Action Selection Policies for Walking Monte Carlo Tree Search