Add files using upload-large-folder tool

deeprobust/graph/defense/gcn_guard.py

@@ -25,7 +25,7 @@ from scipy.sparse import lil_matrix
 
 class GCNGuard(nn.Module):
 
-    def __init__(self, nfeat, nhid, nclass, dropout=0.5, lr=0.01, drop=False, weight_decay=5e-4, with_relu=True,
+    def __init__(self, nfeat, nhid, nclass, dropout=0.5, lr=0.01, drop=True, weight_decay=5e-4, with_relu=True,
                  with_bias=True, device=None):
 
         super(GCNGuard, self).__init__()
@@ -102,7 +102,7 @@ class GCNGuard(nn.Module):
     def forward(self, x, adj):
         """we don't change the edge_index, just update the edge_weight;
         some edge_weight are regarded as removed if it equals to zero"""
-        x = x.to_dense()  # topology attack中需要注销掉(wisconsin不需要), 在meta attack还有 dice中均需要
+        x = x.to_dense()  
 
         """GCN and GAT"""
         if self.attention:
@@ -189,7 +189,7 @@ class GCNGuard(nn.Module):
         if self.drop:
             character = np.vstack((att_dense_norm[row, col].A1,
                                      att_dense_norm[col, row].A1))
-            character = torch.from_numpy(character.T)
+            character = torch.from_numpy(character.T).to(self.device)
             drop_score = self.drop_learn_1(character)
             drop_score = torch.sigmoid(drop_score)  # do not use softmax since we only have one element
             mm = torch.nn.Threshold(0.5, 0)

deeprobust/graph/defense/gcn_preprocess.py

@@ -6,6 +6,7 @@ from torch.nn.parameter import Parameter
 from torch.nn.modules.module import Module
 from deeprobust.graph import utils
 from deeprobust.graph.defense import GCN
+from sklearn.utils.extmath import randomized_svd
 from tqdm import tqdm
 import scipy.sparse as sp
 import numpy as np
@@ -104,37 +105,86 @@ class GCNSVD(GCN):
         self.labels = labels
         super().fit(features, modified_adj, labels, idx_train, idx_val, train_iters=train_iters, initialize=initialize, verbose=verbose)
 
-    def truncatedSVD(self, data, k=50):
-        """Truncated SVD on input data.
-
-        Parameters
-        ----------
-        data :
-            input matrix to be decomposed
-        k : int
-            number of singular values and vectors to compute.
+    # def truncatedSVD(self, data, k=50):
+    #     """Truncated SVD on input data.
+
+    #     Parameters
+    #     ----------
+    #     data :
+    #         input matrix to be decomposed
+    #     k : int
+    #         number of singular values and vectors to compute.
+
+    #     Returns
+    #     -------
+    #     numpy.array
+    #         reconstructed matrix.
+    #     """
+    #     print('=== GCN-SVD: rank={} ==='.format(k))
+    #     print("stage1")
+    #     if sp.issparse(data):
+    #         data = data.asfptype()
+    #         U, S, V = sp.linalg.svds(data, k=k)
+    #         print("rank_after = {}".format(len(S.nonzero()[0])))
+    #         diag_S = np.diag(S)
+    #     else:
+    #         U, S, V = np.linalg.svd(data)
+    #         U = U[:, :k]
+    #         S = S[:k]
+    #         V = V[:k, :]
+    #         print("rank_before = {}".format(len(S.nonzero()[0])))
+    #         diag_S = np.diag(S)
+    #         print("rank_after = {}".format(len(diag_S.nonzero()[0])))
+
+    #     return U @ diag_S @ V
 
-        Returns
-        -------
-        numpy.array
-            reconstructed matrix.
-        """
-        print('=== GCN-SVD: rank={} ==='.format(k))
-        if sp.issparse(data):
-            data = data.asfptype()
-            U, S, V = sp.linalg.svds(data, k=k)
-            print("rank_after = {}".format(len(S.nonzero()[0])))
-            diag_S = np.diag(S)
-        else:
-            U, S, V = np.linalg.svd(data)
-            U = U[:, :k]
-            S = S[:k]
-            V = V[:k, :]
-            print("rank_before = {}".format(len(S.nonzero()[0])))
-            diag_S = np.diag(S)
-            print("rank_after = {}".format(len(diag_S.nonzero()[0])))
+    def truncatedSVD(self, data, k=50):
+        """Fast truncated SVD using randomized algorithm"""
+        print(f'=== Fast SVD: rank={k} ===')
+
+        # Convert torch Tensor to NumPy
+        if isinstance(data, torch.Tensor):
+            data = data.detach().cpu().numpy()
+
+        U, S, Vt = randomized_svd(data, n_components=k, n_iter=10, random_state=None)
+        print("rank_after =", np.count_nonzero(S))
+        diag_S = np.diag(S)
+        return U @ diag_S @ Vt
+
+    # def truncatedSVD(self, data, k=50):
+
+    #     """Truncated SVD on input data using GPU (torch.linalg.svd), returning numpy array."""
+    #     print('=== GCN-SVD (GPU): rank={} ==='.format(k))
+    #     # 转换为 dense 并放到 GPU
+    #     if sp.issparse(data):
+    #         print("[Info] Converting sparse matrix to dense (may be memory-intensive).")
+    #         data = torch.from_numpy(data.toarray()).float().to(self.device)
+    #     elif isinstance(data, np.ndarray):
+    #         data = torch.from_numpy(data).float().to(self.device)
+    #     elif isinstance(data, torch.Tensor):
+    #         data = data.float().to(self.device)
+    #     else:
+    #         raise TypeError("Unsupported data type for SVD.")
+        
+    #     print("stage1")
+    #     # 
+    #     U, S, Vh = torch.linalg.svd(data, full_matrices=False)
+       
+    #     print("stage2")
+    #     # 
+    #     U_k = U[:, :k]
+    #     S_k = S[:k]
+    #     V_k = Vh[:k, :]
+
+    #     print(f"Singular values used (non-zero count): {(S_k != 0).sum().item()} / {len(S_k)}")
+
+    #     # 还原矩阵：U * S * V
+    #     diag_S = torch.diag(S_k)
+    #     reconstructed = U_k @ diag_S @ V_k
+
+    #     # 返回 numpy 类型（保持一致）
+    #     return reconstructed.detach().cpu().numpy()
 
-        return U @ diag_S @ V
 
     def predict(self, features=None, adj=None):
         """By default, the inputs should be unnormalized adjacency
@@ -276,6 +326,7 @@ class GCNJaccard(GCN):
             removed_cnt = dropedge_dis(adj_triu.data, adj_triu.indptr, adj_triu.indices, features, threshold=self.threshold)
         else:
             if self.binary_feature:
+                print("self.binary_feature", self.binary_feature)
                 removed_cnt = dropedge_jaccard(adj_triu.data, adj_triu.indptr, adj_triu.indices, features, threshold=self.threshold)
             else:
                 removed_cnt = dropedge_cosine(adj_triu.data, adj_triu.indptr, adj_triu.indices, features, threshold=self.threshold)
@@ -378,6 +429,22 @@ def __dropedge_jaccard(A, iA, jA, features, threshold):
                 removed_cnt += 1
     return removed_cnt
 
