Code to prepare dataset?

[jzkay12]

Hi, thanks for the open source implementation. Could you provide the code to generate the dataset into txt files as listed? I'm trying your code on other datasets such as citeseer and pubmed, while the repo only provides cora. Also I want to try 15-fold cross validation as in the paper, could you explain how to prepare the data? Thanks

[aGIToz]

The code to generate the txt files :

import torch
import torch_geometric

graph = torch_geometric.datasets.Planetoid(root='../', name='cora')[0]

f = open("net.txt","w")
for i in range(graph.edge_index.shape[1]):
    f.write(str(graph.edge_index[0,i].item())+"\t"+str(graph.edge_index[1,i].item())+"\t"+"1"+"\n")

f = open("label.txt","w")
for i in range(graph.x.shape[0]):
    f.write(str(i)+"\t"+str(graph.y[i].item())+"\n")

f = open("train.txt", "w")
for i in range(graph.x.shape[0]):
    if graph.train_mask[i].item():
        f.write(str(i)+"\n")

f = open("dev.txt", "w")
for i in range(graph.x.shape[0]):
    if graph.val_mask[i].item():
        f.write(str(i)+"\n")

f = open("test.txt", "w")
for i in range(graph.x.shape[0]):
    if graph.test_mask[i].item():
        f.write(str(i)+"\n")

f = open("feature.txt", "w")
for i in range(graph.x.shape[0]):
    f.write(str(i))
    count = 0
    for j in range(graph.x.shape[1]):
        if graph.x[i,j].item():
            if not count:
                f.write("\t"+str(j)+":"+str(graph.x[i,j].item()))
                count += 1
            else:
                f.write(" "+str(j)+":"+str(graph.x[i,j].item()))
    else:
        f.write("\n")