Author: Peter Fendrich
Date: 04:55:01 12/02/99
Go up one level in this thread
On December 02, 1999 at 05:10:20, Ed Schröder wrote: >>Because here is what your program will do: (a) I like Bxh6, but after I get to >>depth=7 I see that it is bad... so (b) I will first play two captures that >>eat 4 plies, then a check to eat a couple more, and _then_ I can now play Bxh6 >>and I don't see anything bad happening. > >This ONLY happens if the software (in this case King Safety)... >a) is TOO speculative (too high values) >b) missing important knowledge -> inaccurate judgement >c) is not in balance with other positional rules (values) But isn't the case that eval functions has all these deficiencies, otherwise we wouldn't need the search at all. Somewhere near the endpoints of the search we have to rely on insufficient eval functions and that will give incorrect values. It all ends up in some kind of statistical questions like: - "What probabilities do I have that Bxh6 fails/succeeds in positions like this" - "What is the maximum error rate I can afford" - "How often will my program play Bxh6 in positions like this". (The last one because it might be difficult to isolate the exact criterias to correcltly identify what kind of position it is) The right answear to these questions will be different in different programs and be dependent on stuff like the programs normal depth etc. //Peter
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.