Hi such question that do you typically traverse all child objects or is
there faster way to select explored node child object.
I have concluded that it is not at least easy as multiple nodes uct values
change each simulation so trying to keep biggest uct value in first in list
is maybe too expen
>In go we talk instead about moves that lose points. UCT programs
often play moves that clearly lose points in an attempt to maximize the
probability of winning.
I am just started reguraly observe how this uct really plays as my bot is
started play in CGOS, so I am kind of newbie in this but ob
This is a mistake. There are often moves that are illegal for black that
are big for white. If you don't let white play there, white can lose a
lot
of points. Connections through false eyes are one example.
Yep agree that, knowing that it is not fair for other but kind of
rationalized it tha
rtain assumptions.
mercy rule?
t. Harri
- Don
Harri Salakoski wrote:
Hmm, sorry if this is old subject but does it effect much for playout
quality if I cut playouts for example max 110 moves in 9*9 board?
Yes, that will significantly hurt your play-outs.Do you throw away
the results o
Hmm, sorry if this is old subject but does it effect much for playout
quality if I cut playouts for example max 110 moves in 9*9 board?
Yes, that will significantly hurt your play-outs.Do you throw away
the results or score it as is?
I have a very liberal cutoff on my program - from any posi
number empty points or using some other
policy? sorry..
- Original Message -
From: "Harri Salakoski" <[EMAIL PROTECTED]>
To: "computer-go"
Sent: Tuesday, January 15, 2008 2:43 PM
Subject: Re: [computer-go] On average how many board updates/sec can top
Goprogr
The *average* length of a 9x9 playout is roughly 100 moves.
The max length is much larger.
The *average* length of a 9x9 playout is roughly 100 moves.
The max length is much larger.
Hmm, sorry if this is old subject but does it effect much for playout
quality if I cut playouts for example max 11
5 ...
4 WW.
3 BBB..W.
2 ...B.W.
1 ..B..W.
ABCDEFG
Ha, I give blacks more room to play, and shape database did not find it, or
maybe I used it wrong :).
But this search find B2 now in 1258437 nodes :| which is quite much and
takes couple seconds.
Don't know but really hope it goes bett
I am sure that it is good, can't compare node counts for anything which uses
any static knowledge.
2282 is current count and you can increase or decrease those
proofCounter/disProofCounter limits and it still
works but don't know exactly when it is doing it right.
t. hArri
- Original Messag
Now it is only 2282 nodes, and it still has maybe bugs. It returns
immediately which is good, I have hopes for that it comes better even it has
no bugs because it lacs move ordering logic,
If I only find couple bugs its start to be what I expected. Then need to
make desision is it good enough to
go] Microsoft Research Lectures: Akihiro Kishimoto
Le mercredi 12 décembre 2007, Harri Salakoski a écrit :
Such comment just take my word back little it is maybe awesome but I can't
say is it or not, as have still bugs left.
E E E
E E E
BEE
WWWEBEE
E E EWBEE
E E WEBEE
ABCDEFG
For
combinations, I suspect there should be small heuristic
for OR/FirstPlayer node proof path which has most kills first, I add that,
which maybe don't fix it alone as exact calculation child node proof and
disproof tresholds is little vaque for me.
t. Harri
Le mercredi 12 décembre 2007,
Such comment just take my word back little it is maybe awesome but I can't
say is it or not, as have still bugs left.
E E E
E E E
BEE
WWWEBEE
E E EWBEE
E E WEBEE
ABCDEFG
For example current version(not released) goes trought 162438 nodes before
it proofs black B2 kills(without any ord
I coded algorithm based for that representation, it really looks another
awesome thing worth of investigating.
Planning to use that for atleast small board search investigations as it has
quite much power.
That is included lates version pf http://sourceforge.net/projects/narugo I
am happy to hea
;Nick Wedd" <[EMAIL PROTECTED]>
To: "computer-go"
Sent: Thursday, December 06, 2007 3:19 PM
Subject: Re: [computer-go] cgos
In message <[EMAIL PROTECTED]>, Harri Salakoski
<[EMAIL PROTECTED]> writes
Hi is last black move A4 really illegal in cgos rules?
Just e
Hi is last black move A4 really illegal in cgos rules?
Just ensure before start change things. It seemed weird.
White D9 shouls change board situation and white kills two stones before A4
is getting then
one stone back fine? t. harri
W A4
9 WW..W.WWW
8 ...W.W.W.
7 WBW..
6 WBBWBBWWW
5 .WWWBB
Yep and thanks, thanks also everybody else who replyed. It is quite solved,
it is pure "A1" or "E3" coordinates which cgos server expects for genmove
commands, unfortunately own loop implementation send replys for info and
other stuff which i quess was buffered for server, when it was time to
h
r 27, 2007 10:00 PM
Subject: Re: [computer-go] CGOS down? Java client - basic GTP problem
Cgos has it's own syntax - it's not GTP although it superficially
resembles GTP.
cgos3.tcl sits between your program and the server and converts the
conversation to gtp for your program.
- Don
Harri Sa
syntax
- Original Message -
From: "Harri Salakoski" <[EMAIL PROTECTED]>
To: "computer-go"
Sent: Tuesday, November 27, 2007 6:32 PM
Subject: [computer-go] CGOS down? Java client
Hmm, I am not sure as only trying to make Java bot connect CGOS, but it
seems t
Hmm, I am not sure as only trying to make Java bot connect CGOS, but it
seems to be down.
It seems that it does not give feedback after password. Also I found cgos
email list "cgos-developers" is it also for users or is this list ok?
t. harri
___
c
stance) then he wins by
3 points instead of 9.
