Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: hash numbers requested: authors please read

Author: James Swafford

Date: 04:36:01 09/18/02

Go up one level in this thread


On September 18, 2002 at 04:16:28, Andrew Williams wrote:

>On September 17, 2002 at 21:31:17, James Swafford wrote:
>

Your numbers seem high relative to everyone else... but I
suppose that's because you're constantly searching with a narrow
search window?

I haven't implemented PVS or MTD(f) yet, so maybe I'm not doing
so bad afterall.  The smallest window I'm searching with now
(from the root) is +/- half a pawn on each side of the last
alpha score.

Thanks Andrew!

--
James


>Hi James,
>
>Here are some results for postmodernist:
>
>Position 1:
>===========
>NODES: ab: 605505       q: 915527       tot: 1521032
>probes: 1491339 hits: 315293    cutoffs: 210841 secondary: 27025
>probes: 98%     hits: 21%       cutoffs: 66%    secondary: 8%
>
>
>Position 2:
>===========
>NODES: ab: 1760510      q: 766490       tot: 2527000
>probes: 2472152 hits: 1502633   cutoffs: 1200327        secondary: 147757
>probes: 97%     hits: 60%       cutoffs: 79%    secondary: 9%
>
>
>Position 3:
>===========
>NODES: ab: 508192       q: 1037811      tot: 1546003
>probes: 1543924 hits: 265514    cutoffs: 163079 secondary: 20231
>probes: 99%     hits: 17%       cutoffs: 61%    secondary: 7%
>
>
>The percentages are (L to R):
>
>% of nodes for which PM probed the table (PM probes in qsearch too).
>% of probes for which PM finds a matching entry
>% of matching entries which cause a cutoff based on bounds in the entry
>% of matching entries where the bestMove in the entry causes a cutoff
>
>
>
>Andrew
>
>
>
>>
>>I'm skeptical about my hash efficiency - it seems to me I'm not
>>getting enough hits from my hash tables.
>>
>>I wonder if some of you would post some numbers for me to compare
>>by running the following 3 positions:
>>
>>r4k2/q4npp/P2P4/2p1R3/2N5/6PP/Q5K1/8 w - - bm Re7; id "ECM.1028";
>>7k/p7/1R5K/6r1/6p1/6P1/8/8 w - - b6b7;
>>r1bq1rk1/pp2bp1p/2p2np1/3p2B1/3P4/2NQ2N1/PPP2PPP/4RRK1 w - - bm Rxe7; id
>>"ECM.1068";
>>
>>The first and last are from GCP's ECM-GCP suite and the second is
>>from WAC.
>>
>>Anyway, here's what I'm getting (10 seconds per problem):
>>
>>Position 1:  3931063 nodes,  318411 interior nodes.
>>          fail highs: 10866 (3%); fl: 9929 (3%); ex: 671 (0%)
>>          hash mvs played: 10660
>>
>>Position 2:  4944963 nodes,  3102047 interior nodes.
>>          fh: 1531962 (49%); fl: 432594 (13%); ex: 2836 (0%)
>>          hash mvs played: 602039
>>
>>Position 3: 3847265 nodes,  222358 interior nodes.
>>          fh: 3443 (1%); fl: 2926 (1%); ex: 151 (0%)
>>          hash mvs played: 11931
>>
>>The ratios change very little if I run the positions at 30 seconds.
>>
>>Thanks to anyone that takes the time... (which is hopefully a
>>positive number of people)...
>>
>>--
>>James



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.