Efficient Contraction of Large Tensor Networks for Weighted Model Counting through Graph Decompositions.