Nonlinear Relational Markov Networks with an Application to the Game of Go