If black plays first move on corner, white wins by 9 points.
Does anyone have any data to check these "facts" ??
It may try this with 4x4 after doing something to improve the move
ordering.
- Don
Harri Salakoski wrote:
5x5 was solved
exact. If the
last 4 iterations return the same score, I put the root position in the
hash table as a perfect score. I sample the space randomly by
making a random move after searching some position, so that it explores
the full space eventually. This is not very systematic, but it&#x
> Has anyone did this before or has anybody thoughts about this?
I have done that for 4*4 and 3*3, my code is not in shape that it could be used
5*5 but have high believes it is anyway possible used for 6*6 some day. But
this was discussed in this group earlier and nothing new has occurred since
There is many good videos in videolectures, seen some UCT, e.t.c but as I
was specially delighted this particular video, which covers many practical
points for machine learning and is propably useful remind some things in
go-game AI coders also.
http://videolectures.net/mlss06tw_roweis_mlpgm/
Considering how monte carlo actually works, I think it's plausible
to argue that it works best where the distance to endgame is small.
Is it then natural use it only after middle game.
Build fuseki-joseki-extend scripted engine and change for monte-carlo engine
in middle game?
t. harri
__
Absolutelu _great_ link, raises my go rank I hope.
thanks.
Do they exist?
I have watched many hilarious youtube stuff, but no clue to search go stuff,
great.
t. harri
- Original Message -
From: "Chris Fant" <[EMAIL PROTECTED]>
To: "computer-go"
Sent: Saturday, October 13, 2007 2:3
Every hand has theoretical winning propability but if you bet exactly according
winning odds you tell too much about your hand,
other players fold and call and raise according your bets, only when they think
they have better cards, that is bad, obivious book play works only starter
levels.
In p
think poker is more difficult than Go and of course chess.
I have only studied poker AI basics and coded game rules, learned play
slightly winning net poker.
But however dare to say my opinion that I totally disagree. Sounds like
somekind of poker hype that
it is as tough problem than Go game A
] Whites turn 2, , E-B2 is optimal move for white
I need to model that difference using some kind of pattern language, but I am
wondering if somebody is done
nice practical definition for that allready.
ty.
t. Harri Salakoski
http://sourceforge.net/projects/narugo
A hypothetical almighty oracle that already knows the correct answer
to every question and the right response in every situation would
never have to adapt. Hence evidence of intelligence according to your
definition would not be observed.
I think it is normal to expect, or at least in my common
Awesome. I believe similar aproach also in game of Go.
I short that there is no limit how smartly you can store information and thats
why I believe Erik e.t.c research in small board sizes is important and why I
have tried to study TSP kernighan kind of algorithms those don't search even
1/1
I think this is what I want. Thanks! So I might have to repeat this
a few hundred times to actually get a legal position?
- George
Implemented that and my test serie of number of rounds needed to generate
legal board position is quite expected.
count=111, count=36, count=28, count=20, count
In normal board size 1% random positions is legal, so it needs some rounds,
but method is still superior propably any other if position must be random.
t. Harri
- Original Message -
From: "Paul Pogonyshev" <[EMAIL PROTECTED]>
To:
Sent: Monday, July 09, 2007 12:38 AM
Subject: Re: [comp
I don't have such algorithm, you can count legal positions like:
http://www.lysator.liu.se/~gunnar/legal.pike.txt
Modifying it could provide some way select random position atleast for small
boards. Ported that for java but not studied much of it yet, intresting
anyway.
t. Harri
- Ori
It seems kind of dream algorithm to be able extend results for bigger board
sizes. Exact knowledge is allways good anyway.
So amazing, it is suitable for max pseudo liberties for one string, count of
legal board positions,
maybe there is other properties for board this fits or more complex issu
On Mar 13, 2007, at 23:27 , Peter Drake wrote:
Hmmm -- p. 735 of Russell & Norvig's AI text contains a strong argument
that "nearest-neighbor methods cannot be trusted for high- dimensional
data".
AFAIK that's because when you add more and more dimensions and try to
calculate the distance
b
http://sourceforge.net/projects/narugo
--
Is project for pattern matching AI generation.
Has pattern profiling, EA and NN e.t.c.
if java1.5,xml are part of your tools check that.
t. Harri
From: "Stuart A. Yeates" <[EMAIL PROTECTED]>
To: "computer-go"
Sent: Thursday, March 08, 2007 12:19 PM
Su
, Harri Salakoski <[EMAIL PROTECTED]> wrote:
>
>
> Hi
> As reading
> http://www.cs.unimaas.nl/~vanderwerf/pubdown/solving_go_on_sm all_boards.pdf
> don't understand chapter 5.1. Following 3*3 go game.
Maybe you were not actually reading the above link, but
Hi
As reading
http://www.cs.unimaas.nl/~vanderwerf/pubdown/solving_go_on_small_boards.pdf
don't understand chapter 5.1. Following 3*3 go game.
-A2
eee
bee
eee
-B2
eee
bwe
eee
-B1
eee
bwe
ebe
-C3 - can be bad move according chapter 5.1
eew
bwe
ebe
-B3? - don't understand how black wins using
If java is ok check also http://sourceforge.net/projects/narugo
There is many projects doing that basic stuff allready.
t. Harri
- Original Message -
From: Thomas Johnson
To: computer-go@computer-go.org
Sent: Wednesday, January 24, 2007 12:52 AM
Subject: [computer-go] Best com
Does anyone have an example where "pass" is the best move, and not part
of the two passes to end the game? I'm trying to >determine if passes
should ever be considered in a search for the best move, and if so, how
to exclude them until it is really >necessary.
Which round is first pass move mad
41 matches
Mail list logo
