| |
BGonline.org Forums
Bot Comparison after 60/500 games.
Posted By: Michael Depreli In Response To: Bot Comparison after 60/500 games. (Matt Cohn-Geier)
Date: Wednesday, 4 November 2009, at 8:37 a.m.
BTW I would like to see how something like the following GNU rollout setting compares:
5184 trials
0-ply play & cube truncated after 10 plies
2-ply eval at truncation point
Stop after 3.29 jsds (min 36 games)Obviously this would be pretty easy if you could just roll out an entire session and otherwise might not be.
There's no way at present to automate this without rolling out the entire session.
It's also not reproducible.
I experimented a bit with similar settings using truncation/jsds/low miminum trials/VR on and it looked promising.Are these settings you've chosen what you believe to be the best accuracy over speed having tested a few out or your first attempt?
Looking at the bigger picture what are we trying to achieve? We can always make the settings more accurate but there's always a speed hit.
I like XGRoller as it appears to give more accurate results than 4-ply and is very fast.
I don't know how much more accuarate XGR+ will be but the speed hit is fourfold which is quite hefty. Should these "roller" settings be purely for analysis or something you would use to play against or even run rollouts with?
It would be good to find the most accuarte setting given a set overall time limit.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.