When I try to use the NodeCollator method to generate a small batch of subgraphs, the number of num_dst_nodes in the generated subgraph is less than the number of train_mask I input

[shijimozuihouyiren]

tmg = mgs[0] #tmg tmg is the isomorphic information transfer graph I generated before
train_mask #tensor generated from tmg’s oringin graph g size = 2987
sampler = dgl.dataloading.MultiLayerFullNeighborSampler(1)
tcollator = NodeCollator(tmg,train_mask,sampler)
c =collators[0].collate(train_mask)[2]

[minjie]

Hi @shijimozuihouyiren , would you mind providing more information, e.g., an runnable code snippet, your DGL version, PyTorch version, etc.? This will enable us to dive deeper.

[shijimozuihouyiren]

Thanks for reply ！I will take screenshots of my pytorch and dgl versions

Then I generated an ondrive share link for some of my data, you can try to run the ipynb file

dgl_fortest

[minjie]

The DGL version is pretty old. Now is already 1.0. Could you try upgrade DGL to see whether the issue still exists?

[shijimozuihouyiren]

i have update my dgl version , but it doesn’t work

[shijimozuihouyiren]

I checked the trainset I generated. I didn’t check the tensor.item() of the trainset before. In fact, there is a repeated generation, so the number of dstnodes nodes in the final generated subgraph is reduced.