Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: mtd, pvs, etc.

Author: Antonio Dieguez

Date: 17:31:26 11/15/01

Go up one level in this thread


On November 15, 2001 at 17:59:55, Robert Hyatt wrote:

>On November 15, 2001 at 16:47:57, Antonio Dieguez wrote:
>
>>I would like to know why they work for you (if).
>>
>>I have tried doing a null window search in the first root move with (prev_score,
>>prev_score+1), to decide if to research up or down. But that is not better than
>>just a (prev_score-window,prev_score+window), is that normal? I mean, if not
>>even that works well for me why should I think MTD will?
>>What do you get if you do the first thing mentioned?
>>
>>And do many null window searches in the tree have never worked very well for me.
>>No more than 1% or 2% gain. Even only using them when there are ht move of exact
>>score and the window is very big I have that 1% or something. I have heard PVS
>>is a lot better than alphabeta(with small and nulls windows in the root of
>>cors), yea sure...
>>
>>What difference you get using pvs or using alphabeta with aspiration search?
>
>I use both PVS _and_ aspiration.  I set an aspiration window at the root
>before I call Search(), and I use PVS inside the search everywhere.  PVS will
>generally reduce the tree size by around 10%.  Less if the program changes its
>mind at the root a lot since it has to re-search those things twice.  It might
>be more than 10% for positions where you _never_ change your mind or fail high.

oh yes yes, I meant that, pvs with aspiration at the same time.

I think amyan changes his mind much often :( anyway 10% does not excite me too
much :)

Be well.



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.