Author: Bob Durrett
Date: 17:31:32 12/25/02
Go up one level in this thread
On December 25, 2002 at 19:33:16, Mike S. wrote: >On December 25, 2002 at 18:12:58, Bob Durrett wrote: > >>(...) >>The problem is that I am selecting moves different from the ones selected by >>Fritz 8 and then forcing Fritz 8 to analyze my moves. I am merely a chess >>amateur. Why are my moves often better than the ones Fritz 8 selects? This >>should not happen more than once in a million moves. But it is happening maybe >>once in ten moves. I really don't understand this!!!! > >This sound like an effect of different calculation times and/or hash size (or >even CPU speed) used in your analysis, compared to the games Fritz 8 played. It is possible that I am subconsciously not allowing Fritz 8 as much time as I allowed Fritz 7. In other words, impatience. I'm not sure, however, that I'm doing that. I really hate the extremely slow chess computer I'm using. But this is the same computer I used with Fritz 7. Hash is unchanged. >I assume you analyse games Fritz 8 has played, enter a different move >(variation) and the compare the evaluations of the original Fritz 8 with the one >of your new move? I do not delete previous material until almost ready to print my analysis. Most of my time is spent in the search for improvements over what Fritz has found during automated overnight post-mortem analysis. A new move is not "declared better" until the best lines from the two moves have been constructed and compared. Often, I see interesting ideas to try during that process. [most of them are bad] > >In this case, you compare evals of *before* the move was actually played in the >game, with evals after the (alternative) move was entered, IOW one ply later. >When the original move is from short calculation time, I think it's not too >unusual that Fritz 8 comes to a better evaluation quickly (it would probably >come to a better evaluation if you point it to the position *after it's original >move* too). When the moves really are good and Fritz has the advantage, then the >tendency is that the advantage will increase, so if you analyse one ply later it >may search one ply further ahead... Yes, I've seen that a lot. I am producing lines and not just moves. But I do that by a very time-consuming process. i.e. one move at a time. The only way I know to verify a move is to produce a line, thereby culling out horizon effect errors. >You can also cross-check with another engine, to see if you notice the same >effect in general. I guess so. I guess also, that the positions you talk about, >are not sharp tactical one's? :o) Next time I come upon a clear example, I'll post it here. > >Regards, >M.Scheidl
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.