Quoting Peter Drake <[EMAIL PROTECTED]>:
In previous versions of Orego, I have added one node per playout. I
just changed that to add a child to a node only if that node has at
least A runs, where A is the area of the board (e.g., 81). This seems
to make the program stronger, if only because it allows me to get in
more runs. Specifically, it allows me to get (a bit) more out of my
4- processor-core machine, because the parts where the threads have
to synchronize on the tree (generating in-tree moves and
incorporating the playout) are shorter.
I just realized the same thing about Valkyria when I implemented progressive
widening in Valkyria, because then it became obvious it is bad to add one node
per playout. The reason is that as progressive widening initially search only
the recent new node a new node is added below every time essentially
creating a
long string of nodes before progressive widening adds a second move
next to the
first. Thus what is evaluated is not the first node added but a random
position
many ply deep down.
I have also added some crude move ordering and it seems to have improved the
rating about 70 points, although a lot can still be done to improve the move
ordering.
_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/
