This is what I do (no tree, just a hash table). The cost is that the nodes
become very large because every node also holds all the child information,
all rave counters, etc. So memory usage is higher.
David
From: computer-go-boun...@computer-go.org
Hi!
How would you recommend to bias nodes in the tree according to pattern
strengths, when the pattern strength can be in (0,inf)? (I'm using this
for Remi's pattern ELO model, but I guess other fuzzy pattern values face
similar problems.)
I have node values in the form of W / N (wins /
Petr Baudis wrote:
Hi!
How would you recommend to bias nodes in the tree according to pattern
strengths, when the pattern strength can be in (0,inf)? (I'm using this
for Remi's pattern ELO model, but I guess other fuzzy pattern values face
similar problems.)
I have node values in the
On Mon, Dec 14, 2009 at 07:46:54PM +0100, Rémi Coulom wrote:
I bias them too. I use move probability, not move gamma, which
normalizes the value between 0 and 1. I multiply this probability by
a constant, divide by the number of playouts, and add it to the move
score.
I see, but the constant
Petr Baudis wrote:
On Mon, Dec 14, 2009 at 07:46:54PM +0100, Rémi Coulom wrote:
I bias them too. I use move probability, not move gamma, which
normalizes the value between 0 and 1. I multiply this probability by
a constant, divide by the number of playouts, and add it to the move
score.
I
I've found that AMAF gives very little boost with light playouts,
Thanks for this interesting comment.
I would (erroneously) have believed that AMAF gives
better results with non-optimized implementation (e.g. in Havannah with no
expertise Amaf provides huge improvements). In particular, I
That same statement baffles me. AMAF gives a huge boost with light
playouts for me. As the number of playouts increase, AMAF gives less and
less. After a few thousand playouts it's almost nothing but if it's worse
than not doing AMAF it is difficult to measure. Of course MCTS never does
I've found that AMAF gives very little boost with light playouts,
you really need something fairly meaningful already to get any kind
of real boost. To have at least 10% chance of beating GNUGo with
reasonable time per game (to be able to play-test your bot), I think
you can't avoid doing a
This approach seems easier to manage.
2009/12/14 David Fotland fotl...@smart-games.com
This is what I do (no tree, just a hash table). The cost is that the
nodes become very large because every node also holds all the child
information, all rave counters, etc. So memory usage is higher.
On Mon, Dec 14, 2009 at 11:45:44AM -1000, Mark Boon wrote:
I've found that AMAF gives very little boost with light playouts,
you really need something fairly meaningful already to get any kind
of real boost. To have at least 10% chance of beating GNUGo with
reasonable time per game (to be
I'm a bit confused by the difference between RAVE and AMAF. Or are
they the same thing?
- Brian
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
It's easy to get confused -- different researchers use the terms
slightly differently.
They both gather data on moves other than a move made from the current
board configuration. I would say that AMAF stores statistics on every
move played from any position, while RAVE only stores info on
12 matches
Mail list logo
