Author: José Carlos
Date: 07:05:49 04/07/04
Go up one level in this thread
On April 07, 2004 at 07:14:36, Tord Romstad wrote: >On April 07, 2004 at 05:11:44, José Carlos wrote: > >> You share your ideas, and that's great. Keep doing so please. > >I will, whenever I think I have something interesting to share. > >I'll now make an attempt to give a simple explanation of the more general >idea I mentioned: > >Like most programmers, I don't do a null move search at all nodes, but only >at nodes where I am reasonably sure of a fail high. I regard the null move >search as a technique to prune moves which does not even threaten to bring >the score up to alpha. > >At the node directly following the null move, the side to move should try >to prove that the move played before the null move contained some important >threat. Therefore, it makes sense to search some moves more deeply than >others after the null move. Moves which were made possible by the previous >move should be searched deeply, and also moves which attack squares and >pieces which were also attacked by the previous move. Other moves, which >have no logical connection to the move played before the null move, could >be searched with reduced depth. > >It would clearly take a lot of thought and experimentation to make something >like this work well, but I think the idea might have some potential. Yes, the idea looks interesting. In Averno I try null move everywhere (even in PV nodes) except some critical type of endings. However, in Anubis, where I have a lot of information in every node because I evaluate interior nodes, I try to guess whether a null move has any chance to fail high. I also try to guess whether a null move might hide a potential king attack due to reduced depth. All of this work is stopped now because I'm looking for a new job (got many problems in my current job) and I spend all my time preparing interviews and studing, but the ideas have potential and a good implementation will surely pay off. I also use evaluation of interior nodes to calculate what I call a "degree of uncertainity", meaning that some positions are pretty clear and don't need further analisys (and can thus be safely pruned) and some others are very complex and even a big material and positional advantage is not enough to safely prune. As an example, I set uncertainity to "high" in pawn endings if there's at least a passed pawn. There're many possibilities to this concept which I need to explore further when I have time. José C. >> Even if you make >>a mistake, you: >> -get feedback and avoid useless developing+testing >> -give something interesting to think about > >Yes, but in this case my mistake was so obvious that I was just wasting >everybody's time and bandwidth. > >> Congratulations on your success with Gothmog. Really impressive how your >>"stupid" ideas are working on your program. I wish I was so "stupid" as you. > >Thanks, but it seems to me that Averno is around the same strength. :-) > >Tord
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.