[Computer-go] AlphaGo Zero
alvaro.begue at gmail.com
Thu Oct 19 06:35:11 PDT 2017
Yes, it seems really odd that they didn't add a plane of all ones. The
"heads" have weights that depend on the location of the board, but all the
other layers can't tell the difference between a lonely stone at (1,1) and
one at (3,3).
In my own experiments (trying to predict human moves) I found that 3 inputs
worked well: signed liberties, age capped at 8, all ones. I think of the
number of liberties as a key part of the game mechanics, so I don't think
it detracts from the purity of the approach, and it's probably helpful for
learning about life and death.
On Thu, Oct 19, 2017 at 7:42 AM, Gian-Carlo Pascutto <gcp at sjeng.org> wrote:
> On 18-10-17 19:50, cazenave at ai.univ-paris8.fr wrote:
> > https://deepmind.com/blog/
> > http://www.nature.com/nature/index.html
> Another interesting tidbit:
> The inputs don't contain a reliable board edge. The "white to move"
> plane contains it, but only when white is to move.
> So until AG Zero "black" learned that a go board is 19 x 19, the white
> player had a serious advantage.
> I think I will use 18 input layers :-)
> Computer-go mailing list
> Computer-go at computer-go.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Computer-go