+# @njit
+# def dropedge_jaccard(A, iA, jA, features, threshold):
+#     removed_cnt = 0
+#     for row in range(len(iA)-1):
+#         for i in range(iA[row], iA[row+1]):
+#             # print(row, jA[i], A[i])
+#             n1 = row
+#             n2 = jA[i]
+#             a, b = features[n1], features[n2]
+#             intersection = np.count_nonzero(a*b)
+#             J = intersection * 1.0 / (np.count_nonzero(a) + np.count_nonzero(b) - intersection)
+#             if J < threshold:
+#                 A[i] = 0
+#                 # A[n2, n1] = 0
+#                 removed_cnt += 1
+#     return removed_cnt
 @njit
 def dropedge_jaccard(A, iA, jA, features, threshold):
     removed_cnt = 0
@@ -388,7 +455,12 @@ def dropedge_jaccard(A, iA, jA, features, threshold):
             n2 = jA[i]
             a, b = features[n1], features[n2]
             intersection = np.count_nonzero(a*b)
-            J = intersection * 1.0 / (np.count_nonzero(a) + np.count_nonzero(b) - intersection)
+            union = np.count_nonzero(a) + np.count_nonzero(b) - intersection
+            if union == 0:
+                print("Running safe dropedge_jaccard")
+                J = 0  
+            else:
+                J = intersection * 1.0 / union
 
             if J < threshold:
                 A[i] = 0

deeprobust/graph/global_attack/base_attack.py

@@ -71,6 +71,7 @@ class BaseAttack(Module):
         assert torch.abs(adj - adj.t()).sum() == 0, "Input graph is not symmetric"
         assert adj.max() == 1, "Max value should be 1!"
         assert adj.min() == 0, "Min value should be 0!"
+        adj.fill_diagonal_(0)
         diag = adj.diag()
         assert diag.max() == 0, "Diagonal should be 0!"
         assert diag.min() == 0, "Diagonal should be 0!"

deeprobust/graph/global_attack/topology_attack.py

@@ -79,7 +79,7 @@ class PGDAttack(BaseAttack):
 
         self.complementary = None
 
-    def attack(self, ori_features, ori_adj, labels, idx_train, n_perturbations, epochs=25, **kwargs):
+    def attack(self, ori_features, ori_adj, labels, idx_train, n_perturbations, epochs=50, **kwargs):
         """Generate perturbations on the input graph.
 
         Parameters

examples/graph/cgscore_experiments/defense_method/GAT.py

@@ -6,19 +6,23 @@ from deeprobust.graph.defense import GCNJaccard, GCN, GAT
 from deeprobust.graph.utils import *
 from deeprobust.graph.data import Dataset, PrePtbDataset, Dpr2Pyg
 import argparse
+import csv
+import os
 
 parser = argparse.ArgumentParser()
 parser.add_argument('--seed', type=int, default=15, help='Random seed.')
-parser.add_argument('--dataset', type=str, default='cora', choices=['cora', 'cora_ml', 'citeseer', 'polblogs', 'pubmed'], help='dataset')
-parser.add_argument('--ptb_rate', type=float, default=0.0,  help='pertubation rate')
-parser.add_argument('--ptb_type', type=str, default='clean', choices=['clean', 'meta', 'dice', 'minmax', 'pgd', 'random'], help='attack type')
+parser.add_argument('--dataset', type=str, default='pubmed', choices=['cora', 'cora_ml', 'citeseer', 'polblogs', 'pubmed', 'Flickr'], help='dataset')
+parser.add_argument('--ptb_rate', type=float, default=0.25,  help='pertubation rate')
+parser.add_argument('--ptb_type', type=str, default='dice', choices=['clean', 'meta', 'dice', 'minmax', 'pgd', 'random'], help='attack type')
 parser.add_argument('--hidden', type=int, default=16, help='Number of hidden units.')
 parser.add_argument('--dropout', type=float, default=0.5, help='Dropout rate (1 - keep probability).')
+parser.add_argument('--gpu', type=int, default=0, help='GPU device ID (default: 0)')
 
 args = parser.parse_args()
 args.cuda = torch.cuda.is_available()
 print('cuda: %s' % args.cuda)
-device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+device = torch.device(f"cuda:{args.gpu}" if torch.cuda.is_available() else "cpu")
+
 
 # make sure you use the same data splits as you generated attacks
 np.random.seed(args.seed)
@@ -57,5 +61,23 @@ print('=== train on clean graph ===')
 
 pyg_data = Dpr2Pyg(data)
 pyg_data.update_edge_index(perturbed_adj) # inplace operation
-gat.fit(pyg_data, train_iters=200, verbose=True) # train with earlystopping
-gat.test()
+gat.fit(pyg_data, train_iters=500, verbose=True) # train with earlystopping
+acc = gat.test()
+print("acc:", acc)
+
+csv_dir = "../result"
+os.makedirs(csv_dir, exist_ok=True) 
+
+csv_filename = os.path.join(csv_dir, f"GAT_{args.dataset}_{args.ptb_type}_{args.ptb_rate}.csv")
+row = [f"{args.dataset} ", f" {args.ptb_type} ", f" {args.ptb_rate} ", f" {acc}"]
+
+try:
+    file_exists = os.path.isfile(csv_filename)
+    with open(csv_filename, 'a', newline='') as csvfile:
+        writer = csv.writer(csvfile)
+        if not file_exists:
+            writer.writerow(["dataset ", "ptb_type ", "ptb_rate ", "accuracy"])
+        writer.writerow(row)
+except Exception as e:
+    print(f"[Error] Failed to write CSV: {e}")
+

examples/graph/cgscore_experiments/defense_method/GCN.py

@@ -5,19 +5,22 @@ from deeprobust.graph.defense import GCNJaccard, GCN
 from deeprobust.graph.utils import *
 from deeprobust.graph.data import Dataset, PrePtbDataset
 import argparse
+import os
+import csv
 
 parser = argparse.ArgumentParser()
 parser.add_argument('--seed', type=int, default=15, help='Random seed.')
-parser.add_argument('--dataset', type=str, default='cora', choices=['cora', 'cora_ml', 'citeseer', 'polblogs', 'pubmed'], help='dataset')
-parser.add_argument('--ptb_rate', type=float, default=0.0,  help='pertubation rate')
-parser.add_argument('--ptb_type', type=str, default='clean', choices=['clean', 'meta', 'dice', 'minmax', 'pgd', 'random'], help='attack type')
+parser.add_argument('--dataset', type=str, default='Flickr', choices=['cora', 'cora_ml', 'citeseer', 'polblogs', 'pubmed', 'Flickr'], help='dataset')
+parser.add_argument('--ptb_rate', type=float, default=0.25,  help='pertubation rate')
+parser.add_argument('--ptb_type', type=str, default='minmax', choices=['clean', 'meta', 'dice', 'minmax', 'pgd', 'random'], help='attack type')
 parser.add_argument('--hidden', type=int, default=16, help='Number of hidden units.')
 parser.add_argument('--dropout', type=float, default=0.5, help='Dropout rate (1 - keep probability).')
+parser.add_argument('--gpu', type=int, default=0, help='GPU device ID (default: 0)')
 
 args = parser.parse_args()
 args.cuda = torch.cuda.is_available()
 print('cuda: %s' % args.cuda)
-device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+device = torch.device(f"cuda:{args.gpu}" if torch.cuda.is_available() else "cpu")
 
