Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Zero window question ?

Author: Nagendra Singh Tomar

Date: 09:27:49 10/25/02

Go up one level in this thread


On October 25, 2002 at 11:16:48, Robert Hyatt wrote:

>On October 25, 2002 at 08:39:47, Nagendra Singh Tomar wrote:
>
>>Hi everybody!
>>
>>In my endeavour to understand the techniques of comp chess I am in the process
>>of understanding the benefits (what/why/how) of a zero window search. Told
>>superficially it can be said that since we have a very small window to search we
>>search less nodes, but I want to understand it to the minute details.
>>
>>What I observe and understand is follows.
>>
>>score = -alphabeta(board, -beta, -alpha, depth-1) <-- 1st call (full width)
>>.
>>. update alpha if score more than alpha
>>.
>>score = -alphabeta(board, -alpha-1, -alpha, depth-1) <-- subsequent calls
>>                                                         zero window
>>
>>Any saving whatsoever that can be achieved by reducing the window size, should
>>be because of beta cutoffs (fail-highs) increasing.
>
>
>Not quite.  Because you are passing that null-window to the next ply, _all_ of
>this sub-tree
>will be searched with a null-window.  This means that every search will either
>fail high or
>fail low, which is about as efficient as you can get.

But for fail-low we anyway have to search all the nodes. fail-high is the actual
saver.
Is;nt it ??

>
>>Here we are keeping beta as -alpha (same as we would have kept it for a full
>>window search, though we have kept alpha as -alpha-1 in contrast to -beta for a
>>full win search)
>>But in the subsequent call (child nodes) this alpha (-alpha-1) will be negated
>>and passed as beta, which will be very less. Is the saving in nodes searched,
>>due to this beta being passed as a very small value in the child->child nodes
>>searched.
>>I know I have made it complex, but I hope you understand me.
>>
>>IOW changing the value of alpha cannot reduce the number of nodes searched. But
>>the fact that this alpha will be passed as beta in the subsequent call to
>>alphabeta,   results in lot of beta cuts and hence reduced nodes searched.
>>
>
>Correct.
>
>
>>
>>Am I thinking on the right lines ?
>>
>>tomar



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.