I agree that dgl has better design, but pytorch geometric has reimplementations of most of the known graph convolution layers and pooling ...
確定! 回上一頁