I have empirical evidence that mwRefbot level 4096 (the Michael Williams
enhanced reference bot at 4096 playouts) is stronger than the reference
bot at the same (and even higher levels.)
Rank Name Elo+- games score oppo. draws
1 mwRefbot-004096 2684 23 22 1529
On Mon, 2008-10-27 at 23:23 -0400, Luke Gustafson wrote:
> Computer Scrabble significantly exceeds humans. A basic monte carlo search
> and an endgame solver is very effective. There is probably still much
> strength to be gained (very little opponent modeling is done), but it's
> already so s
What do those number mean exactly? Are those winning percentages
against your ref-bot with 1,000 playouts?
I'm not getting the same results as Michael does. I see some
improvement, but not nearly as much.
I had to make a few changes however, as I was adding up the results
in a different w
David Fotland wrote:
> I've thought about using mcts for Arimaa, and it might be better than
> alpha-beta. Arimaa has a huge branching factor, and since the pieces move
> slowly, human long-term planning works well against computers. Mcts prunes
> better, so it should get deeper PVs. AMAF might
My original chart of numbers was created before it occurred to me that maybe the check whether the move has been played before is not needed. It is of course
possible that the check IS needed. I ran a new test after removing that check and the numbers don't look as good:
At1 playouts per m
On Tue, 2008-10-28 at 12:48 -0200, Mark Boon wrote:
> What do those number mean exactly? Are those winning percentages
> against your ref-bot with 1,000 playouts?
No, it is a massive tournament where each player is scheduled in a
manner similar to CGOS.The number of playouts is identified by
Hehe, this reminds me of the title of a CNN program: 'Keeping them
honest' ;-)
OK, let me add that check back in to see if I get better results.
Mark
On 28-okt-08, at 13:18, Michael Williams wrote:
My original chart of numbers was created before it occurred to me
that maybe the che
If you look at my original email, I did not mention removing that until after I showed and discussed the numbers. And even then, I said "... it is probably
safe to remove...". Probably being the operative word indicating that I had not yet tested that.
Mark Boon wrote:
Hehe, this reminds me o
You can add the check back in and be more efficient than Don's nested loop by
doing something like this (warning: untested code)...
for (int mv = 0; mv < NNN; mv++)
{
credit[mv] = true; // initial assumption is that credit is awarded
for any move
}
On Tue, 2008-10-28 at 12:48 -0200, Mark Boon wrote:
> What do those number mean exactly? Are those winning percentages
> against your ref-bot with 1,000 playouts?
>
> I'm not getting the same results as Michael does. I see some
> improvement, but not nearly as much.
>
> I had to make a few ch
MCTS seems to be not very useful for some games, but I think it needs a
lot more research.
I tried it will chess with really dismal results. Even the mighty
Rybka team has a MC mode which is a curiosity, it plays much weaker than
the standard program.
I have this feeling that you ought to be ab
On Tue, 2008-10-28 at 11:18 -0400, Michael Williams wrote:
> Don, did you retain the check in your version that has the weighted
> AMAF?
No, I eliminated the check so that I could give an apples to apples
assessment of the idea exactly as Michael Williams suggested.
> What ever you did, could y
Don,
Don't get me wrong, I understand very well where you're coming from.
It's very handy to have a concise defintion of a minimalist playing
program that can be used to check correctness. The main reason I
thought Michael's idea was worth considering was that it was both
simpler AND play
On Tue, 2008-10-28 at 11:40 -0400, Michael Williams wrote:
> If you look at my original email, I did not mention removing that
> until after I showed and discussed the numbers. And even then, I said
> "... it is probably
> safe to remove...". Probably being the operative word indicating that
> I
On Tue, 2008-10-28 at 14:08 -0200, Mark Boon wrote:
> Don,
>
> Don't get me wrong, I understand very well where you're coming from.
> It's very handy to have a concise defintion of a minimalist playing
> program that can be used to check correctness. The main reason I
> thought Michael's ide
Yes, this is the hash table idea I mentioned, although most people don't
view this as a hash table - it is.
