Author: James Swafford
Date: 20:20:15 04/03/99
Go up one level in this thread
On April 03, 1999 at 18:26:33, William Bryant wrote:
>I recently ran (for the first time) the LCTII test suite against my program in
>two forms, one with a PVS search and one without. The version without scored
>better in a few positions. At no point did the PVS code make it to a deeper ply
>in the 10 minute time frame than the non-PVS search.
>(Score of 2090 for PVS to 2120 without PVS).
>
>I do not have Null moves implemented yet.
>
>I know PVS is very sensitive to move ordering. I am sorting moves as follows.
>1. Hash Table Moves
>2. PV Move (if not the same as Hash table move)
>3. Captures and Promotions
>4. History moves
>5. All other moves
>
>I do not have killer moves added yet
A couple of things to think about:
Try placing the pv move first. After all, a previous search
determined that move was best. PVS works best in cases that
the best move is searched first, and all others are fail
quickly.
Try putting losing captures after the non captures.
>
>My questions are:
>1) Does null move pruning help PVS work better.
>
>2) Does this suggest a bug in my PVS code (listed below).
Your code looks alright, but why not try that through
the entire full width search?
>
>3) Am I missing something else?
>
>Thanks in advance,
>
>
>William
>wbryant@ix.netcom.com
>
>Note: this code is only in the search_Root function not in the search function.
>
>//****** PVS Search
>if (firstMove){ //following the pv so don't change anything
> x= -search(-beta, -alpha, depth-1);
> firstMove = false;
>}
>else{ //try a PVS search
> x= -search(-alpha-1, -alpha, depth - 1);
> if ((x>alpha) && (x<beta) && (!OutOfTime)) {
> gPVSResearch ++; //research
> x= -search(-beta,-alpha,depth-1);
> }
>}
>//******
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.