Author: Dann Corbit
Date: 12:14:55 07/26/04
Go up one level in this thread
On July 24, 2004 at 15:21:22, Gerd Isenberg wrote: >On July 23, 2004 at 19:59:03, Dann Corbit wrote: > >>Similar to what I like to do. >>Use the linear "select top N" function I posted in the "find min" thread to >>collect the top k entries. (I like k=4 or 5). Then I use a special function to >>sort the top k items using minimal comparisons and exchanges (a tiny fixed >>number of operations). >>The other N-k entries can remain unsorted, or you can repeat the process on the >>N-k entries if you never get a big fail-high (if you like.) >> >>Expected time: O(n), with top 4 or 5 items in proper sort order. > >Good idea! If first five moves don't fail high in cut-nodes, how high is the >probability that a remaining move fails high or that the cut-node becomes an >all-node? I don't think it really makes much difference. Almost 100% one of the top 5 will fail high. I imagine problems may arise in zugzwang situations but I have not measured it. >Is it worth to keep track of pv/cut/all-nodes (pv is most left, cut-nodes are >siblings of pv-nodes or child of all-nodes, all-nodes child of cut-nodes) and to >distinguish them in scoring and getting? Anything you can do to improve the guess of the value of a node will be important, I think. And especially anything you have already computed. >What i actually do in my 32-bit program is generating moves with a kind of >finite state machine. So i first try to only generate hash move, obvious winning >captures, killers and safe checks. If no such move (if any) results in an >cutoff, i start to generate other moves and score and pick them depending on >eval versus alpha. > >That works fine, but is complicated and has the disadvantage to keep some >additional data, such as state of movegen and bookholding-information in each >node. It seems to me that if you get it right, it will work just the same as my idea. Probably, my method is as complicated as your since I do everything incrementally (including eval). So I store lots of state information also.
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.