On Tue, May 11, 2010 at 8:49 AM, Jean-loup Gailly <[email protected]> wrote: > It is in slides 82 and 83 of > http://www.math-info.univ-paris5.fr/~bouzy/publications/CIG07-tutorial-1avril2007.ppt > The 70% refers to keeping the transposition table across moves; a > simple tree > can and should also be kept across moves. > The transposition table vs tree is only 60% but this is good enough for me.
Sure it's good enough. Does that mean it is the gain you're actually getting? Does the 70% refer to just keeping the tree (table) across moves? Or is it 7% on top of the 63% gain of transpositions? Because I have played with keeping the tree across moves. But it saves on average just 5% of the nodes. That is not enough to give a win-rate of 70%. Admittedly this was before I tried RAVE, so possibly the RAVE information gives this an important boost? These win-rates should be accompanied by playout numbers to be really meaningful. Normally I would expect transpositions to give a proportionally higher gain the larger the number of playouts is. What I was seeing was the opposite, some gain for low numbers of playouts and no gain or a loss for higher numbers. Mark _______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
