(dataset)
| 188 | |
| 189 | |
| 190 | def generate_node_data(dataset): |
| 191 | |
| 192 | if dataset in ['cora', 'citeseer','pubmed']: |
| 193 | |
| 194 | adj, x, y = load_data(dataset) |
| 195 | adj = adj.todense() |
| 196 | x = x.todense() |
| 197 | x = feature_normalize(x) |
| 198 | e, u = eigen_decompositon(adj) |
| 199 | |
| 200 | e = torch.FloatTensor(e) |
| 201 | u = torch.FloatTensor(u) |
| 202 | x = torch.FloatTensor(x) |
| 203 | y = torch.LongTensor(y) |
| 204 | |
| 205 | torch.save([e, u, x, y], 'data/{}.pt'.format(dataset)) |
| 206 | |
| 207 | elif dataset in ['photo']: |
| 208 | data = np.load('node_raw_data/amazon_electronics_photo.npz', allow_pickle=True) |
| 209 | adj = sp.sparse.csr_matrix((data['adj_data'], data['adj_indices'], data['adj_indptr']), |
| 210 | shape=data['adj_shape']).toarray() |
| 211 | feat = sp.sparse.csr_matrix((data['attr_data'], data['attr_indices'], data['attr_indptr']), |
| 212 | shape=data['attr_shape']).toarray() |
| 213 | x = feature_normalize(feat) |
| 214 | y = data['labels'] |
| 215 | e, u = eigen_decompositon(adj) |
| 216 | |
| 217 | e = torch.FloatTensor(e) |
| 218 | u = torch.FloatTensor(u) |
| 219 | x = torch.FloatTensor(x) |
| 220 | y = torch.LongTensor(y) |
| 221 | |
| 222 | torch.save([e, u, x, y], 'data/{}.pt'.format(dataset)) |
| 223 | |
| 224 | elif dataset in ['arxiv']: |
| 225 | data = DglNodePropPredDataset('ogbn-arxiv') |
| 226 | g = data[0][0] |
| 227 | g = dgl.add_reverse_edges(g) |
| 228 | g = dgl.to_simple(g) |
| 229 | |
| 230 | e, u = eig_dgl_adj_sparse(g, sm=5000) |
| 231 | x = g.ndata['feat'] |
| 232 | y = data[0][1] |
| 233 | |
| 234 | torch.save([e, u, x, y], 'data/arxiv.pt') |
| 235 | |
| 236 | elif dataset in ['penn']: |
| 237 | g, x, y = load_fb100_dataset() |
| 238 | g = dgl.add_reverse_edges(g) |
| 239 | g = dgl.to_simple(g) |
| 240 | |
| 241 | e, u = eig_dgl_adj_sparse(g, sm=3000, lm=3000) |
| 242 | |
| 243 | torch.save([e, u, x, y], 'data/penn.pt') |
| 244 | |
| 245 | elif dataset in ['physics']: |
| 246 | datasets = Coauthor(root='./data', name='physics', transform=T.NormalizeFeatures()) |
| 247 | data = datasets[0] |
no test coverage detected