Yongduo Sui

thx! I have fixed the bug.

In our paper table1 shows "Semantic robustness against losing partial attributes per node". It means that masks the whole attributes for the selected nodes. "partial attributes" means the whole graph's...

As for superpixel datasets, we just release the code of best augmentation method: node dropping or subgraph. As for edge perturbation for these datasets, we also apply these undirected way...

please follow this repo to prepare the dataset. https://github.com/graphdeeplearning/benchmarking-gnns

1. Please check if the dataset is on the correct path. 2. Maybe you can try to change the other different version for pickle.

This is finetuning code, please use the pretraining code first.

semisupervised_MNIST_CIFAR10 only include experiments for MNIST and CIFAR10 graph classification, not for SBMs. Please ignore main_SBMs_node_classification.py file.

@Austinzhenghua Please put the download MNSIT or CIFAR10 superpixel dataset on the "data/superpixels/" and try again?