update comsim & onlysim & communityRewiring

Comm+Sim/Citeseer_1000_2000_comsim.csv

@@ -0,0 +1,14 @@
+Metric,Value
+Dataset,Citeseer
+Number of Communities,34
+Original Edge Count,3679
+Final Edge Count,4317
+Edges Modified,638
+Edges to be Added (Budget),1000
+Edges to be Removed (Budget),2000
+Mean edges by cluster,0.0016603416827790686
+Covariance edges by cluster,0.00034657850797783314
+Mean edges intra-cluster,0.01649730374288271
+Covariance edges intra-cluster,0.0008512322502035165
+Mean edges inter-cluster,0.0007611318609546055
+Covariance edges inter-cluster,0.0003032834856890727

Comm+Sim/Citeseer_2000_2000_comsim.csv

@@ -0,0 +1,14 @@
+Metric,Value
+Dataset,Citeseer
+Number of Communities,38
+Original Edge Count,3679
+Final Edge Count,4716
+Edges Modified,1037
+Edges to be Added (Budget),2000
+Edges to be Removed (Budget),2000
+Mean edges by cluster,0.004361091394654539
+Covariance edges by cluster,0.001054218127622854
+Mean edges intra-cluster,0.0677765981107619
+Covariance edges intra-cluster,0.013538898498386564
+Mean edges inter-cluster,0.0009332261667568449
+Covariance edges inter-cluster,0.00016823893234729564

Comm+Sim/comsim.py

@@ -1,6 +1,6 @@
 import torch
 import networkx as nx
-import nx_cugraph as nxcg
+# import nx_cugraph as nxcg
 import numpy as np
 import pandas as pd
 import os
@@ -16,8 +16,9 @@ def modify_graph(data, dataset_name, budget_add, budget_delete, seed):
     print("=============================================================")
     print("Rewiring based on feature similarity...")
     G = to_networkx(data, to_undirected=True)
-    nxcg_G = nxcg.from_networkx(G) 
-    communities = list(nx.community.louvain_communities(nxcg_G, seed=seed))
+    # nxcg_G = nxcg.from_networkx(G) 
+    # communities = list(nx.community.louvain_communities(nxcg_G, seed=seed))
+    communities = list(nx.community.louvain_communities(G, seed=seed))
     cluster_dict_before = {node: i for i, cluster in enumerate(communities) for node in cluster}
     cluster_list_before = [cluster_dict_before[node] for node in range(len(data.y))]
     nmiscoremod_before = NMI(cluster_list_before, data.y.cpu().numpy())     

Comm+Sim/main.py

@@ -20,7 +20,7 @@
 from matplotlib import pyplot as plt
 #from comsim_onlysim2 import *
 from comsim  import *
-from onlysim import *
+# from onlysim import *
 args = parse_args()
 
 
@@ -107,15 +107,19 @@
 
 ## For Community+SimilarityBased Rewiring ####
 algo_stime = time.time()
+# ComFy
 data,nmiscoremod_before = modify_graph(data,args.dataset,budget_edges_add,budget_edges_delete,seed)
+# FeaSt
+# data, nmiscoremod_before, nmiscoremod_after = modify_graph(data, args.dataset, budget_edges_add, budget_edges_delete, seed)
 algo_etime = time.time()
 rewire_time =  algo_etime - algo_stime
 print(f"Time Taken for Rewiring : {rewire_time}")
 
 
 newG = to_networkx(data, to_undirected=True)
-nxcg_G = nxcg.from_networkx(newG) 
-communities_after = list(nx.community.louvain_communities(nxcg_G, seed=seed))
+# nxcg_G = nxcg.from_networkx(newG) 
+# communities_after = list(nx.community.louvain_communities(nxcg_G, seed=seed))
+communities_after = list(nx.community.louvain_communities(newG, seed=seed))
 cluster_dict_after = {node: i for i, cluster in enumerate(communities_after) for node in cluster}
 cluster_list_after = [cluster_dict_after[node] for node in range(len(data.y))]
 nmiscoremod_after = NMI(cluster_list_after, data.y.cpu().numpy())

Comm+Sim/model.py

@@ -76,7 +76,8 @@ def reset_parameters(self):
         for bn in self.bns:
             bn.reset_parameters()
 
-    def forward(self, x, edge_index):
+    # def forward(self, x, edge_index):
+    def forward(self, x, edge_index, p=0.0):
         for i, conv in enumerate(self.convs[:-1]):
             x = conv(x, edge_index)
             if self.use_bn:

