[Example] Experimental results on ogbl-citation & ogbl-ppa (#1711)

* ogb-deepwalk

* update readme

* update readme

* update readme

* update readme

* ogbl-ddi

* readme

* citation&ppa
Co-authored-by: xiang song(charlie.song) <classicxsong@gmail.com>

examples/pytorch/ogb/deepwalk/README.md

@@ -5,13 +5,12 @@ python3 load_dataset.py --name ogbl-collab
 ```
 
 ## Evaluation
-For evaluatation we follow the code provided by ogb [here](https://github.com/snap-stanford/ogb/blob/master/examples/linkproppred/collab/mlp.py).
+For evaluatation we follow the code mlp.py provided by ogb [here](https://github.com/snap-stanford/ogb/blob/master/examples/linkproppred/collab/mlp.py).
 
 ## Used config
 ogbl-collab
 ```
 python3 deepwalk.py --data_file ogbl-collab-net.txt --save_in_pt --output_emb_file embedding.pt --num_walks 50 --window_size 20 --walk_length 40 --lr 0.1 --negative 1 --neg_weight 1 --lap_norm 0.005 --mix --adam --gpus 0 1 2 3 --num_threads 4 --print_interval 2000 --print_loss --batch_size 32
-
 cd ./ogb/blob/master/examples/linkproppred/collab/
 cp embedding_pt_file_path ./
 python3 mlp.py --device 0 --runs 10 --use_node_embedding
@@ -25,6 +24,19 @@ cp embedding_pt_file_path ./
 python3 mlp.py --device 0 --runs 5
 ```
 
+ogbl-ppa
+```
+python3 deepwalk.py --data_file ogbl-ppa-net.txt --save_in_pt --output_emb_file ppa-embedding.pt --negative 1 --neg_weight 1 --batch_size 64 --print_interval 2000 --print_loss --window_size 2 --num_walks 30 --walk_length 80 --lr 0.1 --lap_norm 0.02 --adam --mix --gpus 0 1 --use_context_weight --num_threads 4
+cp embedding_pt_file_path ./
+python3 mlp.py --device 2 --runs 10
+```
+
+ogbl-citation
+```
+python3 deepwalk.py --data_file ogbl-citation-net.txt --save_in_pt --output_emb_file embedding.pt --window_size 2 --num_walks 10 --negative 1 --neg_weight 1 --walk_length 80 --batch_size 128 --print_loss --print_interval 1000 --mix --adam --gpus 0 1 2 3 --use_context_weight --num_threads 4 --lap_norm 0.05 --lr 0.1
+cp embedding_pt_file_path ./
+python3 mlp.py --device 2 --runs 5 --use_node_embedding
+```
 
 ## Score
 ogbl-collab
@@ -61,3 +73,27 @@ ogbl-collab
 <br>&emsp;&emsp;Final Train: 52.28 ± 1.21
 <br>&emsp;&emsp;Final Test: 29.13 ± 3.46
 
+
+<br>ogbl-ppa
+<br>Hits@10
+<br>&emsp;Highest Train: 3.58 ± 0.90
+<br>&emsp;Highest Valid: 2.88 ± 0.76
+<br>&emsp;&emsp;Final Train: 3.58 ± 0.90
+<br>&emsp;&emsp;Final Test: 1.45 ± 0.65
+<br>&emsp;Hits@50
+<br>&emsp;Highest Train: 18.21 ± 2.29
+<br>&emsp;Highest Valid: 15.75 ± 2.10
+<br>&emsp;&emsp;Final Train: 18.21 ± 2.29
+<br>&emsp;&emsp;Final Test: 11.70 ± 0.97
+<br>&emsp;Hits@100
+<br>&emsp;Highest Train: 31.16 ± 2.23
+<br>&emsp;Highest Valid: 27.52 ± 2.07
+<br>&emsp;&emsp;Final Train: 31.16 ± 2.23
+<br>&emsp;&emsp;Final Test: 23.02 ± 1.63
+
+<br>ogbl-citation
+<br>MRR
+<br>&emsp;Highest Train: 0.8796 ± 0.0007
+<br>&emsp;Highest Valid: 0.8141 ± 0.0007
+<br>&emsp;&emsp;Final Train: 0.8793 ± 0.0008
+<br>&emsp;&emsp;Final Test: 0.8159 ± 0.0006

examples/pytorch/ogb/deepwalk/load_dataset.py

@@ -4,8 +4,10 @@ from ogb.linkproppred import PygLinkPropPredDataset
 import argparse
 
 parser = argparse.ArgumentParser()
-parser.add_argument('--name', type=str, choices=['ogbl-collab', 'ogbl-ddi', 'ogbl-ppa'], default='ogbl-collab',
-        help="name of datasets by ogb") 
+parser.add_argument('--name', type=str,
+    choices=['ogbl-collab', 'ogbl-ddi', 'ogbl-ppa', 'ogbl-citation'], 
+    default='ogbl-collab',
+    help="name of datasets by ogb") 
 args = parser.parse_args()
 
 name = args.name
@@ -19,7 +21,6 @@ try:
 except:
     weighted = False
 
-
 with open(name + "-net.txt", "w") as f:
     for i in range(data.edge_index.shape[1]):
         if weighted:

examples/pytorch/ogb/deepwalk/model.py

@@ -481,7 +481,7 @@ class SkipGramModel(nn.Module):
         return torch.sum(score), torch.sum(neg_score)
 
     def save_embedding(self, dataset, file_name):
-        """ Write embedding to local file.
+        """ Write embedding to local file. Only used when node ids are numbers.
 
         Parameter
         ---------
@@ -494,14 +494,19 @@ class SkipGramModel(nn.Module):
         np.save(file_name, embedding)
 
     def save_embedding_pt(self, dataset, file_name):
-        """ 
+        """ For ogb leaderboard.
         """
-        embedding = self.u_embeddings.weight.cpu().data
-        assert max(dataset.node2id.keys()) == self.emb_size - 1, "The node id does not starts from 0, saving embedding failed."
-        index = torch.LongTensor(list(map(lambda node: dataset.node2id[node], list(range(self.emb_size)))))
-        embedding = torch.index_select(embedding, 0, index)
+        max_node_id = max(dataset.node2id.keys())
+        if max_node_id + 1 != self.emb_size:
+            print("WARNING: The node ids are not serial.")
+
+        embedding = torch.zeros(max_node_id + 1, self.emb_dimension)
+        index = torch.LongTensor(list(map(lambda id: dataset.id2node[id], list(range(self.emb_size)))))
+        embedding.index_add_(0, index, self.u_embeddings.weight.cpu().data)
+
         if self.norm:
-            embedding /= torch.sqrt(torch.sum(embedding.mul(embedding), 1)).unsqueeze(1)
+            embedding /= torch.sqrt(torch.sum(embedding.mul(embedding), 1) + 1e-6).unsqueeze(1)
+
         torch.save(embedding, file_name)
 
     def save_embedding_txt(self, dataset, file_name):