Number of unpruned childs wrong in the beginning of a game
I am not absolutly sure, if I understood it wrong.
I added some code to get information on the quality of the unprune order. UN: "number of the move selected"/"number of unpruned moves"
from this it seems, that in the beginning the number of unpruned childs is much to high. Even with 100 playouts it is 70, if I understood everything correctly...
This is the commit: https://bitbucket.org/dsmic/oakfoam/changeset/087d8f86a95e
param playouts_per_move_max 100
clear_board
genmove b [genmove]: r:42.69% ppms:1.61 rd:0.79% r2:1.58 fs:-5.90 fstd:2.49 UN:3/70 bs:0 pv:(C4,D4,D3) = c5
genmove w [genmove]: r:53.02% ppms:1.01 rd:6.37% r2:1.49 fs:6.78 fstd:2.38 UN:1/40 bs:0 pv:(B4,C3,B3) = c4
genmove b [genmove]: r:55.20% ppms:1.59 rd:6.78% r2:1.81 fs:-0.18 fstd:2.27 UN:3/4 bs:1 pv:(D4,E4,E3,F3) = d5
Comments (2)
-
reporter -
repo owner - changed status to resolved
I agree, isPrimary() is required.
- Log in to comment
Sorry for the noise, most probably I did not understand the uct_symmetry, so I added a isPrimary in my count and it seems fine now?!