 # make sure you use the same data splits as you generated attacks
 np.random.seed(args.seed)
@@ -63,7 +66,22 @@ def test(adj):
 
 def main():
     
-    test(perturbed_adj)
+    acc = test(perturbed_adj)
+    csv_dir = "../result"
+    os.makedirs(csv_dir, exist_ok=True) 
+
+    csv_filename = os.path.join(csv_dir, f"GCN_{args.dataset}_{args.ptb_type}_{args.ptb_rate}.csv")
+    row = [f"{args.dataset} ", f" {args.ptb_type} ", f" {args.ptb_rate} ", f" {acc}"]
+
+    try:
+        file_exists = os.path.isfile(csv_filename)
+        with open(csv_filename, 'a', newline='') as csvfile:
+            writer = csv.writer(csvfile)
+            if not file_exists:
+                writer.writerow(["dataset ", "ptb_type ", "ptb_rate ", "accuracy"])
+            writer.writerow(row)
+    except Exception as e:
+        print(f"[Error] Failed to write CSV: {e}")
 
     # # if you want to save the modified adj/features, uncomment the code below
     # model.save_adj(root='./', name=f'mod_adj')

examples/graph/cgscore_experiments/defense_method/GCNJaccard.py

@@ -4,21 +4,24 @@ import torch.nn.functional as F
 from deeprobust.graph.defense import GCNJaccard, GCN
 from deeprobust.graph.utils import *
 from deeprobust.graph.data import Dataset, PrePtbDataset
+import csv
 import argparse
+import os
 
 parser = argparse.ArgumentParser()
 parser.add_argument('--seed', type=int, default=15, help='Random seed.')
-parser.add_argument('--dataset', type=str, default='cora', choices=['cora', 'cora_ml', 'citeseer', 'polblogs', 'pubmed'], help='dataset')
-parser.add_argument('--ptb_rate', type=float, default=0.0,  help='pertubation rate')
-parser.add_argument('--ptb_type', type=str, default='clean', choices=['clean', 'meta', 'dice', 'minmax', 'pgd', 'random'], help='attack type')
+parser.add_argument('--dataset', type=str, default='Flickr', choices=['cora', 'cora_ml', 'citeseer', 'polblogs', 'pubmed', 'Flickr'], help='dataset')
+parser.add_argument('--ptb_rate', type=float, default=0.25,  help='pertubation rate')
+parser.add_argument('--ptb_type', type=str, default='dice', choices=['clean', 'meta', 'dice', 'minmax', 'pgd', 'random'], help='attack type')
 parser.add_argument('--hidden', type=int, default=16, help='Number of hidden units.')
 parser.add_argument('--dropout', type=float, default=0.5, help='Dropout rate (1 - keep probability).')
 parser.add_argument('--threshold', type=float, default=0.1, help='jaccard coeficient')
+parser.add_argument('--gpu', type=int, default=0, help='GPU device ID (default: 0)')
 
 args = parser.parse_args()
 args.cuda = torch.cuda.is_available()
 print('cuda: %s' % args.cuda)
-device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+device = torch.device(f"cuda:{args.gpu}" if torch.cuda.is_available() else "cpu")
 
 # make sure you use the same data splits as you generated attacks
 np.random.seed(args.seed)
@@ -54,15 +57,32 @@ def test_jaccard(adj):
     gcn = gcn.to(device)
     gcn.fit(features, adj, labels, idx_train, idx_val, threshold=args.threshold)
     gcn.eval()
-    gcn.test(idx_test)
+    test_acc = gcn.test(idx_test)
+    return  test_acc
 
 
 def main():
     
-    test_jaccard(perturbed_adj)
-    # # if you want to save the modified adj/features, uncomment the code below
-    # model.save_adj(root='./', name=f'mod_adj')
-    # model.save_features(root='./', name='mod_features')
+    acc = test_jaccard(perturbed_adj)
+    
+    print(acc)
+    # record result
+    csv_dir = "../result"
+    os.makedirs(csv_dir, exist_ok=True)  # 自动创建目录（如果不存在）
+
+    csv_filename = os.path.join(csv_dir, f"Jaccard_{args.dataset}_{args.ptb_type}_{args.ptb_rate}.csv")
+    row = [f"{args.dataset} ", f" {args.ptb_type} ", f" {args.ptb_rate} ", f" {args.threshold} ", f" {acc}"]
+
+    # 写入 CSV 文件
+    try:
+        file_exists = os.path.isfile(csv_filename)
+        with open(csv_filename, 'a', newline='') as csvfile:
+            writer = csv.writer(csvfile)
+            if not file_exists:
+                writer.writerow(["dataset ", "ptb_type ", "ptb_rate ", "threshold ", "accuracy"])
+            writer.writerow(row)
+    except Exception as e:
+        print(f"[Error] Failed to write CSV: {e}")
 
 if __name__ == '__main__':
     main()

examples/graph/cgscore_experiments/defense_method/GCNSVD.py

@@ -5,20 +5,25 @@ from deeprobust.graph.defense import GCNJaccard, GCN, GCNSVD
 from deeprobust.graph.utils import *
 from deeprobust.graph.data import Dataset, PrePtbDataset
 import argparse
+import os
+import csv
 
 parser = argparse.ArgumentParser()
-parser.add_argument('--seed', type=int, default=15, help='Random seed.')
-parser.add_argument('--dataset', type=str, default='cora', choices=['cora', 'cora_ml', 'citeseer', 'polblogs', 'pubmed'], help='dataset')
-parser.add_argument('--ptb_rate', type=float, default=0.0,  help='pertubation rate')
-parser.add_argument('--ptb_type', type=str, default='clean', choices=['clean', 'meta', 'dice', 'minmax', 'pgd', 'random'], help='attack type')
+parser.add_argument('--seed', type=int, default=50, help='Random seed.')
+parser.add_argument('--dataset', type=str, default='pubmed', choices=['cora', 'cora_ml', 'citeseer', 'polblogs', 'pubmed', 'Flickr'], help='dataset')
+parser.add_argument('--ptb_rate', type=float, default=0.25,  help='pertubation rate')
+parser.add_argument('--ptb_type', type=str, default='minmax', choices=['clean', 'meta', 'dice', 'minmax', 'pgd', 'random'], help='attack type')
 parser.add_argument('--hidden', type=int, default=16, help='Number of hidden units.')
 parser.add_argument('--dropout', type=float, default=0.5, help='Dropout rate (1 - keep probability).')
 parser.add_argument('--k', type=int, default=50, help='Truncated Components.')
+parser.add_argument('--gpu', type=int, default=5, help='GPU device ID (default: 0)')
 
 args = parser.parse_args()
 args.cuda = torch.cuda.is_available()
 print('cuda: %s' % args.cuda)
-device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+device = torch.device(f"cuda:{args.gpu}" if torch.cuda.is_available() else "cpu")
+
+os.environ["CUDA_VISIBLE_DEVICES"] = str(args.gpu) 
 
