update config

Author: haleqiu

config/DataIncremental/BaselineDatalifelongCiteseer.yaml

@@ -0,0 +1,15 @@
+dataset=citeseer
+model=GAT #MLP APPNP SAGE GCN
+device=cuda:0
+seed=1
+
+sample=10
+jump=1
+memory=512
+batch=10
+opt=SGD
+lr=0.01
+iteration=5
+
+hidden =[64, 32]
+drop = [0, 0]

config/DataIncremental/FGNDatalifelongOGB.yaml

@@ -0,0 +1,16 @@
+dataset=ogbn-arxiv
+model=KTransCat
+device=cuda:0
+seed=1
+
+sample=100
+jump=1
+memory=512
+batch=16
+opt=SGD
+lr=0.01
+iteration=5
+k=1
+
+hidden =[128, 128]
+drop = [0.3, 0.1]

config/Regular/FGNRegularCiteseer.yaml

@@ -8,6 +8,4 @@ opt=SGD
 lr=0.01
 
 hidden =[10, 10]
-drop = [0.3, 0.1]
-
-# save='loads/KTransCat_class_ogbn-arxiv_SGD_M4096_J5_1_B16.model', seed=1) number of prarams 70960
+drop = [0.3, 0.1]

config/Regular/FGNRegularOGB.yaml

@@ -1,5 +1,5 @@
 dataset=ogbn-arxiv
-model=attnKTransCat
+model=KTransCat
 device=cuda:0
 seed=1
 
@@ -9,6 +9,4 @@ lr=0.01
 k=1
 
 hidden =[128, 128]
-drop = [0.3, 0.1]
-
-# save='loads/KTransCat_class_ogbn-arxiv_SGD_M4096_J5_1_B16.model', seed=1) number of prarams 70960
+drop = [0.3, 0.1]

torch_util/tools.py

@@ -83,4 +83,25 @@ def performance(loader, net, device, k):
             total += targets.size(0)
             correct += predicted.eq(targets.data).cpu().sum().item()
         acc = correct/total
+    return acc
+
+
+def accuracy(net, loader, device, num_class):
+    net.eval()
+    correct, total = 0, 0
+    classes = torch.arange(num_class).view(-1,1).to(device)
+    with torch.no_grad():
+        for idx, (inputs, targets, neighbor) in enumerate(loader):
+            if torch.cuda.is_available():
+                inputs, targets = inputs.to(device), targets.to(device)
+                if not k:
+                    neighbor = [element.to(device) for element in neighbor]
+                else:
+                    neighbor = [[item.to(device) for item in element] for element in neighbor]
+            outputs = net(inputs, neighbor)
+            _, predicted = torch.max(outputs.data, 1)
+            total += (targets == classes).sum(1)
+            corrected = predicted==targets
+            correct += torch.stack([corrected[targets==i].sum() for i in range(num_class)])
+        acc = correct/total
     return acc