PR [#73] debug/gcn-gat-node -> dev

debug for gcn & gat node clf

autogl/module/model/encoders/_dgl/_gat.py

@@ -52,8 +52,6 @@ class GAT(torch.nn.Module):
     def forward(
             self, graph: dgl.DGLGraph, *__args, **__kwargs
     ) -> _typing.Iterable[torch.Tensor]:
-        graph = dgl.remove_self_loop(graph)
-        graph = dgl.add_self_loop(graph)
         num_layers = len(self.__convolutions)
         x: torch.Tensor = graph.ndata['feat']
         results = [x]

autogl/module/model/encoders/_dgl/_gcn.py

@@ -26,8 +26,6 @@ class _GCN(torch.nn.Module):
         self._dropout: _typing.Optional[float] = dropout
 
     def forward(self, graph: dgl.DGLGraph, *__args, **__kwargs):
-        graph = dgl.remove_self_loop(graph)
-        graph = dgl.add_self_loop(graph)
         x: torch.Tensor = graph.ndata['feat']
         results: _typing.MutableSequence[torch.Tensor] = []
         for _layer in range(len(self.__convolution_layers)):

test/performance/node_classification/dgl/helper.py

@@ -15,7 +15,7 @@ def get_encoder_decoder_hp(model='gin', decoder=None):
             "hidden": [8],
             "heads": 8,
             "dropout": 0.6,
-            "act": "elu",
+            "act": "relu",
         }
     elif model == 'gcn':
         model_hp = {

test/performance/node_classification/dgl/model.py

@@ -82,8 +82,8 @@ if __name__ == '__main__':
 
         if args.model == 'gat':
             model = AutoGAT(
-                num_features=num_features,
-                num_classes=num_classes,
+                input_dimension=num_features,
+                output_dimension=num_classes,
                 device=args.device
             ).from_hyper_parameter({
                 # hp from model
@@ -96,8 +96,8 @@ if __name__ == '__main__':
             }).model
         elif args.model == 'gcn':
             model = AutoGCN(
-                num_features=num_features,
-                num_classes=num_classes,
+                input_dimension=num_features,
+                output_dimension=num_classes,
                 device=args.device
             ).from_hyper_parameter({
                 "num_layers": 2,
@@ -107,8 +107,8 @@ if __name__ == '__main__':
             }).model
         elif args.model == 'sage':
             model = AutoSAGE(
-                num_features=num_features,
-                num_classes=num_classes,
+                input_dimension=num_features,
+                output_dimension=num_classes,
                 device=args.device
             ).from_hyper_parameter({
                 "num_layers": 2,

test/performance/node_classification/pyg/helper.py

@@ -1,9 +1,9 @@
-def get_encoder_decoder_hp(model='gin', decoder=None):
+def get_encoder_decoder_hp(model='gin', decoder=None, decoupled=False):
     if model == 'gin':
         model_hp = {
             # hp from model
-            "num_layers": 5,
-            "hidden": [64,64,64,64],
+            "num_layers": 2,
+            "hidden": [64],
             "dropout": 0.5,
             "act": "relu",
             "eps": "False",
@@ -11,14 +11,23 @@ def get_encoder_decoder_hp(model='gin', decoder=None):
         }
 
     if model == 'gat':
-        model_hp = {
-            "num_layers": 2,
-            "hidden": [8],
-            "num_hidden_heads": 8,
-            "num_output_heads": 8,
-            "dropout": 0.6,
-            "act": "elu"
-        }
+        if decoupled:
+            model_hp = {
+                "num_layers": 2,
+                "hidden": [8],
+                "num_hidden_heads": 8,
+                "num_output_heads": 1,
+                "dropout": 0.6,
+                "act": "elu"
+            }
+        else:
+            model_hp = {
+                "num_layers": 2,
+                "hidden": [8],
+                "heads": 8,
+                "dropout": 0.6,
+                "act": "elu"
+            }
     elif model == 'gcn':
         model_hp = {
             "num_layers": 2,

test/performance/node_classification/pyg/model_decouple.py

@@ -87,7 +87,7 @@ if __name__ == '__main__':
 
     accs = []
 
-    model_hp, decoder_hp = get_encoder_decoder_hp(args.model)
+    model_hp, decoder_hp = get_encoder_decoder_hp(args.model, decoupled=True)
 
     for seed in tqdm(range(args.repeat)):
         set_seed(seed)

test/performance/node_classification/pyg/solver.py

@@ -42,7 +42,7 @@ if __name__ == '__main__':
     label = dataset[0].nodes.data['y'][dataset[0].nodes.data['test_mask']].numpy()
     accs = []
 
-    model_hp, decoder_hp = get_encoder_decoder_hp(args.model)
+    model_hp, decoder_hp = get_encoder_decoder_hp(args.model, decoupled=True)
 
     for seed in tqdm(range(args.repeat)):
 

test/performance/node_classification/pyg/trainer.py

@@ -40,12 +40,10 @@ if __name__ == '__main__':
 
     accs = []
 
-    model_hp, decoder_hp = get_encoder_decoder_hp(args.model)
+    model_hp, decoder_hp = get_encoder_decoder_hp(args.model, decoupled=True)
     for seed in tqdm(range(args.repeat)):
         set_seed(seed)
 
-        model_hp, decoder_hp = get_encoder_decoder_hp(args.model)
-
         trainer = NodeClassificationFullTrainer(
             model=args.model,
             num_features=num_features,

test/performance/node_classification/pyg/trainer_dataset.py

@@ -40,7 +40,7 @@ if __name__ == '__main__':
 
     accs = []
 
-    model_hp, decoder_hp = get_encoder_decoder_hp(args.model)
+    model_hp, decoder_hp = get_encoder_decoder_hp(args.model, decoupled=True)
     for seed in tqdm(range(args.repeat)):
         set_seed(seed)