After getting some strange results with my experiments I realised
that doing these tests on the reference-bot is probably not
appropriate. The MC+AMAF combination is a completely different beast
than MC+UCT-search.
So I need to do the testing on a reference implementation for MC+UCT-
searc
I could find anything problematic with your specification so I just
make some comments.
Quoting Mark Boon <[EMAIL PROTECTED]>:
- When it reaches N simulations, the child of the root-node with the
best wins-visits ratio is played. I've also seen that simply the child
with the highest number of
Mark Boon wrote:
The treatment of pass feels a bit convoluted. The idea behind it is that
I don't want to allow pass-moves in the tree early in the game. But
towards the very end of the game I want to allow pass to avoid being
forced to make a bad move or not being able to make a move at all.
Thanks for the comments Magnus.
On 20-nov-08, at 13:00, Magnus Persson wrote:
I'd also like to hear opinions on what would be a good N for a
reference bot. If I set N to 2,000, just like the reference-bots on
CGOS, it plays rather poorly. Much worse than the ref-bot. I haven't
tested it a lot
On 20-nov-08, at 13:22, Michael Williams wrote:
Mark Boon wrote:
The treatment of pass feels a bit convoluted. The idea behind it
is that I don't want to allow pass-moves in the tree early in the
game. But towards the very end of the game I want to allow pass to
avoid being forced to make
You could simply allow the pass all the time. Early in the game, it
will be a significantly inferior move and not often explored in the
tree. That may not be optimal, but it's certainly not convoluted and
you are guaranteed to never fail to generate the pass move when you
needed it.
I rem
On 20-nov-08, at 14:03, Michael Williams wrote:
You could simply allow the pass all the time. Early in the game,
it will be a significantly inferior move and not often explored
in the tree. That may not be optimal, but it's certainly not
convoluted and you are guaranteed to never fail to
Quoting Mark Boon <[EMAIL PROTECTED]>:
Thanks for the comments Magnus.
On 20-nov-08, at 13:00, Magnus Persson wrote:
The way I understood the article, after a playout it updates all the
nodes at the current level of all the moves played during the playout
(if it's a win for the player) with a
On 20-nov-08, at 14:03, Michael Williams wrote:
You could simply allow the pass all the time. Early in the game,
it will be a significantly inferior move and not often explored
in the tree. That may not be optimal, but it's certainly not
convoluted and you are guaranteed to never fail to
About allowing early passes.
I think my problem is that RAVE/AMAF really say nothing about the
value of pass moves, which makes it problematic when selective search
do not search bad moves such as pass. So how are we supposed to know
when to search pass and not?
Thus, everytime I had some
On 20-nov-08, at 14:42, Magnus Persson wrote:
I think you understand the basic principle of RAVE correctly. The
hard part is how to weight together the AMAF value (which I call
*virtual win-visit* ratio) with the actual win-visit ratio. And I
have to admit that my implementation of RAVE i
On 20-nov-08, at 14:51, Magnus Persson wrote:
and had to add a lot of kludges to avoid problems with older
kludges and so on...
Hey, that sounds strangely familiar :)
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.o
Quoting Mark Boon <[EMAIL PROTECTED]>:
What is not exactly clear to me is what you mean by 'postponing
expansion'. Let me write it in my own words to see if that's what you
mean. When you have selected a best node based on the UCT + wins/visits
value which has no children yet, you first simply do
On 20-nov-08, at 15:20, Magnus Persson wrote:
Quoting Mark Boon <[EMAIL PROTECTED]>:
What is not exactly clear to me is what you mean by 'postponing
expansion'. Let me write it in my own words to see if that's what you
mean. When you have selected a best node based on the UCT + wins/
visits
Quoting Mark Boon <[EMAIL PROTECTED]>:
OK, things start to fall in place for me. I was wondering all this time
what would happen with the information of the simulations that happen
before expansion. So the answer is: nothing. But at least the result of
the playout gets percolated up the tree, s
>My first monte carlo programs used ownership info very effectively, but it
>can be that by using
>AMAF this information is used already.
As a relative beginner in these matters, the more I look at AMAF,
the less I like it, and I think that has to do with AMAF ignoring
relative sequencing. By l
Claus,
I think you are raising some very valid questions. I'm a bit
ambivalent towards AMAF for very similar reasons. One thing in
defense of AMAF though, is that it doesn't necessarily need to make
Go-sense to be useful. MC simulations also don't make much Go-sense.
For example, moves ar
Hi,
I have read that the amount of money that a winning computer go
program would make in a go tournament is insignificant compared to the
amount of money that such a program would earn selling to the general
public. I have also read that the biggest pirates of computer software
come from
18 matches
Mail list logo
