Xin Yao comments

Results 123 comments of


                                            Xin Yao

CORAL loss is defined differently from the original paper

I agree with you. @SSARCandy This is my implementation, any advice? ```python def coral_loss(source, target): d = source.size(1) ns, nt = source.size(0), target.size(0) # source covariance tmp_s = torch.ones((1, ns))...

CORAL loss is defined differently from the original paper

@redhat12345 My code is based on `PyTorch>=0.4`, in which `torch.tensor` and `Variable` are merged together.

CORAL loss is defined differently from the original paper

@redhat12345 if the `source` and `target` are cuda tensors, then `torch.ones((1, ns))` should be `torch.ones((1, ns)).cuda()`, as well as that of `nt`. I have tried with this loss and find...

Add CUDA_ARCH 80 to the default compile flags

> @yaox12 I saw you've closed PR4384. Is this still an issue or you will find other resolutions? Either Quan's suggestion or specifying `-DCUDA_ARCH_NAME=All` should work.

Set Large Tensor as UnifiedTensor has error

Cannot reproduce. Can you share more env information, e.g., OS, RAM, etc.?

Multi-GPU graph classification

Not yet. cc @chang-l @TristonC

Bus error(core dumped) in EdgeDataLoader of unsupervised graphsage example (20M edges)

How many GPUs did you use? Have you changed the args such as `--graph-device`, `--data-device`?

Bus error(core dumped) in EdgeDataLoader of unsupervised graphsage example (20M edges)

Can you try adding `--shm-size=64g` (large enough to store the whole graph) to your `docker run` command?

Bus error(core dumped) in EdgeDataLoader of unsupervised graphsage example (20M edges)

This line of code in dataloader will create a shared memory array for shuffling. https://github.com/dmlc/dgl/blob/5ba5106acab6a642e9b790e5331ee519112a5623/python/dgl/dataloading/dataloader.py#L146-L149 When `len(train_seeds)` > 8M, the shared tensor will run out of Docker's default shm size...

Bus error(core dumped) in EdgeDataLoader of unsupervised graphsage example (20M edges)

According to the comments, the shared tensor is used for `persistent_workers=True` (`or num_workers > 0` I think?). We can change the code to use shared tensors only when these conditions...