Author: Rick Bischoff
Date: 13:18:07 10/01/04
I re-did the hash table to use a two table approach, similar to what Bruce
Moreland describes --- and I am still getting very low hit% and very high node
counts on his test position:
[D]8/k/3p4/p2P1p2/P2P1P2/8/8/K7 w - - 0 1
Crafty takes around 40,000 nodes to get to finish depth 20:
20 102 73 39037 1. Kb2 Ka8 2. Kc3 Kb7 3. Kd2 Kc8 4. Kc2 Kb8 5.
Kb3 Ka7 6. Kc4 Kb6 7. Kd3 Kc7 8. Ke2 Kd7 9. Kf2 Ke7 10. Ke3 Kf6
time=0.76 cpu=78% mat=1 n=39037 fh=86% nps=51K
ext-> chk=1 cap=0 pp=0 1rep=0 mate=0
predicted=0 nodes=39037 evals=4109 50move=0
endgame tablebase-> probes=0 hits=0
hashing-> 68%(raw) 64%(depth) 1%(sat) 99%(pawn)
hashing-> 1%(exact) 49%(lower) 6%(upper)
SMP-> split=260 stop=30 data=5/64 cpu=0.60 elap=0.76
My engine takes...
20 188 19463 19301771 1. Ka2 Kb7 2. Kb2 Ka7 3. Kb3 Ka6 4. Kc2 Kb6 5. Kd2
Kc7 6. Kd3
***** 194645 ms elapsed 19303206 nodes 99171.3 nps.
table_probes={5783659,6767435} table_hits{5769439,6177910}
table_goodhits{2241046,2777128} table_exacthits{24695,578507}
table_lower{5662,293465} table_upper{5325,111804}
table_stores{18234707,13963792}. good hits: %42.0024 exact hits: %5.04884 lower
hits: %2.50371 upper hits: %0.980376 saturation: Bank#0 { 0.247, 0.247}
Bank#1 { 0.23825, 0.23825} Bank#2 { 0.25125, 0.25125} Bank#3 { 0.2415,
0.2415} Bank#4 { 0.23475, 0.23475} Bank#5 { 0.24125, 0.24125} Bank#6 {
0.25175, 0.25175} Bank#7 { 0.2495, 0.2495} key collisions storing in deep
table: 503054
19 million nodes!
Ignoring the wrong moves from both engines (Kb1, crafty finds it at depth 21),
what could possibly explain my low hit rates-- Crafty is getting 1% exact, 49%
lower and 6% upper, while I am getting %5 exact, %2.5 lower and %0.9 upper!??
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.