The enhanced version has actually improved it's result over the past
couple of hours:
Rank Name Elo+- games score oppo. draws
1 mwRefbot-004096 2691 22 21 164
I ran a quick test of the check in question combined with the decaying
playouts. The one WITH the check won 70% at 2000 playouts over a 200 game
match.
Don Dailey wrote:
On Tue, 2008-10-28 at 14:08 -0200, Mark Boon wrote:
Don,
Don't get me wrong, I understand very well where you're coming
I shouldn't have gone on memory. It was actually 65.9% over 183 games.
Michael Williams wrote:
I ran a quick test of the check in question combined with the decaying
playouts. The one WITH the check won 70% at 2000 playouts over a 200
game match.
Don Dailey wrote:
On Tue, 2008-10-28 at 14
On Tue, 2008-10-28 at 12:22 -0400, Michael Williams wrote:
> I ran a quick test of the check in question combined with the decaying
> playouts. The one WITH the check won 70% at 2000 playouts over a 200 game
> match.
I just queried my results at 4096 playouts and get very different
numbers.
As promised I tested this on the C reference bot. I get almost a 12%
speedup - 36.0 seconds versus 40.2
- Don
On Tue, 2008-10-28 at 11:41 -0400, Michael Williams wrote:
> You can add the check back in and be more efficient than Don's nested loop by
> doing something like this (warning: unte
"The enemy's key point is my own" is often invoked, for example, as a
reason to occupy the central point of a _nakade_ shape, or to play a
double sente point, or to make an extension that would also be an
extension for the opponent.
I would like now to talk about it in the context of the potentia
It appears from testing 3 of us are doing, that the Mark Williams
enhancement is good, and that testing to see if the move was played
before is VERY GOOD when combined with the enhancement.
The enhancement without the checking is perhaps 40 ELO stronger but with
the testing it appears to be abo
It's worth noting that the standard ref bot also has the first move test.
Don Dailey wrote:
It appears from testing 3 of us are doing, that the Mark Williams
enhancement is good, and that testing to see if the move was played
before is VERY GOOD when combined with the enhancement.
The enha
On Tue, 2008-10-28 at 15:07 -0400, Michael Williams wrote:
> It's worth noting that the standard ref bot also has the first move test.
Yes, so presumably even the weakest version is doing it the "best" way,
i.e. doing the check.
- Don
>
>
> Don Dailey wrote:
> > It appears from testing 3
> From: Richard Brown <[EMAIL PROTECTED]>
>
> "The enemy's key point is my own" is often invoked, for example, as a
> reason to occupy the central point of a _nakade_ shape, or to play a
> double sente point, or to make an extension that would also be an
> extension for the opponent.
> Not on
On 28, Oct 2008, at 11:23 AM, Richard Brown wrote:
... if there is someone who can explain to me why I have the
nagging suspicion that a differential equation is involved here, ...
I cannot tell why you have this nagging suspicion, but I can say that
if you wish to look into it deeper you can
On Tue, 2008-10-28 at 12:28 -0700, terry mcintyre wrote:
> The downside of overfitting to a particular opponent is that little
> improvement versus other opponents was seen.
I wonder what would happen if Sluggo used a second player (with a style
much different from gnugo but still similar streng
You should also test your 7/8 keep heuristic to see if it's reducing
weight or avoiding the final moves that matters.
Sent from my iPhone
On Oct 28, 2008, at 3:03 PM, Don Dailey <[EMAIL PROTECTED]> wrote:
It appears from testing 3 of us are doing, that the Mark Williams
enhancement is good,
On 28, Oct 2008, at 12:28 PM, terry mcintyre wrote:
Sluggo has ( or had ) a particularly nasty form of "the enemy's key
point is my own" - the program actually ran the GnuGo engine, so
Sluggo knew precisely where GnuGo was most likely to play, and
( using a large cluster ) could give GnuGo
On Tue, 2008-10-28 at 15:34 -0400, Jason House wrote:
> You should also test your 7/8 keep heuristic to see if it's reducing
> weight or avoiding the final moves that matters.
