Author: Gian-Carlo Pascutto
Date: 02:28:39 09/06/03
Go up one level in this thread
On September 05, 2003 at 17:26:58, Robert Hyatt wrote: >Now, what about that 1/6? It turns out that it is possible to predict with >some reliability when this is happening, by looking at the node count for >each root move. And when it appears to be a possibility, those moves that >appear to have larger-than-usual node counts can be searched one at a time >with a parallel search, I'm not so sure - 1/6 is not particularly impressive, even half of that isn't. You can do significantly better down in the tree. Your average correct splits/incorrect splits should be way higher than 1/12. Doing better with the tricks is not relevant I'd suspect, since you're not parallelizing at that point. Then there's the issue of less splitting overhead - not so sure how relevant that is. I indeed do way more splits as you but performance doesn't seem to suffer from that, on the contrary. Then there's Tony's remark. Good point, but knowing when your root changes _is_ an intersting thing that's good to know fast, not only in testsets. Som maybe that's not a problem either. This is the kind of thing where you really want to test in games to know for sur eor not. You'd have to play a very long match between them to be sure, and then really hope a statistically significant result turns up. Messy bussiness. -- GCP
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.