search

AlphaZeroIncubator / AlphaZero

Our implementation of AlphaZero for simple games such as Tic-Tac-Toe and Connect4.
0 stars 0 forks source link

NN architecture #4

Closed guidopetri closed 4 years ago

guidopetri commented 4 years ago

Our central NN architecture should be based off of AlphaZero but maybe we can leave some things flexible.

For instance, we could have a param for number of residual blocks; a parameter for what game to train/test on; a param for batch norm on/off (or maybe other norm types); and a param for which nonlinearity to use (ReLU to start with, like in the paper maybe?).

guidopetri commented 4 years ago

@abhon I'd still like to have the network be one contiguous thing, please :)

abhon commented 4 years ago

Yeah I'll work on it now.

guidopetri commented 4 years ago

Closed via #32