Dude, nice work. I really must take a closer look later on.

On 12/16/05, horndude77 / gmail.com <horndude77 / gmail.com> wrote:
> I know this is a bit late, but I've been trying to train my neural net.
> I've never used them before and I've been wanting to learn. I created a
> neural net with the input being the kalah board plus one bias node(so
> 15 input nodes), 20 hidden nodes, and 6 output nodes. A move is
> selected by running the neural net and then picking the valid move
> whose corresponding output node is highest. I'm training the net right
> now using Particle Swarm Optimization to find good weights.
>
> I trained it against dave's player and adam's remove right. It found a
> pretty good solution against the remove right player, but not dave's.
> The best I could get after searching for a while was losing by 12. I
> might have better luck if I just take dave's solution and replace the
> function it uses to rate the board's goodness with a neural net and
> then train it. That might be fun for later.
>
> Anyways sorry about the lateness, but here it is.
>
> -----Horndude77
>
> NeuralPlayer.rb (Sorry about the long section of weights. I had them on
> one line each.):
>