 # make sure you use the same data splits as you generated attacks
 np.random.seed(args.seed)
@@ -49,12 +54,32 @@ def test_svd(adj):
     gcn = gcn.to(device)
     gcn.fit(features, perturbed_adj, labels, idx_train, idx_val, k=args.k, verbose=True)
     gcn.eval()
-    gcn.test(idx_test)
+    test_acc = gcn.test(idx_test)
+    return test_acc
 
 
 def main():
     
-    test_svd(perturbed_adj)
+    acc = test_svd(perturbed_adj)
+    
+    print(acc)
+    # record result
+    csv_dir = "../result"
+    os.makedirs(csv_dir, exist_ok=True) 
+
+    csv_filename = os.path.join(csv_dir, f"SVD_{args.dataset}_{args.ptb_type}_{args.ptb_rate}.csv")
+    row = [f"{args.dataset} ", f" {args.ptb_type} ", f" {args.ptb_rate} ", f" {args.k} ", f" {acc}"]
+
+    # 写入 CSV 文件
+    try:
+        file_exists = os.path.isfile(csv_filename)
+        with open(csv_filename, 'a', newline='') as csvfile:
+            writer = csv.writer(csvfile)
+            if not file_exists:
+                writer.writerow(["dataset ", "ptb_type ", "ptb_rate ", "svd_k " "accuracy"])
+            writer.writerow(row)
+    except Exception as e:
+        print(f"[Error] Failed to write CSV: {e}")
     # # if you want to save the modified adj/features, uncomment the code below
     # model.save_adj(root='./', name=f'mod_adj')
     # model.save_features(root='./', name='mod_features')

examples/graph/cgscore_experiments/defense_method/GNNGuard.py

@@ -5,6 +5,8 @@ from deeprobust.graph.defense.gcn_guard import GCNGuard
 from deeprobust.graph.utils import *
 from deeprobust.graph.data import Dataset
 from deeprobust.graph.data import PtbDataset, PrePtbDataset
+import os
+import csv
 
 import argparse
 from scipy import sparse
@@ -12,14 +14,15 @@ from scipy import sparse
 parser = argparse.ArgumentParser()
 parser.add_argument('--seed', type=int, default=15, help='Random seed')
 parser.add_argument('--GNNGuard', type=bool, default=True,  choices=[True, False])
-parser.add_argument('--dataset', type=str, default='cora', choices=['cora', 'cora_ml', 'citeseer', 'polblogs', 'pubmed', 'flickr'], help='dataset')
-parser.add_argument('--ptb_rate', type=float, default=0.0,  help='pertubation rate')
-parser.add_argument('--ptb_type', type=str, default='clean', choices=['clean', 'meta', 'dice', 'minmax', 'pgd', 'random'], help='attack type')
+parser.add_argument('--dataset', type=str, default='Flickr', choices=['cora', 'cora_ml', 'citeseer', 'polblogs', 'pubmed', 'Flickr'], help='dataset')
+parser.add_argument('--ptb_rate', type=float, default=0.25,  help='pertubation rate')
+parser.add_argument('--ptb_type', type=str, default='minmax', choices=['clean', 'meta', 'dice', 'minmax', 'pgd', 'random'], help='attack type')
+parser.add_argument('--gpu', type=int, default=1, help='GPU device ID (default: 0)')
 
 args = parser.parse_args()
 args.cuda = torch.cuda.is_available()
 print('cuda: %s' % args.cuda)
-device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+device = torch.device(f"cuda:{args.gpu}" if torch.cuda.is_available() else "cpu")
 
 np.random.seed(args.seed)
 torch.manual_seed(args.seed)
@@ -39,7 +42,7 @@ perturbed_adj = sp.csr_matrix(perturbed_adj.to('cpu').numpy())
 def test(adj):
     # """defense models"""
     ''' testing model '''
-    gcn =  GCNGuard(nfeat=features.shape[1], nclass=labels.max().item() + 1, nhid=16,
+    gcn =  GCNGuard(nfeat=features.shape[1], nclass=labels.max().item() + 1, nhid=16, drop=True, 
                dropout=0.5, with_relu=False, with_bias=True, weight_decay=5e-4, device=device)
     gcn = gcn.to(device)
 
@@ -57,7 +60,23 @@ def main():
     # test(adj)
     #
     print('=== testing GCN on Mettacked graph ===')
-    test(perturbed_adj)
+    acc = test(perturbed_adj)
+    
+    csv_dir = "../result"
+    os.makedirs(csv_dir, exist_ok=True) 
+
+    csv_filename = os.path.join(csv_dir, f"GNNGuard_{args.dataset}_{args.ptb_type}_{args.ptb_rate}.csv")
+    row = [f"{args.dataset} ", f" {args.ptb_type} ", f" {args.ptb_rate} ", f" {acc}"]
+
+    try:
+        file_exists = os.path.isfile(csv_filename)
+        with open(csv_filename, 'a', newline='') as csvfile:
+            writer = csv.writer(csvfile)
+            if not file_exists:
+                writer.writerow(["dataset ", "ptb_type ", "ptb_rate ", "accuracy"])
+            writer.writerow(row)
+    except Exception as e:
+        print(f"[Error] Failed to write CSV: {e}")
 
 if __name__ == '__main__':
     main()

examples/graph/cgscore_experiments/defense_method/ProGNN.py

@@ -5,12 +5,16 @@ from deeprobust.graph.defense import GCNJaccard, GCN, ProGNN
 from deeprobust.graph.utils import *
 from deeprobust.graph.data import Dataset, PrePtbDataset
 import argparse
+import os
+import csv
 
 parser = argparse.ArgumentParser()
 parser.add_argument('--seed', type=int, default=15, help='Random seed.')
-parser.add_argument('--dataset', type=str, default='cora', choices=['cora', 'cora_ml', 'citeseer', 'polblogs', 'pubmed'], help='dataset')
+
+parser.add_argument('--dataset', type=str, default='pubmed', choices=['cora', 'cora_ml', 'citeseer', 'polblogs', 'pubmed', 'Flickr'], help='dataset')
 parser.add_argument('--ptb_type', type=str, default='clean', choices=['clean', 'meta', 'dice', 'minmax', 'pgd', 'random'], help='attack type')
 parser.add_argument('--ptb_rate', type=float, default=0.0,  help='pertubation rate')
+
 parser.add_argument('--epochs', type=int,  default=200, help='Number of epochs to train.')
 
