Author: Uri Blass
Date: 08:29:35 11/27/02
Go up one level in this thread
On November 27, 2002 at 10:01:02, David Rasmussen wrote: >I've been thinking the same thing. >It is certainly true, that you can traverse the same tree by an infinity of >search schemes. For example, a 5 ply nominal search with some extensions to, >say, 9 ply, can be regarded as, say, a 7 ply "nominal" search, with the boring >lines (i.e. those that did not get extended from 5 ply) pruned. The same goes >for qsearch. In principle, you could skip your qsearch, and just have some >clever extensions that did the same. But what does this all boil down to? This: >qsearch _is_ just a clever extension scheme, if you regard it that way. The >question is: What is the easiest way (and in practice, speed also counts) to >specify what tree you want? I think it is probably much easier to have the >traditional design of separate search and qsearch, compared to combining the >two. A similar question can be asked with regard two extension and pruning. I >guess the main point is: design does matter. It might be much easier to express >something as an extension, than as the corresponding pruning of all other moves, >and vice versa. Design _does_ matter. > >Some time ago, along the lines of these thoughts, I proposed "negative >extensions". That is, if you can somehow classify a move as "probably not >interesting", you can "extend" the depth by -1 or -0.75 or whatever seems >reasonable. Exactly as you do with normal extensions. The good thing about this >is that nothing gets pruned for good, everything will eventually get searched >with iterative deepening, but you search what you think is interesting first. >When I proposed it, a lot of people compared it to null move, which is an >entirely different thing. Also, some people didn't think the idea was >worthwhile. I think it is a good idea, as I think some things are easier >expressed in negative terms: This looks boring, so I will not look so deeply >into to it now. > >Of course, it's just too bad that I haven't gotten around to implementing it >yet. Someone will probably beat me to it. > >/David I am almost sure that some commercial and private programs use it so it is not someone will beat you but someone beated you. Latest not public movei already use negative extensions or something similiar and I found that it was productive for the strength of the program. I am not sure that it is exactly negative extensions because if I find based on search to reduced depth that the move is interesting then I search it again and not to reduced depth. Null move verification is exactly negative extensions if you do not do research because it tells you to search to reduced depth if there is no threat. 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.