Pytorch Geometric was also getting some love here recently:
We introduce PyTorch Geometric, a library for deep learning on irregularly structured input data such as graphs, point clouds and manifolds, built upon PyTorch. In addition to general graph data structures and processing methods, it contains a variety of recently published methods from the domains of relational learning and 3D data processing. PyTorch Geometric achieves high data throughput by leveraging sparse GPU acceleration, by providing dedicated CUDA kernels and by introducing efficient mini-batch handling for input examples of different size. In this work, we present the library in detail and perform a comprehensive comparative study of the implemented methods in homogeneous evaluation scenarios.
Graph Neural Networks(GNNs) recently emerged as a powerful approach for representation learning on graphs, point clouds and manifolds (Bronstein et al., 2017; Kipf & Welling, 2017). Similar to the concepts of convolutional and pooling layers on regular domains, GNNs are able to (hierarchically) extract localized embeddings by passing, transforming, and aggregating information (Bronstein et al., 2017; Gilmer et al., 2017; Battaglia et al., 2018; Ying et al., 2018). However, implementing GNNs is challenging, as high GPU throughput needs to be achieved on highly sparse and irregular data of varying size. Here, we introduce PyTorch Geometric (PyG), a geometric deep learning extension library for PyTorch (Paszke et al., 2017) which achieves high performance by leveraging dedicated CUDA kernels. Following a simple message passing API, it bundles most of the recently proposed convolutional and pooling layers into a single and unified framework. All implemented methods support both CPU and GPU computations and follow an immutable data flow paradigm that enables dynamic changes in graph structures through time. PyG is released under the MIT license and is available on GitHub.1 It is thoroughly documented and provides accompanying tutorials and examples as a first starting point.2
Runtime Experiments. We conduct several experiments on a number of dataset-model pairs to report the runtime of a whole training procedure obtained on a single NVIDIA GTX 1080 Ti (cf. Table 4). As it shows, PyG is very fast despite working on sparse data. Compared to the Deep Graph Library (DGL) 0.1.3 (Wang et al., 2018a), PyG trains models up to 15 times faster.