If you look at the latest Mogo-paper they do initialize statistics in various
ways to make the programs stronger. The initialization should correspond to a
prior belief of the probability of the move to be good. I am currently
experimenting with Valkyria and the effect of assuming some prior values for
winrates and number of visits for new nodes can range between .05 and .50 in
performance against GnugGo with about 1500 simulations per move. For
most cases
performance seems to be worse than the standard approach but for some
parameter
values it seems to make the program stronger.
Valkyria also uses AMAF but I have not yet tested how it interacts with using
priors. I have too many parameters to explore right now.
Quoting Don Dailey <[EMAIL PROTECTED]>:
I just thought of something. I think I initialize the statistics array
with 1 draw per move as a cheesy way to avoid divide by zero error.
Could this be affecting the performance? Perhaps at low levels like
this it has a noticeable effect? Would it make the program especially
vulnerable to an identical program that doesn't do this?
--
Magnus Persson
Berlin, Germany
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
