Train/Test split for multiple Graphs (Node regression)

[sivaramkrishnan101]

Hi,

I am very new to Pytorch Geometric, which has been a pleasure to use.

Currently, I have 200 graphs, each with 5 nodes, each having 2 node features outputting a numeric label. The graphs were made using the NetworkX library.

This was converted to pytorch geometric objects using from_networkx
Code snippet(Here data_1 stores the first pytorch geometric object and so on)
for i in range(1,199):
globals()['data_'+str(i)] = from_networkx(Graphs[i],['pos_x','pos_y'])

Now, I'm not quite sure on how to perform a train/test split such that some nodes in a considered Graph go into the train while the rest go into the test set(Inductive based).

I have tried creating a dataset with these 199 pytorch geometric objects such that

data_list = data_list = [data_1, .... data_199]

However, I am unable to apply RandomNodeSplit on it as it has been converted to a list and limits functionality.

Please, I have been struggling with this for quite a while and any hints/ideas would be greatly appreciated.

[rusty1s]

There does not exist a particular function for this, but you can easily shuffle and split by yourself

data_list = [data_list[i] for i in torch.randperm(len(data_list)).tolist()]
train_data_list = data_list[:150]
test_data_list = data_list[150:]

[sivaramkrishnan101]

I greatly appreciate your prompt response.

I will try it out.

Thank you!