Matthias Fey
Matthias Fey
What does `data.validate()` return for you?
`data.validate()` just checks for invalid edges. It cannot automatically detect whether edges are semantically incorrect.
Can you show me the installation log of `torch-sparse`? This suggests that you are somehow installing from source :(
Can you run ``` pip uninstall torch-sparse pip install --verbose torch-sparse -f https://pytorch-geometric.com/whl/torch-2.0.0+cpu.html
Oh, I somehow missed you are on macOS. In this case, the library will get built from source, and METIS needs to be installed beforehand in order to get picked...
You can find instructions [here](https://github.com/rusty1s/pytorch_sparse#from-source).
Looks like PyGAS is compiled without CUDA support. You may find look when installing via ``` FORCE_CUDA=1 pip install git+https://github.com/rusty1s/pyg_autoscale.git
Can you try to make [this line](https://github.com/rusty1s/pyg_autoscale/blob/master/examples/train_gin.py#L96) an out-of-place function. This behavior might have changed due to different PyTorch versions: ```python h = h.clone() h += x[:h.size(0)]
Can you change this line [here](https://github.com/rusty1s/pyg_autoscale/blob/master/examples/train_gin.py#L78) to `relu()` instead of `relu_()`?
What does ``` for data in loader: print(data.num_nodes) ``` return? I would expect that one adjacency matrix reports a smaller number of nodes due to isolated nodes.