Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: QSearch() as PVS() ?

Author: Anthony Cozzie

Date: 07:50:53 01/15/04

Go up one level in this thread


On January 15, 2004 at 07:48:42, Matthias Gemuh wrote:

>On January 15, 2004 at 06:59:15, Volker Böhm wrote:
>
>>On January 14, 2004 at 15:44:56, Matthias Gemuh wrote:
>>
>>>
>>>
>>>Does it make sense to use PVS algorithm to implement QSearch ?
>>>It does not seem to help my engine.
>>>
>>>/Matthias.
>>
>>PVS is good if
>>
>>1. there are many alternative nodes (never got good results in my connect-4 game
>>with 7 moves max. at one ply)
>>2. the move ordering is good
>>3. you can narrow the window greatly compared with the usually arising
>>position-values in a search.
>>
>>In QSearch:
>>
>>1. Depends a little bit what nodes you are searching. If you only search "good"
>>hits selected by SEE its verry narrow. If you are searching check-moves too it´s
>>wider. You can consider PVS if searching check-moves tool.
>>2. Move ordering in QSearch is seldom good. You haven´t got a PV and seldom
>>usefull hash-moves. If you turn off SEE for "good hit-moves" the ordering is too
>>bad for PVS
>>3. I´am staring a new search with a windows of -30cp .. +30cp around the search
>>result of iteration-1. The window in QSearch is often of witdh 0, but verry
>>seldom of width > 60cp (only if I´ve got fail-high or fail-low against the
>>initial window). The results of a QSearch differ greatly as they are hitting
>>material. The least effective hit move gets +100cp. Conclusion: the window is
>>allready verry narrow compared to the results of QSearch calls.
>>
>>Conclusion:
>>Don´t use PVS in QSearch.
>>
>>Greetings Mangar
>
>
>
>
>Thanks for this elaborate answer.
>BTW, in QSearch I also generate moves for
>
>1) all passed pawns
>2) all attacked pieces moving to unattacked empty squares
>
>/Matthias.

OK, this seems very strange to me.

If your eval is < alpha, how is moving an attacked piece going to increase it?

The only way this makes sense is to never stand pat in q search.  Tord's idea
_sort of_ makes sense (I have seen similar horizon problems in Zappa) but this
will make your qsearch 99.9% of all nodes!

anthony



This page took 0.01 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.