 parser.add_argument('--hidden', type=int, default=16, help='Number of hidden units.')
@@ -30,9 +34,11 @@ parser.add_argument('--outer_steps', type=int, default=1, help='steps for outer
 parser.add_argument('--lr_adj', type=float, default=0.01, help='lr for training adj')
 parser.add_argument('--symmetric', action='store_true', default=False, help='whether use symmetric matrix')
 
+parser.add_argument('--gpu', type=int, default=6, help='GPU device ID (default: 0)')
+
 args = parser.parse_args()
 args.cuda = torch.cuda.is_available()
-device = torch.device("cuda:5" if args.cuda else "cpu")
+device = torch.device(f"cuda:{args.gpu}" if torch.cuda.is_available() else "cpu")
 print('Using device:', device)
 
 # make sure you use the same data splits as you generated attacks
@@ -70,11 +76,28 @@ def test_prognn(features, adj, labels):
                 dropout=args.dropout, device=device).to(device)
     prognn = ProGNN(model, args, device)
     prognn.fit(features, adj, labels, idx_train, idx_val)
-    prognn.test(features, labels, idx_test)
+    acc = prognn.test(features, labels, idx_test)
+    return acc
+
 
 def main():
+    
+    acc = test_prognn(features, perturbed_adj, labels)
+    csv_dir = "../result"
+    os.makedirs(csv_dir, exist_ok=True) 
+
+    csv_filename = os.path.join(csv_dir, f"ProGNN_{args.dataset}_{args.ptb_type}_{args.ptb_rate}.csv")
+    row = [f"{args.dataset} ", f" {args.ptb_type} ", f" {args.ptb_rate} ", f" {acc}"]
 
-    test_prognn(features, perturbed_adj, labels)
+    try:
+        file_exists = os.path.isfile(csv_filename)
+        with open(csv_filename, 'a', newline='') as csvfile:
+            writer = csv.writer(csvfile)
+            if not file_exists:
+                writer.writerow(["dataset ", "ptb_type ", "ptb_rate ", "accuracy"])
+            writer.writerow(row)
+    except Exception as e:
+        print(f"[Error] Failed to write CSV: {e}")
 
 if __name__ == '__main__':
     main()

examples/graph/cgscore_experiments/defense_method/RGCN.py

@@ -5,20 +5,23 @@ from deeprobust.graph.defense import RGCN
 from deeprobust.graph.utils import *
 from deeprobust.graph.data import Dataset, PrePtbDataset
 import argparse
+import os
+import csv
 
 parser = argparse.ArgumentParser()
 parser.add_argument('--seed', type=int, default=15, help='Random seed.')
-parser.add_argument('--dataset', type=str, default='cora', choices=['cora', 'cora_ml', 'citeseer', 'polblogs', 'pubmed'], help='dataset')
-parser.add_argument('--ptb_rate', type=float, default=0.0,  help='pertubation rate')
-parser.add_argument('--ptb_type', type=str, default='clean', choices=['clean', 'meta', 'dice', 'minmax', 'pgd', 'random'], help='attack type')
+parser.add_argument('--dataset', type=str, default='Flickr', choices=['cora', 'cora_ml', 'citeseer', 'polblogs', 'pubmed', 'Flickr'], help='dataset')
+parser.add_argument('--ptb_rate', type=float, default=0.25,  help='pertubation rate')
+parser.add_argument('--ptb_type', type=str, default='minmax', choices=['clean', 'meta', 'dice', 'minmax', 'pgd', 'random'], help='attack type')
 parser.add_argument('--hidden', type=int, default=16, help='Number of hidden units.')
 parser.add_argument('--dropout', type=float, default=0.5, help='Dropout rate (1 - keep probability).')
+parser.add_argument('--gpu', type=int, default=0, help='GPU device ID (default: 0)')
 
 
 args = parser.parse_args()
 args.cuda = torch.cuda.is_available()
 print('cuda: %s' % args.cuda)
-device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+device = torch.device(f"cuda:{args.gpu}" if torch.cuda.is_available() else "cpu")
 
 # make sure you use the same data splits as you generated attacks
 np.random.seed(args.seed)
@@ -52,12 +55,29 @@ def test_rgcn(adj):
     gcn = gcn.to(device)
     gcn.fit(features, adj, labels, idx_train, idx_val, train_iters=200, verbose=True)
     gcn.eval()
-    gcn.test(idx_test)
+    acc = gcn.test(idx_test)
+    return acc
 
 
 def main():
     
-    test_rgcn(perturbed_adj)
+    acc = test_rgcn(perturbed_adj)
+    
+    csv_dir = "../result"
+    os.makedirs(csv_dir, exist_ok=True) 
+
+    csv_filename = os.path.join(csv_dir, f"RGCN_{args.dataset}_{args.ptb_type}_{args.ptb_rate}.csv")
+    row = [f"{args.dataset} ", f" {args.ptb_type} ", f" {args.ptb_rate} ", f" {acc}"]
+
+    try:
+        file_exists = os.path.isfile(csv_filename)
+        with open(csv_filename, 'a', newline='') as csvfile:
+            writer = csv.writer(csvfile)
+            if not file_exists:
+                writer.writerow(["dataset ", "ptb_type ", "ptb_rate ", "accuracy"])
+            writer.writerow(row)
+    except Exception as e:
+        print(f"[Error] Failed to write CSV: {e}")
     # # if you want to save the modified adj/features, uncomment the code below
     # model.save_adj(root='./', name=f'mod_adj')
     # model.save_features(root='./', name='mod_features')

examples/graph/cgscore_experiments/result/GAT_Flickr_clean_0.0.csv

@@ -0,0 +1,3 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+Flickr , clean , 0.0 , 0.4783828382838284
+Flickr , clean , 0.0 , 0.4641914191419142

examples/graph/cgscore_experiments/result/GAT_Flickr_dice_0.25.csv

@@ -0,0 +1,3 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+Flickr , dice , 0.25 , 0.2811881188118812
+Flickr , dice , 0.25 , 0.27937293729372936

examples/graph/cgscore_experiments/result/GAT_Flickr_minmax_0.25.csv

@@ -0,0 +1,3 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+Flickr , minmax , 0.25 , 0.11204620462046205
+Flickr , minmax , 0.25 , 0.11716171617161716

examples/graph/cgscore_experiments/result/GAT_pubmed_clean_0.0.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+pubmed , clean , 0.0 , 0.8510206669202486

