Computer Chess Club Archives


Search

Terms

Messages

Subject: More SEE results

Author: Stuart Cracraft

Date: 07:34:10 08/15/04

Go up one level in this thread


Okay I'll do one more test of SEE after this and then I'm done
with formal SEE testing, though I don't know if I can include it
without some depth condition. Seems to me it only works for
deep searches so perhaps I should just use it if depth > 6 or
something.

Test tests 300 positions with no see, see >= 0, delta/futility prune,
and both. There is slightly more depth with see>=0 (10.19 vs. 10.01.)

My next test will be to use see to sort capture moves by adding
the see score to the mvv/lva score. Likewise, I will not recalculate
the see score (since it was already calculated during the scoring
of the move within the addmove routine) when I use it in the quiescence.

This is why my see use within move sorting was not added optimally.
I was recalculating see in a 2nd place. Two calls to see for the same
position -- I'll make the new run for SEESORT tonight.

Since each additional item to test expands the total test time
(with NOSEE, SEEQUIESCE, DELTAPRUNE, SEEQUIESCE+SEEDELTA, I have
to test 4 things but adding SEESORT I'll have to test SEEQUIESCE+SEESORT
DELTAPRUNE+SEESORT, SEEQUIESCE+SEEDELTA+SEESORT.) So at least only
3 to test tonight instead of 4. Each run is about 2 hours (7100 seconds
4th column below) at 30 seconds per position.

Stuart

$ ./qaNbatch
HERALD ga nosee30300.log 30 300
**** 10.01/35.43 86% 259/300 7185.97 1762949376 5876498/24/245332 0/0/103274328/
0/0/0
HERALD ga -DSEEQUIESCE seequiesce30300.log 30 300
**** 10.19/31.80 87% 261/300 7153.01 1574976128 5249920/24/220184 0/0/51018488/0
/0/0
HERALD ga -DDELTAPRUNE seedelta30300.log 30 300
**** 10.00/35.40 86% 259/300 7188.23 1749075840 5830253/24/243325 0/0/102505680/
0/0/0
HERALD ga -DSEEQUIESCE -DDELTAPRUNE seeboth.log 30 300
**** 10.19/31.81 87% 261/300 7153.72 1575198848 5250663/24/220193 0/0/51030984/0
/0/0



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.