Author: Gian-Carlo Pascutto
Date: 03:36:56 06/10/01
Go up one level in this thread
On June 10, 2001 at 04:55:37, Cheok Yan Cheng wrote: >>>If value<=7, ignore the value. (both windows) >>>If value>=100, return the value (both windows) >>>If value>7 AND value<100, re-search arghhhh...... (only Null windows) >> >>The point is that this 3rd one should not occur very often. > >But both searching can perform 1 and 2. Why we need PVS which will perform 3 >(The bad situation). Because the zero-window which PVS searches with produces more of 2, which is ideal since it gives an immediate cutoff. Additionally the cost of 3 is minimal because it will also be done with a small window, and the hashtables will help. >Consider the node child value with 9 12 101. Why? What would be the point? The move ordering at this node is totally wrong. It's already been stated multiple times that PVS will give WORSE performance in that situation. Lets turn the example around: 101 12 9 Which is a much more typical configuration. Now, behaviour at this level will be identical to alphabeta, so no performance loss due to researches. However, at the next level, for example for node '12', we would have gotten with plain A/B, a window of -inf, -101 (alpha=101 and beta=+inf at prev level) and with PVS, a window of -102, -101 (alpha=101 and beta=alpha+1) and again at the next level: plain A/B: 101, +inf PVS : 101, 102 Now with PVS any score over 101 will give an immediate beta cutoff, whereas plain A/B would have to search all nodes. -- GCP
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.