Author: Steven Edwards
Date: 20:25:42 03/21/04
Symbolic: First benchmark result
Although Symbolic does not yet have a planning facility, it does have a couple
of experimental mate recognizers for testing purposes. They are useful for the
automated running of EPD suites for stress testing with the hope of catching
various crashes and hangs while its author takes well deserved naps.
For BWTC, among the easiest of test suites, Symbolic running on a 400 MHz PPC
averages 2.6 seconds per position and gets:
Solved: four hundred thirty-seven
Unsolved: five hundred sixty-four
Solution rate: 0.436563
That's a rather poor showing compared to almost any other program, but remember
that there is no real planning or search. Yet. Also, the program is currently
configured to re-load both the ChessLisp interpreter and all the Lisp source
with each test position, and this takes a good chunk of the 2.6 seconds mean
time.
On the positive side, the fact that Symbolic can complete a long and varied
suite without any operational faults means that a very large percentage of the
code can be trusted as a foundation for further work.
I expect that completion of the initial pattern and planning tasks will take
about three to five months of calendar time, so there won't be much change in
benchmark results until then.
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.