The challenge in the widely applicable online matching problem lies in making\nirrevocable assignments while there is uncertainty about future inputs. Most\ntheoretically-grounded policies are myopic or greedy in nature. In real-world\napplications where the matching process is repeated on a regular basis, the\nunderlying data distribution can be leveraged for better decision-making. We\npresent an end-to-end Reinforcement Learning framework for deriving better\nmatching policies based on trial-and-error on historical data. We devise a set\nof neural network architectures, design feature representations, and\nempirically evaluate them across two online matching problems: Edge-Weighted\nOnline Bipartite Matching and Online Submodular Bipartite Matching. We show\nthat most of the learning approaches perform consistently better than classical\nbaseline algorithms on four synthetic and real-world datasets. On average, our\nproposed models improve the matching quality by 3--10\\% on a variety of\nsynthetic and real-world datasets. Our code is publicly available at\nhttps://github.com/lyeskhalil/CORL.\n