Consecutive application of the forward-pass of GraphUNet causes a steady increase of memory, which eventually leads to a Memory Overflow. Changing the functionto_torch_csr_tensorinside the methodGraphUNet.augment_adjto the functionto_torch_coo_tensorseems to solve this issue. Note: I was not ab...
🚀 The feature, motivation and pitch I wonder how to train AttributedGraphDataset. I cannot find any examples of it. I met some problems when I tried to write it by myself. dataset = AttributedGraphDataset(conf.root, "mag") data.x.train_i...
Closes #9736. This PR: Adds the edge_weight parameter to the nn.models.GraphUNet model's forward function, so that users can provide an initial weights vector. Makes edge_weight default to None. I...