On Nov 15, 2007 7:40 AM, Petr Baudis <[EMAIL PROTECTED]> wrote: > Thanks a lot! I'm doing that now and while the ranks are not yet stable, > they are all only slightly above 1050 now already. :-( (Even the > variants with extra domain-specific knowledge.) I guess I still have > some bugs there.
Here are some of my early UCT bugs: * Failure to switch wins to losses and losses to wins from one depth of the tree to the next. (I amazed myself with how many times I found this bug in different spots) * I had code intended to be transposition table aware that would resum the wins and losses (for correctness). I initially didn't differentiate between sims done when a node was a leaf and the sum of the sims of children. This lead to my bot forgetting many sims that it did. Based off the posts of others, it seems like creating new children of a leaf after 50 sims gives extra strength (smaller values yield weaker bots at 10k sims) I took down my bots after my internet connection was really bad. Tonight, I'll put up an AMAF and a pure UCT version for reference. They'll likely be something like hb-683-UCT-10k and hb-683-amaf-1k. Maybe I'll try a 10k amaf variant instead... > > > By the way, when I make significant change to the algorithm (even a > bugfix that will significantly improve it), is it considered better > practice to have the changed bot playing from scratch, or introduce the > change to existing bots and see how the rating changes? Because the rank adjustment algorithm can't be told that your bot is new, it messes up the ranking algorithm a bit. It's best to start a new bot. An excellent example of this is when I went to add my AMAF reference bot on CGOS. I happened to choose a name that matched a previous buggy entrant. Even after ~50 games, its rank was still 1000 ELO below what it was supposed to be. > P.S.: By the way, so that google picks this up: If you are getting > > can't read "mv": no such variable > > TCL errors from the CGOS client, that means you are hitting a bug in it > - it does not properly ignore empty lines from the GTP client (or, > heavens forbid, comments). Make sure that you do not emit any. It took > me several 10-minutes waits and time losses of my bot to debug this. > > -- > Petr "Pasky" Baudis > We don't know who it was that discovered water, but we're pretty sure > that it wasn't a fish. -- Marshall McLuhan > _______________________________________________ > computer-go mailing list > computer-go@computer-go.org > http://www.computer-go.org/mailman/listinfo/computer-go/ >
_______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
