c-clark

3 ms·

1. 1.▲
   by c-clark 11y ago

   That's about what I would expect. The two major weaknesses of "naively" training a Go player in this no-look-ahead purely-supervised way are: 1) The training data only consists of positions that occurred in professional games
2. 2.▲
   by c-clark 11y ago

   Thanks for catching that, I fixed the link. The network itself has no capability to pass its turn, which is a consequence of the fact it was only trained to predict player moves, not passes (we thought trying to learn when to pass would be
3. 3.▲
   Show HN: Play Go Against a Deep Neural Network(chrisc36.github.io)

   77 points by c-clark 11y ago|31 comments