >>would kill your compression I think) > >I don't totally agree, here, but I see your point. I think, using 16 bit >counters will obviously boost the compression ratio - all those leading zero >bitr. If another more compact encoding is used, (say 10 bit is enough) the >comression will be worse. The overall result may still be similar (you save >those 6 bits, but compress worse. Compared to configurations where 8-bit is >enough: 10/8 is not such a big factor ... But, of course, much will also depend >on the compressing scheme, and on the actual distribution of the distances to >mate). > >Regards, >Dieter
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.