Author: Bruce Moreland
Date: 00:25:31 01/17/98
Go up one level in this thread
On January 17, 1998 at 02:21:54, Ed Schröder wrote: >Every programmer has his own view how to program the hash table. Typical >for Rebel is that it stores and uses the q-search results in the hash >table (most programs don't do that, true?). Another typical Rebel hash >table behavior is that it doesn't store positions in the hash table that >have no "best move". The latter speed up Rebel a little. I don't do this q-search thing because our searches are probably a lot different. I wouldn't have much use for it the way I do things. But I haven't spent much time trying to figure out how to make the most efficient use of hash memory. I use the Thompson dual-table scheme described numerous places, which has always worked fine for me. I like the idea of not storing elements that don't have a "best" move attached. You lose the ability to fail low on a node based upon the value in the hash table, but a little bit of statistical collection seems to indicate that this is pretty rare. For those who don't know what you mean, the idea here is that a hash entry is good for two things: 1) Cutting off based upon the bound or score contained in it. 2) Remembering the best move in the position so you can search it first next iteration. This second reason is the most important one in almost every case. A lot of times, you are not going to *have* a best move to store in your hash element, since you fail low when you search this node. So in these cases, you get less benefit from hashing this node, and you might get more from leaving what was there before alone. Thompson's dual table method would make this optimization a little hard, since you write into your table in order to remember repetitions, but perhaps it can be salvaged. >The story: take 40 chess programs and you will notice 40 different hash >table behaviors. No kidding. bruce
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.