[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [gnugo-devel] TODO revisions
From: |
David G Doshay |
Subject: |
Re: [gnugo-devel] TODO revisions |
Date: |
Wed, 25 Aug 2004 22:16:18 -0700 |
In some rough sense we are doing something like these 3 things.
Cheers,
David
On Aug 25, 2004, at 7:26 PM, Gunnar Farnebäck wrote:
* Making the engine use many machines loosely connected on the
internet or in a cluster.
After all the discussion on the SlugGo/MFG thread, I am sure that it is
clear that we are doing the opposite: our cluster uses the engine. The
difference is that we are not parallelizing the engine itself, but
running
multiple instances of the engine. But our cluster is dedicated, not
loosely
connected.
As per my previous offer, if anyone wants to think about how to
parallelize the engine we may be able to offer part of our cluster
for testing and development. We are open to suggestions and
proposals.
- * Think on the opponents time.
+ * Think on the opponent's time.
We will begin doing this when we complete our hash table work.
There will be a number of agents dedicated to extending the
leaf nodes of the lookahead paths. These agents will likely be
running all the time, without regard for who's turn it is.
* A global alpha-beta reader. This would probably be very slow and
could only read 2 or 3 moves ahead. Still it could find fatal
errors and improve the moves that GNU Go makes.
Our results so far indicate that 2 moves makes almost no difference,
that the difference increases for 4, 8, and 10 move lookahead, but
that a lookahead depth of 12 is not much different than a lookahead
of 4. A lookahead of 16 is clearly stronger. We are still collecting the
data to verify this result and narrow the error bars.
David