Author: Dann Corbit
Date: 08:45:26 02/12/99
Go up one level in this thread
On February 11, 1999 at 21:35:54, Christophe Theron wrote: >On February 11, 1999 at 20:19:53, Dann Corbit wrote: > >>Without some kind of cheezy trick, does any program score those 5 positions >>listed correctly? >> >>If not, what are the choices your programs make? >> >>If so, what are the pv's and do the ce's reflect that the programs actually know >>what they are doing? > >Tiger finds only POS06, quite quickly. This test is more a tactical one than a >positional one. > >N 9 9.67s Rd7 Rec1 f6 Rxc2 Qxc2 Qf1 e5 Rc1 Qa2 Ra1 Qd2 +0.14 >N 9 12.47s e5 +0.15 >N 9 15.88s e5 Rab1 Qd5 Qa4 Rdc8 Ba1 Qa2 Qa6 R8c6 Qxa7 Rxf2 +0.76 > >So Tiger 11.7.6 finds e5 in 12.47s (K6-2-300, 8Mb hash). > > >The other positions are rather hard. Don't mind too much if your program does >not find them. You could add as well CMB11 and CMB12 in the list of "positions >too hard to solve, ignore them". We might ask ourselves, "Why are they too hard to solve?" It seems that the answers are very likely correct, even though computers cannot find them. Since people can find them, we might ask -- "How do I decide that this position makes the move worthwhile?" If we can quantify it, we can write an algorithm to solve it. For instance, at the lct site, there is some prose description justifying the move choices. That prose could be changed into algorithms. The only problem is how to weight these factors. This question could be answered by a least squares fitting routine with a large amount of data, or perhaps by some other means. When we have a problem such as: "These we know the answer to, but our programs are not smart enough to solve them yet." That seems like a huge gold mine to me. If we can figure out how to solve that problem and no one else can, then we can write a program stronger than anyone else's until they figure it out also. Every problem we encounter is also an opportunity. The bigger the problem, the bigger the opportunity.
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.