Author: Paul Byrne
Date: 18:40:28 01/25/04
I have computed (hopefully) the number of unique position at 9 ply from the
initial position to add to earlier results. This is basically taking the
positions counted in perft 9 and seeing how many unique positions there are.
En passant file is not set unless there is a legal ep capture on the next ply.
ply positions branch size bits
1 20 - - -
2 400 20.00 - -
3 5,362 13.41 0.28 KB 0.43
4 72,078 13.44 4.34 KB 0.49
5 822,518 11.41 77.72 KB 0.77
6 9,417,681 11.45 0.95 MB 0.85
7 96,400,068 10.24 14.04 MB 1.22
8 988,187,354 10.25 154.59 MB 1.32
9 9,183,421,888 9.29 1.77 GB 1.66
"branch" is just the effective branching factor from ply-to-ply.
"size" is the size of the file containing the positions (which can then be used
to compute the next ply).
"bits" is the number of bits/position in the file.
Comparing with the perft's from the same ply:
ply perft branch %age gain
1 20 - 100.00 -
2 400 20.00 100.00 1.00
3 8,902 22.25 60.23 1.66
4 197,281 22.16 36.53 1.65
5 4,865,609 24.66 16.90 2.16
6 119,060,324 24.47 7.91 2.14
7 3,195,901,860 26.84 3.02 2.62
8 84,998,978,956 26.60 1.16 2.59
9 2,439,530,234,167 28.70 0.38 3.09
"branch" is the branching factor.
"%age" is the percent of unique positions.
"gain" is the divisor for the percentages. For example, using the uniq 9
positions to compute perft 11 should be roughly 3.09 times faster than
doing it with the uniq 8 positions. (I didn't actually compute the weights
for uniq 9 though -- lack of disk space -- so the data I have is not enough
to compute the perft this way).
Computing the unique positions for plys 1 to 9 took about 4 days on a 2400+
athlon. I estimate that if I had about 100 GB of disk space for scratch files,
uniq 10 would take about a month to compute -- somewhat more if a file with the
positions was also computed in order to then do uniq 11. Maybe some other time.
:) Back to doing some real chess programming for me...
-paul
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.