examples/graph/cgscore_experiments/result/GCN_Flickr_clean_0.0.csv

@@ -0,0 +1,3 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+Flickr , clean , 0.0 , 0.5414191419141914
+Flickr , clean , 0.0 , 0.5597359735973597

examples/graph/cgscore_experiments/result/GCN_Flickr_dice_0.25.csv

@@ -0,0 +1,6 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+Flickr , dice , 0.25 , 0.511056105610561
+Flickr , dice , 0.25 , 0.5151815181518152
+Flickr , dice , 0.25 , 0.5115511551155115
+Flickr , dice , 0.25 , 0.5171617161716172
+Flickr , dice , 0.25 , 0.5136963696369637

examples/graph/cgscore_experiments/result/GCN_Flickr_minmax_0.25.csv

@@ -0,0 +1,4 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+Flickr , minmax , 0.25 , 0.116996699669967
+Flickr , minmax , 0.25 , 0.14570957095709572
+Flickr , minmax , 0.25 , 0.11831683168316831

examples/graph/cgscore_experiments/result/GCN_pubmed_minmax_0.25.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+pubmed , minmax , 0.25 , 0.5567389374920756

examples/graph/cgscore_experiments/result/GNNGuard_Flickr_clean_0.0.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+Flickr , clean , 0.0 , 0.7541254125412541

examples/graph/cgscore_experiments/result/GNNGuard_Flickr_dice_0.25.csv

@@ -0,0 +1,3 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+Flickr , dice , 0.25 , 0.7419141914191419
+Flickr , dice , 0.25 , 0.7422442244224422

examples/graph/cgscore_experiments/result/GNNGuard_citeseer_clean_0.0.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+citeseer , clean , 0.0 , 0.710308056872038

examples/graph/cgscore_experiments/result/GNNGuard_citeseer_meta_0.25.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+citeseer , meta , 0.25 , 0.6475118483412323

examples/graph/cgscore_experiments/result/GNNGuard_cora_clean_0.0.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+cora , clean , 0.0 , 0.7801810865191148

examples/graph/cgscore_experiments/result/GNNGuard_cora_dice_0.25.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+cora , dice , 0.25 , 0.7550301810865192

examples/graph/cgscore_experiments/result/GNNGuard_cora_minmax_0.25.csv

@@ -0,0 +1,3 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+cora , minmax , 0.25 , 0.7057344064386318
+cora , minmax , 0.25 , 0.7087525150905433

examples/graph/cgscore_experiments/result/GNNGuard_polblogs_meta_0.25.csv

@@ -0,0 +1,3 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+polblogs , meta , 0.25 , 0.4877300613496933
+polblogs , meta , 0.25 , 0.4877300613496933

examples/graph/cgscore_experiments/result/GNNGuard_polblogs_minmax_0.25.csv

@@ -0,0 +1,3 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+polblogs , minmax , 0.25 , 0.4877300613496933
+polblogs , minmax , 0.25 , 0.4877300613496933

examples/graph/cgscore_experiments/result/GNNGuard_pubmed_dice_0.25.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+pubmed , dice , 0.25 , 0.8228730822873083

examples/graph/cgscore_experiments/result/Jaccard_Flickr_clean_0.0.csv

@@ -0,0 +1,15 @@
+dataset ,ptb_type ,ptb_rate ,threshold ,accuracy
+Flickr , clean , 0.0 , 0.05 , 0.7282178217821782
+Flickr , clean , 0.0 , 0.09 , 0.7422442244224422
+Flickr , clean , 0.0 , 0.2 , 0.7333333333333334
+Flickr , clean , 0.0 , 0.06 , 0.7366336633663366
+Flickr , clean , 0.0 , 0.1 , 0.7404290429042905
+Flickr , clean , 0.0 , 0.4 , 0.717986798679868
+Flickr , clean , 0.0 , 0.03 , 0.7156765676567657
+Flickr , clean , 0.0 , 0.08 , 0.7475247524752475
+Flickr , clean , 0.0 , 0.5 , 0.7358085808580859
+Flickr , clean , 0.0 , 0.04 , 0.7288778877887789
+Flickr , clean , 0.0 , 0.07 , 0.7410891089108911
+Flickr , clean , 0.0 , 0.02 , 0.6950495049504951
+Flickr , clean , 0.0 , 0.3 , 0.7242574257425742
+Flickr , clean , 0.0 , 0.01 , 0.6866336633663367

examples/graph/cgscore_experiments/result/Jaccard_Flickr_dice_0.25.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,threshold ,accuracy
+Flickr , dice , 0.25 , 0.1 , 0.7334983498349835

examples/graph/cgscore_experiments/result/Jaccard_Flickr_minmax_0.25.csv

@@ -0,0 +1,6 @@
+dataset ,ptb_type ,ptb_rate ,threshold ,accuracy
+Flickr , minmax , 0.25 , 0.1 , 0.7336633663366336
+Flickr , minmax , 0.25 , 0.04 , 0.721947194719472
+Flickr , minmax , 0.25 , 0.5 , 0.7338283828382839
+Flickr , minmax , 0.25 , 0.01 , 0.6072607260726073
+Flickr , minmax , 0.25 , 0.06 , 0.7382838283828383

examples/graph/cgscore_experiments/result/Jaccard_citeseer_clean_0.0.csv

@@ -0,0 +1,35 @@
+dataset ,ptb_type ,ptb_rate ,threshold ,accuracy
+citeseer , clean , 0.0 , 0.07 , 0.7132701421800949
+citeseer , clean , 0.0 , 0.03 , 0.729265402843602
+citeseer , clean , 0.0 , 0.06 , 0.7109004739336493
+citeseer , clean , 0.0 , 0.05 , 0.7162322274881517
+citeseer , clean , 0.0 , 0.01 , 0.7209715639810427
+citeseer , clean , 0.0 , 0.08 , 0.7138625592417062
+citeseer , clean , 0.0 , 0.5 , 0.6255924170616114
+citeseer , clean , 0.0 , 0.4 , 0.6558056872037915
+citeseer , clean , 0.0 , 0.3 , 0.6463270142180095
+citeseer , clean , 0.0 , 0.3 , 0.6575829383886257
+citeseer , clean , 0.0 , 0.1 , 0.6990521327014219
+citeseer , clean , 0.0 , 0.03 , 0.7162322274881517
+citeseer , clean , 0.0 , 0.01 , 0.71978672985782
+citeseer , clean , 0.0 , 0.2 , 0.6652843601895735
+citeseer , clean , 0.0 , 0.08 , 0.7008293838862559
+citeseer , clean , 0.0 , 0.04 , 0.7221563981042655
+citeseer , clean , 0.0 , 0.5 , 0.6332938388625593
+citeseer , clean , 0.0 , 0.05 , 0.7097156398104266
+citeseer , clean , 0.0 , 0.06 , 0.7227488151658769
+citeseer , clean , 0.0 , 0.4 , 0.6712085308056872
+citeseer , clean , 0.0 , 0.05 , 0.7162322274881517
+citeseer , clean , 0.0 , 0.09 , 0.6978672985781991
+citeseer , clean , 0.0 , 0.07 , 0.715047393364929
+citeseer , clean , 0.0 , 0.3 , 0.6528436018957346
+citeseer , clean , 0.0 , 0.03 , 0.7257109004739337
+citeseer , clean , 0.0 , 0.02 , 0.7215639810426541
+citeseer , clean , 0.0 , 0.06 , 0.7209715639810427
+citeseer , clean , 0.0 , 0.2 , 0.6753554502369669
+citeseer , clean , 0.0 , 0.1 , 0.6937203791469195
+citeseer , clean , 0.0 , 0.08 , 0.7026066350710901
+citeseer , clean , 0.0 , 0.01 , 0.7298578199052134
+citeseer , clean , 0.0 , 0.4 , 0.6658767772511849
+citeseer , clean , 0.0 , 0.5 , 0.6380331753554502
+citeseer , clean , 0.0 , 0.04 , 0.7209715639810427

