Author: Michael Neish
Date: 20:30:06 11/13/00
Hi, Well I'd like to ask people what they think of this idea. If when searching a position the number of beta cutoffs caused by captures is much higher than that of non-captures, this might give an indication of the volatility of the position. Conversely, a higher proportion than average of non-capture beta cutoffs (where the "average" proportion is to be determined by tests, etc) might indicate a fairly quiet position. I'm not sure the above is correct, but it seems to be so when I test my program in a few varied positions -- far from an exhaustive analysis, I know. My suggestion is that you keep a running total of capture and non-capture beta cutoffs, and if after searching a few ply you determine that the position is highly volatile, you switch to a more tactical evaluation function (i.e., material gain, attack on the King, creation of an advanced passed Pawn) and ignore other positional factors. The idea is that you look for some near-term gain/loss and drop out some factors which might slow down the evaluation. You might even be able to gain a ply ... ? Conversely, if the positions is quiet, you use a slower but positionally more accurate evaluation. You might lose a ply in search depth, but you might gain something in accuracy. I'm not saying any the above is correct. It might be totally wrong or it might be correct with some refinements. Does anyone have any comments? Cheers, Mike.
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.