Learning Continuous Treatment Policy and Bipartite Embeddings for Matching with Heterogeneous Causal Effects.