I'm using RAVE too. I didn't make any other changes, besides adding progressive widening.
For the unpruning, I'm only using the ELO feature strengths. I'm using the same formula as in the paper: unprune at t(n+1)=t(n) + 40*1.4^n; t(0)=0. Also I increased my parameter that says when to expand a node. -- Francois van Niekerk Email: [email protected] | Twitter: @francoisvn Cell: +2784 0350 214 | Website: http://leafcloud.com On Sat, Jan 15, 2011 at 5:11 PM, Petr Baudis <[email protected]> wrote: > Hi! > > On Sat, Jan 15, 2011 at 04:05:13PM +0200, Francois van Niekerk wrote: >> I managed to get a large strength increase by doing progressive >> widening/unpruning. In the order of 300 ELO on 9x9 against GnuGo. I >> then retrained the features for 19x19, which gave me a hit on 9x9, but >> should do much better on 19x19. > > Interesting! Are you using pure UCT or RAVE? > > I can understand many interesting things to give a boost to plain UCT, > but improving RAVE tree search has been always very difficult for me, > since it is already so powerful. > > -- > Petr "Pasky" Baudis > Computer science education cannot make an expert programmer any more > than studying brushes and pigment can make an expert painter. --esr > _______________________________________________ > Computer-go mailing list > [email protected] > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go > _______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
