Author: Robert Hyatt
Date: 12:41:11 05/18/01
Go up one level in this thread
On May 18, 2001 at 14:55:06, Dan Andersson wrote: >The function you use to trigger the SE test need not be only static in nature. >One or more components of the function could be predicated by previous >behaviour. For example: Adding a density function taking into account a >increased or decreased probability of singular moves for the part of the tree >you are searching presently, maybe hashed information. Multiplying the static >factors by a function based on how previous predictions went. And adding a >factor based on time to avoid overtraining. Any such dynamic scheme is almost as >fast as a purely static function (even if it would be a higher order function) >and I would be surprised if it did not significantly outperform the static >function search, YMMV. I use such a scheme to enable a pn-search based pruning >currently. You could also make a profiler to measure what factors contribute >positively and negatively to correct prediction. The possibilities for >automation are big. > >Regards Dan Andersson I don't like the idea of a "static test" to trigger the SE test. Hsu's paper explained the SE idea pretty clearly, and it is _obviously_ a dynamic extension. Not trying it unless some static condition is met seems a bit odd... why not just use the static condition to trigger an extension, period, rather than to test to see if an extension should be done... In DB, the SE test was done wherever it was reasonable to do it... Rather than as the result of a static qualification test.
This page took 0.01 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.