examples/graph/cgscore_experiments/result/Jaccard_citeseer_meta_0.25.csv

@@ -0,0 +1,15 @@
+dataset ,ptb_type ,ptb_rate ,threshold ,accuracy
+citeseer , meta , 0.25 , 0.2 , 0.6362559241706162
+citeseer , meta , 0.25 , 0.3 , 0.625
+citeseer , meta , 0.25 , 0.1 , 0.7020142180094787
+citeseer , meta , 0.25 , 0.01 , 0.6107819905213271
+citeseer , meta , 0.25 , 0.5 , 0.658175355450237
+citeseer , meta , 0.25 , 0.06 , 0.6978672985781991
+citeseer , meta , 0.25 , 0.09 , 0.6978672985781991
+citeseer , meta , 0.25 , 0.08 , 0.6907582938388626
+citeseer , meta , 0.25 , 0.4 , 0.6611374407582938
+citeseer , meta , 0.25 , 0.03 , 0.6546208530805687
+citeseer , meta , 0.25 , 0.04 , 0.667654028436019
+citeseer , meta , 0.25 , 0.02 , 0.6433649289099527
+citeseer , meta , 0.25 , 0.05 , 0.6694312796208531
+citeseer , meta , 0.25 , 0.07 , 0.7014218009478673

examples/graph/cgscore_experiments/result/Jaccard_citeseer_minmax_0.25.csv

@@ -0,0 +1,15 @@
+dataset ,ptb_type ,ptb_rate ,threshold ,accuracy
+citeseer , minmax , 0.25 , 0.05 , 0.6877962085308057
+citeseer , minmax , 0.25 , 0.02 , 0.6789099526066351
+citeseer , minmax , 0.25 , 0.08 , 0.7002369668246446
+citeseer , minmax , 0.25 , 0.09 , 0.6949052132701422
+citeseer , minmax , 0.25 , 0.2 , 0.6587677725118484
+citeseer , minmax , 0.25 , 0.04 , 0.6623222748815166
+citeseer , minmax , 0.25 , 0.03 , 0.6931279620853081
+citeseer , minmax , 0.25 , 0.3 , 0.648696682464455
+citeseer , minmax , 0.25 , 0.1 , 0.7120853080568721
+citeseer , minmax , 0.25 , 0.07 , 0.7014218009478673
+citeseer , minmax , 0.25 , 0.01 , 0.691350710900474
+citeseer , minmax , 0.25 , 0.5 , 0.6617298578199052
+citeseer , minmax , 0.25 , 0.06 , 0.6954976303317536
+citeseer , minmax , 0.25 , 0.4 , 0.6718009478672986

examples/graph/cgscore_experiments/result/Jaccard_citeseer_random_0.25.csv

@@ -0,0 +1,15 @@
+dataset ,ptb_type ,ptb_rate ,threshold ,accuracy
+citeseer , random , 0.25 , 0.08 , 0.6907582938388626
+citeseer , random , 0.25 , 0.09 , 0.7120853080568721
+citeseer , random , 0.25 , 0.4 , 0.6516587677725119
+citeseer , random , 0.25 , 0.2 , 0.6777251184834123
+citeseer , random , 0.25 , 0.04 , 0.6978672985781991
+citeseer , random , 0.25 , 0.1 , 0.7008293838862559
+citeseer , random , 0.25 , 0.07 , 0.7138625592417062
+citeseer , random , 0.25 , 0.5 , 0.6433649289099527
+citeseer , random , 0.25 , 0.01 , 0.6836492890995262
+citeseer , random , 0.25 , 0.3 , 0.6516587677725119
+citeseer , random , 0.25 , 0.06 , 0.7209715639810427
+citeseer , random , 0.25 , 0.05 , 0.705568720379147
+citeseer , random , 0.25 , 0.02 , 0.7079383886255924
+citeseer , random , 0.25 , 0.03 , 0.7114928909952607

examples/graph/cgscore_experiments/result/Jaccard_cora_minmax_0.25.csv

@@ -0,0 +1,15 @@
+dataset ,ptb_type ,ptb_rate ,threshold ,accuracy
+cora , minmax , 0.25 , 0.5 , 0.6091549295774649
+cora , minmax , 0.25 , 0.4 , 0.6307847082494971
+cora , minmax , 0.25 , 0.05 , 0.6830985915492959
+cora , minmax , 0.25 , 0.02 , 0.6584507042253521
+cora , minmax , 0.25 , 0.1 , 0.7087525150905433
+cora , minmax , 0.25 , 0.09 , 0.7263581488933603
+cora , minmax , 0.25 , 0.2 , 0.6498993963782697
+cora , minmax , 0.25 , 0.01 , 0.6609657947686117
+cora , minmax , 0.25 , 0.06 , 0.7223340040241449
+cora , minmax , 0.25 , 0.07 , 0.7067404426559357
+cora , minmax , 0.25 , 0.03 , 0.7188128772635816
+cora , minmax , 0.25 , 0.3 , 0.6453722334004025
+cora , minmax , 0.25 , 0.04 , 0.6770623742454729
+cora , minmax , 0.25 , 0.08 , 0.7188128772635816

examples/graph/cgscore_experiments/result/Jaccard_cora_random_0.25.csv

@@ -0,0 +1,15 @@
+dataset ,ptb_type ,ptb_rate ,threshold ,accuracy
+cora , random , 0.25 , 0.02 , 0.7711267605633804
+cora , random , 0.25 , 0.03 , 0.7570422535211269
+cora , random , 0.25 , 0.2 , 0.6448692152917506
+cora , random , 0.25 , 0.08 , 0.7198189134808853
+cora , random , 0.25 , 0.5 , 0.6066398390342053
+cora , random , 0.25 , 0.09 , 0.7268611670020121
+cora , random , 0.25 , 0.1 , 0.6976861167002013
+cora , random , 0.25 , 0.4 , 0.6242454728370221
+cora , random , 0.25 , 0.01 , 0.7711267605633804
+cora , random , 0.25 , 0.05 , 0.7525150905432596
+cora , random , 0.25 , 0.04 , 0.7580482897384306
+cora , random , 0.25 , 0.06 , 0.7535211267605635
+cora , random , 0.25 , 0.3 , 0.6343058350100604
+cora , random , 0.25 , 0.07 , 0.7263581488933603

