Problem about deal with new dataset

[jianchaoji]

I got a new dataset that consist of two files. One is relation.txt and the other one is feature.txt

The relation file is representing the connection relationship between different nodes like this:
1 0
2 0
3 0
2 1
3 1

And the feature file is the one hot features of each node like this:
1 0 0 1 1 0 0 0
2 0 1 0 0 0 1 0

Can I get some advice about how to transform them to the format like "planetoid" (https://github.com/kimiyoung/planetoid/tree/master/data)

Thank you so much!

[saiden89]

Hi,
it would probably be best if you take a look here for a well detailed example on how to structure your data for Pytorch Geometric.
In a nutshell, your features are just a torch.tensor with shape (num_nodes, num_features). In your case:

x = torch.tensor([[1, 0, 0, 1, 1, 0, 0, 0], 
                  [2, 0, 1, 0, 0, 0, 1, 0]], dtype=torch.long)

As for your edge_list (what you call connection relationship), that's another torch.tensor of shape (2, num_edges). If you're working with an undirected graph, the index should report both directions. Like so:

edge_attr = torch.tensor([[1, 0],
                          [0, 1], 
                          [2, 0],
                          [0, 2],                           
                          [3, 0],
                          [0, 3],
                          [2, 1],
                          [1, 2],
                          [3, 1],
                          [1, 3]], dtype=torch.long)

That's just the basics, but you will find the Pytorch Geometric is very consistent when it comes to data structures. For a more detailed explanation feel free to browse the linked documentation as well as the worked examples in this GitHub repository. There are also some Colab notebooks if you prefer.

[rusty1s]

To follow-up on @saiden89, you can read your txt files in via standard Python file reading, or more advanced techniques such as pandas.read_csv. This will get you the aforementioned tensors as numpy arrays.

[jianchaoji]

I will have a try. Thank you!