Author: Fabien Letouzey
Date: 01:09:06 01/18/05
Go up one level in this thread
On January 17, 2005 at 15:39:16, Steve Maughan wrote: >I've just had a look through Fruit's excellent source code. I notice that Fruit >is now using a slightly unorthodox transposition algorithms. Fruit is storing a >minimum and maximum depth, as well as a minimum and maximum score. Has this >approach being documented? I think I've heard of it in a MTD(f) architecture >but not with a PV approach. I would have thought that with a PVS search the >vast majority of positions encountered in a search would either always fail high >or always fail low (unlike MTD) - therefore negating the value of storing two >bounds. What am I missing? Does anyone have any insights? Has anyone >experimented? > >Steve I did it just because I had the space for it (I don't want to use only 12 bytes per entry). Also a separate depth for best move should allow a bug-free PV extraction later (using truncation when unsure). I don't think it makes any difference with my current code. This is probably better for MTD(f) though, I know Gothmog uses separate depths as well. Although I use full-window PVS (no aspiration of any kind, of which MTD is an extension), I want my code to be compatible with other search schemes (not tested yet). Fabien.
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.