examples/graph/cgscore_experiments/result/Jaccard_polblogs_clean_0.0.csv

@@ -0,0 +1,15 @@
+dataset ,ptb_type ,ptb_rate ,threshold ,accuracy
+polblogs , clean , 0.0 , 0.2 , 0.5132924335378324
+polblogs , clean , 0.0 , 0.3 , 0.5173824130879346
+polblogs , clean , 0.0 , 0.06 , 0.5071574642126789
+polblogs , clean , 0.0 , 0.07 , 0.5204498977505113
+polblogs , clean , 0.0 , 0.08 , 0.5071574642126789
+polblogs , clean , 0.0 , 0.04 , 0.5030674846625768
+polblogs , clean , 0.0 , 0.09 , 0.5173824130879346
+polblogs , clean , 0.0 , 0.5 , 0.5051124744376279
+polblogs , clean , 0.0 , 0.05 , 0.5204498977505113
+polblogs , clean , 0.0 , 0.03 , 0.5286298568507158
+polblogs , clean , 0.0 , 0.01 , 0.49386503067484666
+polblogs , clean , 0.0 , 0.4 , 0.5265848670756647
+polblogs , clean , 0.0 , 0.02 , 0.5173824130879346
+polblogs , clean , 0.0 , 0.1 , 0.5040899795501023

examples/graph/cgscore_experiments/result/Jaccard_polblogs_dice_0.25.csv

@@ -0,0 +1,15 @@
+dataset ,ptb_type ,ptb_rate ,threshold ,accuracy
+polblogs , dice , 0.25 , 0.09 , 0.5255623721881391
+polblogs , dice , 0.25 , 0.4 , 0.5132924335378324
+polblogs , dice , 0.25 , 0.08 , 0.5204498977505113
+polblogs , dice , 0.25 , 0.05 , 0.4795501022494888
+polblogs , dice , 0.25 , 0.1 , 0.4795501022494888
+polblogs , dice , 0.25 , 0.5 , 0.5204498977505113
+polblogs , dice , 0.25 , 0.07 , 0.5245398773006136
+polblogs , dice , 0.25 , 0.2 , 0.5061349693251534
+polblogs , dice , 0.25 , 0.03 , 0.5102249488752557
+polblogs , dice , 0.25 , 0.06 , 0.5173824130879346
+polblogs , dice , 0.25 , 0.01 , 0.5173824130879346
+polblogs , dice , 0.25 , 0.02 , 0.5081799591002045
+polblogs , dice , 0.25 , 0.3 , 0.5204498977505113
+polblogs , dice , 0.25 , 0.04 , 0.514314928425358

examples/graph/cgscore_experiments/result/Jaccard_polblogs_meta_0.25.csv

@@ -0,0 +1,15 @@
+dataset ,ptb_type ,ptb_rate ,threshold ,accuracy
+polblogs , meta , 0.25 , 0.2 , 0.5112474437627812
+polblogs , meta , 0.25 , 0.1 , 0.5051124744376279
+polblogs , meta , 0.25 , 0.01 , 0.49897750511247446
+polblogs , meta , 0.25 , 0.06 , 0.48261758691206547
+polblogs , meta , 0.25 , 0.5 , 0.5214723926380369
+polblogs , meta , 0.25 , 0.4 , 0.49897750511247446
+polblogs , meta , 0.25 , 0.07 , 0.5204498977505113
+polblogs , meta , 0.25 , 0.09 , 0.47137014314928427
+polblogs , meta , 0.25 , 0.03 , 0.49079754601227
+polblogs , meta , 0.25 , 0.08 , 0.523517382413088
+polblogs , meta , 0.25 , 0.04 , 0.5092024539877301
+polblogs , meta , 0.25 , 0.3 , 0.5204498977505113
+polblogs , meta , 0.25 , 0.05 , 0.5184049079754601
+polblogs , meta , 0.25 , 0.02 , 0.4979550102249489

examples/graph/cgscore_experiments/result/Jaccard_pubmed_minmax_0.25.csv

@@ -0,0 +1,15 @@
+dataset ,ptb_type ,ptb_rate ,threshold ,accuracy
+pubmed , minmax , 0.25 , 0.05 , 0.597629009762901
+pubmed , minmax , 0.25 , 0.03 , 0.5685304932166857
+pubmed , minmax , 0.25 , 0.3 , 0.8451248890579436
+pubmed , minmax , 0.25 , 0.1 , 0.7615062761506277
+pubmed , minmax , 0.25 , 0.01 , 0.5611132242931407
+pubmed , minmax , 0.25 , 0.09 , 0.7393812603017624
+pubmed , minmax , 0.25 , 0.04 , 0.5867249904906809
+pubmed , minmax , 0.25 , 0.2 , 0.8496893622416636
+pubmed , minmax , 0.25 , 0.07 , 0.6496766831494866
+pubmed , minmax , 0.25 , 0.5 , 0.8383415747432484
+pubmed , minmax , 0.25 , 0.4 , 0.8423988842398885
+pubmed , minmax , 0.25 , 0.02 , 0.5549004691264106
+pubmed , minmax , 0.25 , 0.06 , 0.6208951439076963
+pubmed , minmax , 0.25 , 0.08 , 0.685495118549512

examples/graph/cgscore_experiments/result/ProGNN_Flickr_clean_0.0.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+Flickr , clean , 0.0 , 0.44686468646864685

examples/graph/cgscore_experiments/result/ProGNN_Flickr_dice_0.25.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+Flickr , dice , 0.25 , 0.7123762376237623

examples/graph/cgscore_experiments/result/ProGNN_Flickr_minmax_0.25.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+Flickr , minmax , 0.25 , 0.19686468646864685

examples/graph/cgscore_experiments/result/ProGNN_citeseer_dice_0.25.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+citeseer , dice , 0.25 , 0.7215639810426541

examples/graph/cgscore_experiments/result/ProGNN_citeseer_minmax_0.25.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+citeseer , minmax , 0.25 , 0.6990521327014219

examples/graph/cgscore_experiments/result/ProGNN_cora_minmax_0.25.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+cora , minmax , 0.25 , 0.6443661971830986

examples/graph/cgscore_experiments/result/ProGNN_polblogs_dice_0.25.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+polblogs , dice , 0.25 , 0.7474437627811862

examples/graph/cgscore_experiments/result/ProGNN_polblogs_minmax_0.25.csv

@@ -0,0 +1,2 @@
+dataset ,ptb_type ,ptb_rate ,accuracy
+polblogs , minmax , 0.25 , 0.6073619631901841