…consider any states the play is not very optimal

…arned how to play against itself. Proceeded to draw

… not enough training? Or does it have something to do with the 5 and 2 we pass to the network.

…how how class imbalance affects the posterior probability.

…o each other in a matrix.

…ion for the policy gradient method using a NN. The current parameters work well. Further tuning of hyperparameters could be useful.