Algorithm 2 NMCTS (startNode, Seq, max_iter, level)  //Seq defines best $F_{sub}$ the tree has found so far

best_reward = inf.  best_seq = ()  Current_node = startNode  For iteration number in the called level:   While Current_node is not terminal and not fully expanded:  Current_node = selection(Current_node)  Seq = Current_node.feature_subset  If level=1:   While Current_node is not terminal:    Reward, Seq = Random_rollout(Current_node)  Else:    Reward, Seq = NMCTS(Current_Node, Seq, max_iter, level-1)  Back_propogation(Current_Node, reward)  If Reward < best_reward:   best_reward = reward   best_seq = Seq