Author: Robert Hyatt
Date: 16:56:54 09/08/04
Go up one level in this thread
On September 08, 2004 at 19:16:15, Andrew Platt wrote: >On September 08, 2004 at 17:17:38, Robert Hyatt wrote: > >>For example, the one-reply-to-check extension can be made to skip the first such >>extension by just using 3/4 of a ply. Adding 3/4 of a ply doesn't extend, but >>if you get another one-reply extension in the same path, now you have extended >>1.5 plies which adds 1 ply of depth. Alternatively you can do as I do and start >>every search at x.75 plies. IE my 5 ply search starts with a depth of 5.75 >>plies, but that only searches to 5 plies normally. Any 3/4 ply extension will >>take that to 6.5 which is the same as 6 plies. Another 3/4 ply extension will >>take that to 7.25 which extends another real ply. Another 3.4 ply extension >>takes us to 8, another real extension. But another only takes us to 8.75 which >>doesn't change the depth. IE using 3/4 ply extensions means you are really >>saying "extend 3 times and then skip one, which is one way of controlling tree >>growth. >> >>Different extensions can interact using this approach, so that one says extend >>for 3/4 plies, another says extend for 1/2 ply, the net result is a 1 ply >>extension... > >Interestingly in a simple case of just check extensions this will have the >opposite effect as Ed's - at least for the first few extensions - because he's >dropping it at the beginning and then catching up later whereas you are dropping >it later in the search. He estimates 10 ELO difference; do you believe that >would be the case for Crafty? > >Andy. No idea. But remember, I "pre-load" the extension with +.75, so that the first +.75 will trip over the boundary and extend. IE the first 3 extend and the 4th will not. Without that pre-load, the first would not extend, but the next 3 would. I tested once and the pre-load .75 seemed to be best. But I probably should test it again to see if that is still the case. Without the .75 pre-load the search certainly goes faster.
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.