[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[gnugo-devel] Test results
From: |
Daniel Bump |
Subject: |
[gnugo-devel] Test results |
Date: |
Mon, 4 Feb 2002 06:52:29 -0800 |
As I mentioned in a previous message when I updated
the test results, I used results configured with
--enable-experimental-connections --enable-alternate-connections.
Although this was unintentional I think it was probably
a good thing. There were two errors in the test results
so I revised the results in two files, global.tst and
nngs.tst.
I also added test results to BREAKAGE without the
experimental and alternate connection options. If
we exclude connection.tst and connect.tst where
the new connection options make a huge improvement,
they give a net gain of 3 tests. The use of the
new connection code by the move generation is
currently very limited.
Anyone who wants more information should check
out the current CVS and consult BREAKAGE.
I also fixed a typo in ChangeLog reported by
Arend and moved the CVS tag rel-3-1-24 in
four files, ChangeLog, BREAKAGE, nngs.tst and
global.tst. So the CVS release differs very
slightly from the tar files distributed though
they make the same engine.
About speed, the experimental connection code
makes no measurable difference in speed for
genmove tests. I expect this could change
when the engine starts calling that code
more intensively. I ran two tests,
strategy.tst and nngs.tst and timed them
on my 433Mh Celeron. Without the experimental
options, it took 5610 seconds. With the
experimental options, it took 5592 seconds,
actually faster though not measurably so.
However use of the new connection code by
genmove is currently slight so this is to
be expected.
Dan