Add static data and model for hybrid link gnn

hybridgnn/nn/models/hybridgnn.py

@@ -35,6 +35,7 @@ def __init__(
         norm: str = 'layer_norm',
         torch_frame_model_cls: Type[torch.nn.Module] = ResNet,
         torch_frame_model_kwargs: Optional[Dict[str, Any]] = None,
+        is_static: Optional[bool] = False,
     ) -> None:
         super().__init__(data, col_stats_dict, rhs_emb_mode, dst_entity_table,
                          num_nodes, embedding_dim)
@@ -76,6 +77,7 @@ def __init__(
         self.lin_offset_idgnn = torch.nn.Linear(embedding_dim, 1)
         self.lin_offset_embgnn = torch.nn.Linear(embedding_dim, 1)
         self.channels = channels
+        self.is_static = is_static
 
         self.reset_parameters()
 
@@ -103,11 +105,12 @@ def forward(
         # Add ID-awareness to the root node
         x_dict[entity_table][:seed_time.size(0
                                              )] += self.id_awareness_emb.weight
-        rel_time_dict = self.temporal_encoder(seed_time, batch.time_dict,
-                                              batch.batch_dict)
+        if self.is_static is not True:
+            rel_time_dict = self.temporal_encoder(seed_time, batch.time_dict,
+                                                  batch.batch_dict)
 
-        for node_type, rel_time in rel_time_dict.items():
-            x_dict[node_type] = x_dict[node_type] + rel_time
+            for node_type, rel_time in rel_time_dict.items():
+                x_dict[node_type] = x_dict[node_type] + rel_time
 
         x_dict = self.gnn(
             x_dict,

hybridgnn/nn/models/idgnn.py

@@ -29,6 +29,7 @@ def __init__(
         norm: str = 'layer_norm',
         torch_frame_model_cls: Type[torch.nn.Module] = ResNet,
         torch_frame_model_kwargs: Optional[Dict[str, Any]] = None,
+        is_static=True,
     ) -> None:
         super().__init__()
 
@@ -65,6 +66,7 @@ def __init__(
         )
 
         self.id_awareness_emb = torch.nn.Embedding(1, channels)
+        self.is_static = is_static
         self.reset_parameters()
 
     def reset_parameters(self) -> None:
@@ -85,11 +87,12 @@ def forward(
         # Add ID-awareness to the root node
         x_dict[entity_table][:seed_time.size(0
                                              )] += self.id_awareness_emb.weight
-        rel_time_dict = self.temporal_encoder(seed_time, batch.time_dict,
-                                              batch.batch_dict)
+        if self.is_static is not True:
+            rel_time_dict = self.temporal_encoder(seed_time, batch.time_dict,
+                                                  batch.batch_dict)
 
-        for node_type, rel_time in rel_time_dict.items():
-            x_dict[node_type] = x_dict[node_type] + rel_time
+            for node_type, rel_time in rel_time_dict.items():
+                x_dict[node_type] = x_dict[node_type] + rel_time
 
         x_dict = self.gnn(
             x_dict,

hybridgnn/nn/models/shallowrhsgnn.py

@@ -35,6 +35,7 @@ def __init__(
         norm: str = 'layer_norm',
         torch_frame_model_cls: Type[torch.nn.Module] = ResNet,
         torch_frame_model_kwargs: Optional[Dict[str, Any]] = None,
+        is_static: Optional[bool] = False,
     ) -> None:
         super().__init__(data, col_stats_dict, rhs_emb_mode, dst_entity_table,
                          num_nodes, embedding_dim)
@@ -71,6 +72,8 @@ def __init__(
         )
         self.lhs_projector = torch.nn.Linear(channels, embedding_dim)
         self.id_awareness_emb = torch.nn.Embedding(1, channels)
+        self.is_static = is_static
+
         self.reset_parameters()
 
     def reset_parameters(self) -> None:
@@ -94,11 +97,13 @@ def forward(
         # Add ID-awareness to the root node
         x_dict[entity_table][:seed_time.size(0
                                              )] += self.id_awareness_emb.weight
-        rel_time_dict = self.temporal_encoder(seed_time, batch.time_dict,
-                                              batch.batch_dict)
 
-        for node_type, rel_time in rel_time_dict.items():
-            x_dict[node_type] = x_dict[node_type] + rel_time
+        if self.is_static is not True:
+            rel_time_dict = self.temporal_encoder(seed_time, batch.time_dict,
+                                                  batch.batch_dict)
+
+            for node_type, rel_time in rel_time_dict.items():
+                x_dict[node_type] = x_dict[node_type] + rel_time
 
         x_dict = self.gnn(
             x_dict,

static_data/amazon-book/README.md

@@ -0,0 +1 @@
+Look for the full dataset? Please visit the [websit](http://jmcauley.ucsd.edu/data/amazon).