I'm now also testing the reference bot that does NOT check for already
played moves and it's at least 100 ELO weaker tha
David Doshay wrote:
> One nasty form of "The Enemy's Key Point Is My Own" was the "reverse
> monkey jump," where SlugGo would properly recognize that the opponent's
> best move against it was a monkey jump, and properly see that stopping
> that monkey jump was the best move, but it would then play
SlugGo was never intended to just be the multi-headed global
lookahead on top of GNU Go that it is today. The idea has always
been to have multiple go engines inside. We just picked GNU Go
for the first because when we started that was the only decent
open source program, so we built the infrastru
It makes me wonder if there are other classes of moves that can be segregated
as AMAF-worthy and AMAF-unworthy.
For instance, maybe capture moves played in the playout are only worth treating
as first when the enemy did not add any stones to the chain that got captured.
Of course, you can't do a
On Tue, 2008-10-28 at 15:34 -0400, Jason House wrote:
> You should also test your 7/8 keep heuristic to see if it's reducing
> weight or avoiding the final moves that matters.
Yes, there is some overlap. The 7/8 rule avoids those cleanup moves at
the end that are probably useless for choosing m
On Tue, 2008-10-28 at 16:37 -0400, Michael Williams wrote:
> It makes me wonder if there are other classes of moves that can be segregated
> as AMAF-worthy and AMAF-unworthy.
>
> For instance, maybe capture moves played in the playout are only worth
> treating
> as first when the enemy did not a
This testing would be ideal for a boinc-style network of cooperative computers.
It isn't time-critical, and does not require a lot of information exchange;
each computer would run a series of multi-way tournaments, and the results
would be collected.
I'd be happy to volunteer part of my deskt
>It's pretty tricky doing anything very clever without slowing things
>down a lot. But I am highly interested because I want to build a
>stronger Ogo type of program for handheld devices.
Here's a suggestion: some mobile devices have volume-bound
communication tariffs (internet or text messaging
The idea this time is to make use of your opponent's moves (as the proverb says). I tried a few different weights for the opponent move and 0.1 was the best of
the ones I tried. The win rates for the version with opponent moves versus the version without opponent moves are as follows (both versi
Some interesting ideas Michael, keep them coming.
With regards to the code-snippet you posted, this kind of
'mark-to-prevent-double-work' type of construction is very common in
computer-Go. Looping over the complete array to set the initial value
is very expensive and totally avoidable.
The same
The marker idea I call aging. I call the variable age because it seems
to be more descriptive to me. Each playout has a different age
associated with it. The idea is used in persistent hash tables in
computer chess and I'm sure in other games too.
However, I seriously doubt this particular en
Seems Ito-sensen is very busy...
The committee prepares a very limited number of volunteer operators.
See "4-3 Volunteer Operators" at
http://jsb.cs.uec.ac.jp/~igo/2008/eng/sankayouken.html for detail.
Hideki
David Fotland: <[EMAIL PROTECTED]>:
>Do we have to show up in person, or can our pro
Seems Ito-sensei is very busy...
Remote computing is allowed. See "1-4: Using a Remote Host via the
Internet" at http://jsb.cs.uec.ac.jp/~igo/2008/eng/sankayouken.html
("Requirements for participation" in the side menu) for detail.
Hideki
David Doshay: <[EMAIL PROTECTED]>:
>Will remote comput
The protocol used is described as a "modified NNGS" protocol. The paragraph on
remote computing seems to suggest that no direct connection between the
Internet and the tournament is permitted; that an operator must manually
transfer moves from the remote computer to a GUI which is connected to t
This is wonderful news for SlugGo, but unfortunately my wife just
booked a
vacation for us at the same time. I will have to wait until next time
to be
able to attend.
I will plan to hold this time open next year and hope that you will be
holding
the tournament again at that time.
Thank yo
On 29-okt-08, at 00:13, Don Dailey wrote:
The marker idea I call aging. I call the variable age because it
seems
to be more descriptive to me. Each playout has a different age
associated with it. The idea is used in persistent hash tables in
computer chess and I'm sure in other games too
45 matches
Mail list logo
