On 23 May 2016 08:00, "Rémi Coulom" <remi.cou...@free.fr> wrote: > > I think one of the main problems is that the network learns good replies to good moves. The training set does not have good replies to bad moves, but the search tree is full of bad moves that need to be punished.
As the joseki that CS messed up was quite basic, it looks like a bug in your search or a problem of your policy network. Maybe you should train and run your networks on GPU. Aja > Alvaro's suggestion looks good. This is one of the experiments I want to try. > > Rémi > > > On 05/22/2016 11:41 PM, Henry Hemming wrote: >> >> If the network is too selective, the cost function used to generate it doesn't penalize extreme predictions sufficiently? It was generated using quadratic cost when it should have been using cross-entropy cost? >> >> On Mon, May 23, 2016 at 12:08 AM Álvaro Begué <alvaro.be...@gmail.com <mailto:alvaro.be...@gmail.com>> wrote: >> >> Disclaimer: I haven't actually implemented MCTS with NNs, but I >> have played around with both techniques. >> >> Would it make sense to artificially scale down the values before >> the SoftMax is applied, so the probability distribution is not as >> concentrated, and unlikely moves are not penalized as much? >> >> >> >> On Sun, May 22, 2016 at 3:54 PM, Rémi Coulom <remi.cou...@free.fr >> <mailto:remi.cou...@free.fr>> wrote: >> >> Hi, >> >> Thanks for using Crazy Stone. >> >> I tried changes during the week, but nothing worked. So the >> version that played the game was almost identical to the >> commercial version. >> >> The search did not anticipate Black E8 after B3. It seems the >> NN makes the search too selective. I will investigate more. >> >> Rémi >> >> ----- Mail original ----- >> De: "Paweł Morawiecki" <pawel.morawie...@gmail.com >> <mailto:pawel.morawie...@gmail.com>> >> À: computer-go@computer-go.org >> <mailto:computer-go@computer-go.org> >> >> Envoyé: Dimanche 22 Mai 2016 21:29:56 >> Objet: Re: [Computer-go] Hajin Lee will play a live commented >> game against CrazyStone >> >> >> >> Hi, >> >> >> >> >> >> >> >> >> >> It's fun to hear the pro making comments as she goes. I had >> hoped for a better game, though. >> Any comments from the CS camp? >> >> >> >> I'm not from CrazyStone Team but a happy user of CS Deep Learning. >> >> >> I analyzed the game (30 000 playouts per move) with the >> version commercially available and it got everything right. I >> mean every move Hajin Lee questioned was also questioned by >> CrazyStone running on my PC. It includes: >> >> >> - questionable attachment in the first joseki they played >> - hane in upper-left corner >> - and finally a blunder at b3 (loosing move) and selection of >> this joseki as well >> >> >> Remi said that he's been working hard to improve CS over the >> last week, but it looks like something went wrong and instead >> he got clearly a weaker program. Particularly this B3 is >> really strange, where right after this white catches >> everything with one move. Remi, what went wrong? >> >> >> Cheers, >> Paweł >> >> >> >> >> >> >> >> >> >> >> >> >> >> Thanks, >> Álvaro. >> >> >> >> >> >> >> >> >> On Mon, May 16, 2016 at 3:58 AM, Xavier Combelle < >> xavier.combe...@gmail.com <mailto:xavier.combe...@gmail.com> > >> >> wrote: >> >> >> >> That's fantastic >> >> >> I suppose crazystone will play with crazystone account, but >> what will be her handle ? >> >> >> >> >> >> 2016-05-16 9:50 GMT+02:00 Rémi Coulom < remi.cou...@free.fr >> <mailto:remi.cou...@free.fr> > : >> >> >> >> Hi, >> >> I am very happy to announce that Hajin Lee will play a live >> commented game against Crazy Stone on Sunday, at 8PM Korean >> time. The game will take place on KGS, and she will make live >> comments on her youtube channel. >> >> Haylee's youtube: >> https://www.youtube.com/c/HayleesWorldofGoBaduk >> >> Rémi >> _______________________________________________ >> Computer-go mailing list >> Computer-go@computer-go.org <mailto:Computer-go@computer-go.org> >> >> http://computer-go.org/mailman/listinfo/computer-go >> >> _______________________________________________ >> Computer-go mailing list >> Computer-go@computer-go.org <mailto:Computer-go@computer-go.org> >> >> http://computer-go.org/mailman/listinfo/computer-go >> >> >> _______________________________________________ >> Computer-go mailing list >> Computer-go@computer-go.org <mailto:Computer-go@computer-go.org> >> >> http://computer-go.org/mailman/listinfo/computer-go >> >> >> _______________________________________________ >> Computer-go mailing list >> Computer-go@computer-go.org <mailto:Computer-go@computer-go.org> >> >> http://computer-go.org/mailman/listinfo/computer-go >> _______________________________________________ >> Computer-go mailing list >> Computer-go@computer-go.org <mailto:Computer-go@computer-go.org> >> >> http://computer-go.org/mailman/listinfo/computer-go >> >> >> _______________________________________________ >> Computer-go mailing list >> Computer-go@computer-go.org <mailto:Computer-go@computer-go.org> >> >> http://computer-go.org/mailman/listinfo/computer-go >> >> >> >> _______________________________________________ >> Computer-go mailing list >> Computer-go@computer-go.org >> http://computer-go.org/mailman/listinfo/computer-go > > > _______________________________________________ > Computer-go mailing list > Computer-go@computer-go.org > http://computer-go.org/mailman/listinfo/computer-go
_______________________________________________ Computer-go mailing list Computer-go@computer-go.org http://computer-go.org/mailman/listinfo/computer-go
