Merge branch 'master' of https://github.com/snap-stanford/ogb

Author: weihua916

examples/graphproppred/code2/utils.py

@@ -170,7 +170,7 @@ def decode_arr_to_seq(arr, idx2vocab):
     '''
 
 
-    eos_idx_list = torch.nonzero(arr == len(idx2vocab) - 1) # find the position of __EOS__ (the last vocab in idx2vocab)
+    eos_idx_list = torch.nonzero(arr == len(idx2vocab) - 1, as_tuple=False) # find the position of __EOS__ (the last vocab in idx2vocab)
     if len(eos_idx_list) > 0:
         clippted_arr = arr[: torch.min(eos_idx_list)] # find the smallest __EOS__
     else:

examples/lsc/mag240m/README.md

@@ -1,6 +1,7 @@
 # Baseline code for MAG-240M
 
-Please refer to the **[OGB-LSC paper](https://arxiv.org/abs/2103.09430)** for the detailed setting.
+- Please refer to the **[OGB-LSC paper](https://arxiv.org/abs/2103.09430)** for the detailed setting.
+- Baseline code based on **[DGL](https://www.dgl.ai/)** is available **[here](https://github.com/dmlc/dgl/tree/master/examples/pytorch/ogb_lsc/MAG240M)**.
 
 ## Installation requirements
 ```
@@ -139,7 +140,7 @@ python rgnn.py --device=0 --model=rgat --evaluate
 | Label Propagation | 58.44 | 56.29 | 0 | --- |
 | SGC | 65.82 | 65.29 | 0.7M | GeForce RTX 2080 Ti (11GB GPU) |
 | SIGN | 66.64 | 66.09 | 3.8M | GeForce RTX 2080 Ti (11GB GPU) |
-| MLP+C&S | 66.98 | 66.09 | 0.5M | GeForce RTX 2080 Ti (11GB GPU) |
+| MLP+C&S | 66.98 | 66.18 | 0.5M | GeForce RTX 2080 Ti (11GB GPU) |
 | GraphSAGE | 67.32 | 66.25 | 4.9M | GeForce RTX 2080 Ti (11GB GPU) |
 | GAT | 67.71 | 66.63 | 4.9M | GeForce RTX 2080 Ti (11GB GPU) |
 | R-GraphSAGE | 70.21 | 68.94 | 12.2M | GeForce RTX 2080 Ti (11GB GPU) |

examples/lsc/pcqm4m/README.md

@@ -1,6 +1,7 @@
 # Baseline code for PCQM4M-LSC
 
-Please refer to the **[OGB-LSC paper](https://arxiv.org/abs/2103.09430)** for the detailed setting.
+- Please refer to the **[OGB-LSC paper](https://arxiv.org/abs/2103.09430)** for the detailed setting.
+- Baseline code based on **[DGL](https://www.dgl.ai/)** is available **[here](https://github.com/dmlc/dgl/tree/master/examples/pytorch/ogb_lsc/PCQM4M)**.
 
 ## Installation requirements
 ```

examples/lsc/wikikg90m/dgl-ke-ogb-lsc/python/dglke/dataloader/KGDataset.py

@@ -222,17 +222,7 @@ def rmap_fname(self):
         return 'relations.dict'
 
 class KGDatasetWiki(KGDataset):
-    '''Load a knowledge graph FB15k
-
-    The FB15k dataset has five files:
-    * entities.dict stores the mapping between entity Id and entity name.
-    * relations.dict stores the mapping between relation Id and relation name.
-    * train.txt stores the triples in the training set.
-    * valid.txt stores the triples in the validation set.
-    * test.txt stores the triples in the test set.
-
-    The mapping between entity (relation) name and entity (relation) Id is stored as 'name\tid'.
-    The triples are stored as 'head_nid\trelation_id\ttail_nid'.
+    '''Load a knowledge graph wikikg
     '''
     def __init__(self, path, name='wikikg90m'):
         self.name = name

examples/lsc/wikikg90m/dgl-ke-ogb-lsc/python/dglke/dataloader/sampler.py

@@ -339,7 +339,7 @@ def __init__(self, dataset, args, ranks=64, has_importance=False):
         else:
             self.edge_parts = [np.arange(num_train)]
             self.rel_parts = [np.arange(dataset.n_relations)]
-            self.cross_part = False
+            self.cross_part = True
 
         self.g = ConstructGraph(triples, dataset.n_entities, args)
 

examples/lsc/wikikg90m/dgl-ke-ogb-lsc/python/dglke/train_pytorch.py

@@ -198,7 +198,7 @@ def test(args, model, test_samplers, step, rank=0, mode='Test'):
         for sampler in test_samplers:
             print(sampler.num_edges, sampler.batch_size)
             for query, ans, candidate in tqdm(sampler, disable=not args.print_on_screen, total=ceil(sampler.num_edges/sampler.batch_size)):
-                log = model.forward_test_wikikg(query, ans, candidate, sampler.mode, gpu_id)
+                log = model.forward_test_wikikg(query, ans, candidate, sampler.mode, gpu_id).cpu()
                 logs[sampler.mode].append(log)
                 answers[sampler.mode].append(ans)
         print("[{}] finished {} forward".format(rank, mode))

ogb/linkproppred/evaluate.py

@@ -232,7 +232,7 @@ def _eval_mrr(self, y_pred_pos, y_pred_neg, type_info):
         if type_info == 'torch':
             y_pred = torch.cat([y_pred_pos.view(-1,1), y_pred_neg], dim = 1)
             argsort = torch.argsort(y_pred, dim = 1, descending = True)
-            ranking_list = torch.nonzero(argsort == 0)
+            ranking_list = torch.nonzero(argsort == 0, as_tuple=False)
             ranking_list = ranking_list[:, 1] + 1
             hits1_list = (ranking_list <= 1).to(torch.float)
             hits3_list = (ranking_list <= 3).to(torch.float)

ogb/lsc/wikikg90m.py

@@ -206,7 +206,7 @@ def _calculate_mrr(self, correct_index, pred_top10):
             - pred_top10: shape (num_eval_triplets, 10)
         '''
         # extract indices where correct_index is within top10
-        tmp = torch.nonzero(correct_index.view(-1,1) == pred_top10)
+        tmp = torch.nonzero(correct_index.view(-1,1) == pred_top10, as_tuple=False)
 
         # reciprocal rank
         # if rank is larger than 10, then set the reciprocal rank to 0.