[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[gnugo-devel] About 3.4

From: bump
Subject: [gnugo-devel] About 3.4
Date: Tue, 22 Jul 2003 09:40:04 -0700

As far as I can see the biggest objection to 3.3.23 as a
candidate for 3.4 is the break in code. The problems:

(1) It makes the engine slower by at least 20 %, maybe a
little more.

Stephane suggested that I test Arend's pending patch
arend_3_23.1a and see if it gives a speedup. I haven't
done that yet but will do so.

(2) It gives no statistical improvement in gnugo/gnugo matches.

It does give regression improvements and Arend believes that
it makes the engine stronger, perhaps by 1 stone.

(3) It is difficult to understand how it works. I explained in:

how to get sgf traces of the functions break_in and block_off.
These traces show many bizarre moves being tried. This is not
necessarily bad since the same is true of the tactical
reading code, which is very reliable.

But at the very least I would like to see get traces of
examples where the code works correctly. I have been unable
to confirm that it does what Arend described in the case of
the file break_in.sgf.

I do think we should use it in Gifu. I don't see any reason
to wait before making a zip file containing a gnugo-3.3.23
binary and supporting programs to supply to Gifu.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]