Author: Ralph E. Carter
Date: 05:16:52 11/15/98
I understand the need to have a statistically significant sample size at the end of the critical line. Unfortunately, it appears necessary to ensure the sample size in EVERY branch examined. This damages the useability of this great tool. Its analysis can be trusted only if you KNOW that every leaf had at least this sample size. So you can't know to what depth you can reasonably search. Another parameter is needed: "minimum sample size at each leaf", either when creating the tree, or when calculating the critical line. P.S. You can work around this by manually pruning off the infrequently played lines. But all the while, your results using "Critical Line" remain problematical. Assuming your database consists of games played by high rated players only, I think the most reliable statistic on a move is "N". (The REASONS why more GMs play a move, even though its scoring percentage might be a little lower, are found with a deeper study.) P.P.S. I was stuck with a permanent unwillingness to make the decisions required to build a repertoire. I bought some powerful software tools... I can trust them while eliminating the coarse choices... But the finer judgements are still a matter of GM opinion. Fortunately for our game, the solution to chess is not available yet. P.P.P.S. For building a repertoire, more effective than this attempt at basic research, would be to just pick a player you like, and make a repertoire database from his games (while watching the sample size!)
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.