Author: Heiner Marxen
Date: 01:37:19 01/06/01
Go up one level in this thread
On January 05, 2001 at 17:42:49, Gian-Carlo Pascutto wrote: >On January 05, 2001 at 14:41:51, David Rasmussen wrote: > >>I was wondering why some of the clever ideas I've been reading about, aren't >>used in the programs I know of. >> >>ETC, Enhanced Transposition Cutoff, seems like a very good idea and have been >>tested to perform well in practice. Why isn't it used? > >The overhead to generate all moves at each node, and make/unmake/probe >them all is too great in most highly optimized chessprograms as compared >to the benefits. > >Of course each program is different, so you need to experiment. > >-- >GCP Chest (my problem solver program) uses ETC since Feb 1991 with good success. (Invented that on my own, learned much later that it is called ETC ;-) Hit rates can be as high as 90%, although hit rates around 50% for the multi-probe are more typical/average. For a playing program the numbers may differ substantially, but you should generate some statistics, seperately for each remaining depth, before you give up. Of course, the expected size of the subtree to search must be large enough to justify the expected cost of probing the TT. I estimate that by remaining depth and branching factor. The details are found by experiment, of course. Heiner
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.