--
-Original Message-
From: [EMAIL PROTECTED]
To: computer-go@computer-go.org
Sent: Wed, 3 Dec 2008 12:56 am
Subject: Re: [computer-go] Monte-Carlo Tree Search reference bot
Hmm.. it could be that N is picked randomly each time... now I can't seem to
find the descripti
On Wed, 2008-12-03 at 11:24 -0200, Mark Boon wrote:
> If this is because I use Java, then
> Don's concise C implementation of the MC-AMAF bot should be a lot
> faster than my bloated Java version.
I don't remember the numbers, but my own java and C implementation were
written in the same style
I had a chess program years ago that was blindingly fast on some
computers, very slow on others. It was all about the cache. The move
generator was hard coded for each piece on each square. For instance a
white pawn on d7 had it's very own move specialized move generator.
There was a function
On Wed, Dec 03, 2008 at 11:24:22AM -0200, Mark Boon wrote:
>
> Heikki wrote:
> >One thing to remember is that more memory use means more cache misses,
> >and more access to the main memory. On modern computers, those can cost
> >as much as executing a thousand instructions! So memory optimizing c
On 3-dec-08, at 10:31, Heikki Levanto wrote:
Having said that, I can't help responding to one detail:
I had seen people write about memory usage of the tree, but never
understood the concerns.
One thing to remember is that more memory use means more cache
misses, and
more access to the m
On Wed, Dec 03, 2008 at 09:55:07AM -0200, Mark Boon wrote:
> I thought about that, but I was afraid the code would become too
> obscure. After all, this is supposed to be a reference
> implementation. But maybe I should actually give it a try to see what
> it would look like.
I agree that th
On 3-dec-08, at 06:09, Sylvain Gelly wrote:
What I did (was a "long" time ago, I don't know if it is still used in
Mogo), is to compute the m best moves every so often and most of the
time just do the max over those m moves.
m was on the order of 5, and "every so often" was an increasing
functi
On Wed, Dec 3, 2008 at 05:17, Mark Boon <[EMAIL PROTECTED]> wrote:
> I had set myself as an arbitrary goal that it should do at least 20K
> playouts. But with real liberties, AMAF and a RAVE formula I got stuck in
> the 16K-17K range. According to my profiler that is mostly due to the
> expensive f
ove without having to calculate what the move would
> be.
>
> - Dave Hillis
>
> -Original Message-
> From: Mark Boon <[EMAIL PROTECTED]>
> To: computer-go
> Sent: Tue, 2 Dec 2008 11:17 pm
> Subject: Re: [computer-go] Monte-Carlo Tree Search reference bot
>
>
ROTECTED]>
To: computer-go
Sent: Wed, 3 Dec 2008 12:14 am
Subject: Re: [computer-go] Monte-Carlo Tree Search reference bot
That's going to repeat the same exact path through the tree three times, isn't
it? If so, it seems like it would be more efficient to do N playouts from the
leaf a
e what the move would be.
- Dave Hillis
-Original Message-
From: Mark Boon <[EMAIL PROTECTED]>
To: computer-go
Sent: Tue, 2 Dec 2008 11:17 pm
Subject: Re: [computer-go] Monte-Carlo Tree Search reference bot
I have made some minor performance improvements and this is as far as I
int
10 or so), just
repeat that move without having to calculate what the move would be.
- Dave Hillis
-Original Message-
From: Mark Boon <[EMAIL PROTECTED]>
To: computer-go
Sent: Tue, 2 Dec 2008 11:17 pm
Subject: Re: [computer-go] Monte-Carlo Tree Search reference bot
I have mad
I have made some minor performance improvements and this is as far as
I intend to take this particular project. I might make some small
changes if necessary, but most likely I'll leave this largely
unchanged from now.
I had set myself as an arbitrary goal that it should do at least 20K
pl
he most usefull part of the standard ref bot :
because a lot of persons have tried it, you can be sure of the
exact link
between the software, and the experimental results. It makes
assertions
reproducibles, and that's really great.
> From: [EMAIL PROTECTED]
> Subject: Re: [compute
re of the exact link
between the software, and the experimental results. It makes assertions
reproducibles, and that's really great.
> From: [EMAIL PROTECTED]
> Subject: Re: [computer-go] Monte-Carlo Tree Search reference bot
> Date: Fri, 28 Nov 2008 01:48:09 -0200
> To: comput
On 27-nov-08, at 19:50, Denis fidaali wrote:
So, you use AMAF for "simulating" the first UCT evaluations ?
I though the classical way to use AMAF, was to affect only the
win/lose ratio portion of the uct equation.
Obvioulsy it should be allowed to use an arbitrary
large number of AMAF simulatio
So, you use AMAF for "simulating" the first UCT evaluations ?
I though the classical way to use AMAF, was to affect only the
win/lose ratio portion of the uct equation.
Obvioulsy it should be allowed to use an arbitrary
large number of AMAF simulation accumulating them longer
than what it take to
17 matches
Mail list logo
