firstlist and secondlist used for rave
I was trying to implement covariance positions and wondered, it this is the expected behaviour.
In the playouts it is possible that the same position is in firstlist and secondlist.
Of cause this can happen, and I do not remember correctly but I thought I read, that only the first move at a position was counted for rave.
maybe I am wrong.
I had a test like this in my code if (blacklist->get(p) && whitelist->get(p)) fprintf(stderr,"This should not happen!!! a place played by black and white in bitboard?!\n");
which was called with firstlist and secondlist for blacklist and whitelist and it triggered?
Comments (11)
-
repo owner -
repo owner - marked as bug
-
repo owner - changed status to on hold
-
reporter - attached gomill.report
Old behaviour
-
reporter - attached gomill.report
new behaviour test_p2=1
-
reporter Test on 9x9 show significant improvment (20ELO or more). Should I use a parameter rave_use_first_only in tree? or just "fix" it?
-
reporter - changed status to open
my commit 04ae35c
-
reporter just to not confuse you: test_p2 was 1 in both, but in the old behaviour test rave behavoiur change was not implemented, sorry for the unconvinience!
-
repo owner I think a parameter like "rave_use_first_only" would be best for now (to preserve old behaviour), but then in the future it could be removed
-
repo owner When you have a moment, please submit a pull request with this change to close this bug
-
reporter -
assigned issue to
in my branch
-
assigned issue to
- Log in to comment
I am not sure if this should be allowed to happen, but I don't think it will make a big difference either way. I am happy to change my opinion with some empirical results.