SUPPLEMENTARY MATERIAL Deep Reinforcement Learning with Stacked Hierarchical Attention for Text based Games