Author: Uri Blass
Date: 02:58:25 08/21/02
Go up one level in this thread
On August 21, 2002 at 01:42:07, Mike S. wrote: >On August 20, 2002 at 17:25:43, Uri Blass wrote: > >>(...) >>No reason to be surprised. >>This test was designed to put Fritz on top because the analysis was done by >>Fritz to decide about the positions. > >What exactly do you mean? > >Do you really mean, "designed to put Fritz on top"? If not - and I assume (and >hope) you didn't mean that - then I suggest to be a bit more careful. Now it >sounds like an accusation. I mean that the testers used Fritz to analyze before deciding about the positions. Vincent said that when he asked about a position they gave him Fritz's analysis as a proof. I think that we need positions when humans are sure about the best move or positions when all top programs agree about the best move if you give them enough time. > >Btw. you probably missed that the test has been developed in several stages >(with increasing number of positions), where in between *Gromit* 3.9.5 (!) once >was on top. The first version (40 pos.) was designed and published before Fritz >was available. > >>I am sure that Fritz is going to have bad result if someone design a test to put >>tiger on top. > >It would be unethical, and I don't expect somebody would do that or ever has >done - if he would hide the intention. No matter which program is favoured of >course (comment for the suspicious :o). > >OTOH, it could be interesting as an experiment if one would explain it before, >that he'd try to see if it is possible at all. It would require to choose >positions where not only Tiger is good, but also Fritz is slower at the same >time (or in the majority of the positions). I think while both are about equally >strong in gameplay, Tiger 14 is usually behind Fritz 7 in testsuites, in >general. I do not know. I think that test suite should to be based on comp-comp games I plan to develop a test suite based on comp-comp games of movei(avoid tactical mistakes of movei or the opponent or find moves that were played) I expect most of the positions to be easy but I believe that there are going to be also some hard test positions. Here is one example that at least Deep Fritz has a problem to solve and tiger solve it with no problem (I am not 100% sure that it is correct at this moment and I only analyzed Nxh4 the move in the game): Movei 0.0.72h - Horizon 3.2 [D]6k1/6r1/2Q3N1/2p5/2P1P1qb/1P1P2p1/P5R1/3b2K1 w - - 0 1 bm Qa8+,Qe8+ I believe that Qd5+ or Nxh4 are losing blunders but I am not sure. Unfortunately even after a lot of learning yace shows after Nxh4 Qxh4 only a small advantage for black. I think that I am going to develop 2 test suites when one is tactical test suites when I feel sure that the move is correct and another one is a positional test suite when I believe that the solution is correct but probably only 90-99% of the solutions are going to be correct. The last position is probably going to be in the positional test because even after Nxh4 Qxh4 and learning that 47.Qc8+ is 1.03 pawns advantage for black yace can see after some minutes only 0.89 pawns for black and suggests 47.Qe6+ at depth 13. Uri
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.