Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Searching less deterministically

Author: Charles Roberson

Date: 08:28:47 12/15/05

Go up one level in this thread



  The thought is to do this:

     If the value of best move is < Alpha but within
    a specified range of Alpha, toss a coin. If heads
    make current move best move and adjust best value and
    alpha. If tails, do as normal.

    Now, what does it mean to "change Alpha". I can think
    of multiple ways to do this. Remember: Alpha is not a
    global variable; it is local and stacked.

    My first thought is this will likely blow up the search
    tree which is exactly what it does in simulated annealing.
    However, the end result is great for simulated annealing due
    to the fact that you are happy to wait a little longer for a
    better answer. With Chess, we have such tight constraints on
    move time that this might be best for long term analysis mode.

    If Alpha is adjusted downward, then we immediately decrease the
    probability of pruning the rest of the branches at that node of
    the search tree. Thus, a bigger tree. But, we then run the chance
    of finding a "Great" move that prunes many branches that we would
    not have seen before.

     So, I am not suggesting the use of simulated annealing itself. Just
    utilizing a concept in simulated annealing.





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.