When i train the R-GCN model with my own dataset, the RuntimeError occurs:
RuntimeError Traceback (most recent call last)
in ()
78 print(“start eval”)
79 mrr = utils.evaluate(test_graph, model, valid_data, num_nodes,
—> 80 hits=[1, 3, 10], eval_bz=eval_batch_size)
81 # save best model
82 if mrr < best_mrr:
~/R-GCN/utils.py in evaluate(test_graph, model, test_triplets, num_entity, hits, eval_bz)
198 ranks_s = perturb_and_get_rank(embedding, w, o, r, s, num_entity, eval_bz)
199 # perturb object
–> 200 ranks_o = perturb_and_get_rank(embedding, w, s, r, o, num_entity, eval_bz)
201
202 ranks = torch.cat([ranks_s, ranks_o])
~/R-GCN/utils.py in perturb_and_get_rank(embedding, w, a, r, b, num_entity, batch_size)
183 score = torch.sigmoid(score)
184 target = b[batch_start: batch_end]
–> 185 ranks.append(sort_and_rank(score, target))
186 return torch.cat(ranks)
187
~/R-GCN/utils.py in sort_and_rank(score, target)
160 def sort_and_rank(score, target):
161 _, indices = torch.sort(score, dim=1, descending=True)
–> 162 indices = torch.nonzero(indices == target.view(-1, 1))
163 indices = indices[:, 1].view(-1)
164 return indices
RuntimeError: Trying to resize storage that is not resizable at /pytorch/aten/src/TH/THStorageFunctions.cpp:76
But it’s totally fine when using FB15k-237 datasets.
my datasets has the same format with FB15k-237, besides there is only one relationship in my dataset which is encoded : 0. i’m not sure if there any problem with my own datasets, or the codes.