update comments (#2132)

Co-authored-by: Ubuntu <ubuntu@ip-172-31-51-214.ec2.internal>

examples/pytorch/gcmc/train_sampling.py

@@ -403,6 +403,8 @@ if __name__ == '__main__':
         run(0, n_gpus, args, devices, dataset)
     # multi gpu
     else:
+        # Create csr/coo/csc formats before launching training processes with multi-gpu.
+        # This avoids creating certain formats in each sub-process, which saves momory and CPU.
         dataset.train_enc_graph.create_formats_()
         dataset.train_dec_graph.create_formats_()
         procs = []

examples/pytorch/graphsage/train_cv_multi_gpu.py

@@ -384,6 +384,8 @@ if __name__ == '__main__':
     g.ndata['features'] = features.share_memory_()
     create_history_storage(g, args, n_classes)
 
+    # Create csr/coo/csc formats before launching training processes with multi-gpu.
+    # This avoids creating certain formats in each sub-process, which saves momory and CPU.
     g.create_formats_()
     # Pack data
     data = train_mask, val_mask, in_feats, labels, n_classes, g

examples/pytorch/graphsage/train_sampling.py

@@ -229,6 +229,8 @@ if __name__ == '__main__':
     else:
         train_g = val_g = test_g = g
 
+    # Create csr/coo/csc formats before launching training processes with multi-gpu.
+    # This avoids creating certain formats in each sub-process, which saves momory and CPU.
     train_g.create_formats_()
     val_g.create_formats_()
     test_g.create_formats_()

examples/pytorch/graphsage/train_sampling_multi_gpu.py

@@ -258,6 +258,8 @@ if __name__ == '__main__':
     else:
         train_g = val_g = test_g = g
 
+    # Create csr/coo/csc formats before launching training processes with multi-gpu.
+    # This avoids creating certain formats in each sub-process, which saves momory and CPU.
     train_g.create_formats_()
     val_g.create_formats_()
     test_g.create_formats_()

examples/pytorch/graphsage/train_sampling_unsupervised.py

@@ -298,6 +298,9 @@ def main(args, devices):
     val_mask = g.ndata['val_mask']
     test_mask = g.ndata['test_mask']
     g.ndata['features'] = features
+
+    # Create csr/coo/csc formats before launching training processes with multi-gpu.
+    # This avoids creating certain formats in each sub-process, which saves momory and CPU.
     g.create_formats_()
     # Pack data
     data = train_mask, val_mask, test_mask, in_feats, labels, n_classes, g

examples/pytorch/ogb/cluster-sage/sampler.py

@@ -63,9 +63,4 @@ class ClusterIter(object):
 def subgraph_collate_fn(g, batch):
     nids = np.concatenate(batch).reshape(-1).astype(np.int64)
     g1 = g.subgraph(nids)
-    nid = g1.ndata[dgl.NID]
-    g1.ndata['feat'] = g.ndata['feat'][nid]
-    g1.ndata['labels'] = g.ndata['labels'][nid]
-    g1.ndata['train_mask'] = g.ndata['train_mask'][nid]
-    g1.create_formats_()
     return g1

examples/pytorch/ogb/ogbn-products/gat/main.py

@@ -243,6 +243,9 @@ if __name__ == '__main__':
 
     in_feats = graph.ndata['feat'].shape[1]
     n_classes = (labels.max() + 1).item()
+
+    # Create csr/coo/csc formats before launching sampling processes
+    # This avoids creating certain formats in each data loader process, which saves momory and CPU.
     graph.create_formats_()
     # Pack data
     data = train_idx, val_idx, test_idx, in_feats, labels, n_classes, graph, args.head

examples/pytorch/ogb/ogbn-products/graphsage/main.py

@@ -234,6 +234,8 @@ if __name__ == '__main__':
 
     in_feats = graph.ndata['feat'].shape[1]
     n_classes = (labels.max() + 1).item()
+    # Create csr/coo/csc formats before launching sampling processes
+    # This avoids creating certain formats in each data loader process, which saves momory and CPU.
     graph.create_formats_()
     # Pack data
     data = train_idx, val_idx, test_idx, in_feats, labels, n_classes, graph

examples/pytorch/rgcn/entity_classify_mp.py

@@ -504,6 +504,9 @@ def main(args, devices):
     train_idx.share_memory_()
     val_idx.share_memory_()
     test_idx.share_memory_()
+    # Create csr/coo/csc formats before launching training processes with multi-gpu.
+    # This avoids creating certain formats in each sub-process, which saves momory and CPU.
+    g.create_formats_()
 
     n_gpus = len(devices)
     # cpu