Comm+Sim/onlysim.py

@@ -1,5 +1,5 @@
 import networkx as nx
-import nx_cugraph as nxcg
+# import nx_cugraph as nxcg
 import numpy as np
 import pandas as pd
 import os
@@ -17,8 +17,9 @@ def modify_graph(data, dataset_name, budget_edges_add, budget_edges_delete, seed
     print("=============================================================")
     print("Rewiring based on feature similarity...")
     G = to_networkx(data, to_undirected=True)
-    nxcg_G = nxcg.from_networkx(G) 
-    communities = list(nx.community.louvain_communities(nxcg_G, seed=seed))
+    # nxcg_G = nxcg.from_networkx(G) 
+    # communities = list(nx.community.louvain_communities(nxcg_G, seed=seed))
+    communities = list(nx.community.louvain_communities(G, seed=seed))
     cluster_dict_before = {node: i for i, cluster in enumerate(communities) for node in cluster}
     cluster_list_before = [cluster_dict_before[node] for node in range(len(data.y))]
     nmiscoremod_before = NMI(cluster_list_before, data.y.cpu().numpy())     
@@ -94,8 +95,9 @@ def modify_graph(data, dataset_name, budget_edges_add, budget_edges_delete, seed
     print("similarity before and after",sim,sim2)
     pyg_data = from_networkx(G)
     newG = to_networkx(pyg_data, to_undirected=True)
-    nxcg_G = nxcg.from_networkx(newG) 
-    communities_after = list(nx.community.louvain_communities(nxcg_G, seed=seed))
+    # nxcg_G = nxcg.from_networkx(newG) 
+    # communities_after = list(nx.community.louvain_communities(nxcg_G, seed=seed))
+    communities_after = list(nx.community.louvain_communities(newG, seed=seed))
     cluster_dict_after = {node: i for i, cluster in enumerate(communities_after) for node in cluster}
     cluster_list_after = [cluster_dict_after[node] for node in range(len(data.y))]
     nmiscoremod_after = NMI(cluster_list_after, data.y.cpu().numpy())

Comm+Sim/results_baseline.csv

@@ -0,0 +1,2 @@
+Dataset,AvgValAcc,DeviationVal,AvgTestAcc,Deviation,ELI,AdjHom,NMIBefore,NMIAfter,EdgesAdded,EdgesDeleted,HiddenDim,LR,Dropout,GCNTime,RewireTime
+Citeseer,77.94,0.35,77.62,0.33,0.4437396170928487,0.6620897112374549,0.3257939098082461,0.3257939098082461,0,0,32,0.01,0.31,111.7904372215271,1.0000636577606201

Comm+Sim/results_citeseer_final.csv

@@ -0,0 +1,3 @@
+Dataset,AvgValAcc,DeviationVal,AvgTestAcc,Deviation,ELI,AdjHom,NMIBefore,NMIAfter,EdgesAdded,EdgesDeleted,HiddenDim,LR,Dropout,GCNTime,RewireTime
+Citeseer,76.93,0.39,76.62,0.38,0.29706154481533353,0.5336210227744842,0.3257939098082461,0.35253033523293986,1000,2000,32,0.01,0.31,120.44084429740906,2.041614294052124
+Citeseer,72.09,0.43,72.23,0.40,0.29706154481533353,0.5336210227744842,0.3257939098082461,0.35253033523293986,1000,2000,32,0.01,0.31,113.30892968177795,2.047795295715332

Comm+Sim/results_comfy.csv

@@ -0,0 +1,2 @@
+Dataset,AvgValAcc,DeviationVal,AvgTestAcc,Deviation,ELI,AdjHom,NMIBefore,NMIAfter,EdgesAdded,EdgesDeleted,HiddenDim,LR,Dropout,GCNTime,RewireTime
+Citeseer,76.29,0.37,76.16,0.35,0.38227746056556877,0.6060492452747333,0.32000240378789907,0.3148810758800009,2000,2000,32,0.01,0.31,117.36940407752991,2.1097404956817627

Comm+Sim/results_feast.csv

@@ -0,0 +1,2 @@
+Dataset,AvgValAcc,DeviationVal,AvgTestAcc,Deviation,ELI,AdjHom,NMIBefore,NMIAfter,EdgesAdded,EdgesDeleted,HiddenDim,LR,Dropout,GCNTime,RewireTime
+Citeseer,76.67,0.34,76.48,0.35,0.4820896180314216,0.6932820241627689,0.32000240378789907,0.3400549267819509,2000,2000,32,0.01,0.31,113.85506987571716,1.1044039726257324

Comm+Sim/simrewirings/similarityreports/Citeseer_0_0_onlysim.csv

@@ -0,0 +1,13 @@
+Metric,Value
+Dataset,Citeseer
+Number of Communities,34
+Edges to be Added,0
+Edges to be Removed,0
+Mean edges by cluster,0.0
+Covariance edges by cluster,0.0
+Mean edges intra-cluster,0.0
+Covariance edges intra-cluster,0.0
+Mean edges inter-cluster,0.0
+Covariance edges inter-cluster,0.0
+NMI before,0.3257939098082461
+NMI after,0.3257939098082461

Comm+Sim/simrewirings/similarityreports/Citeseer_2000_2000_onlysim.csv

@@ -0,0 +1,13 @@
+Metric,Value
+Dataset,Citeseer
+Number of Communities,38
+Edges to be Added,2000
+Edges to be Removed,2000
+Mean edges by cluster,0.0021284272871995575
+Covariance edges by cluster,0.00010463365938457873
+Mean edges intra-cluster,0.03465807046907419
+Covariance edges intra-cluster,0.0009286828389534813
+Mean edges inter-cluster,0.0003700681962874153
+Covariance edges inter-cluster,9.734140853178402e-07
+NMI before,0.32000240378789907
+NMI after,0.3400549267819509

CommunityRewiring/clustering.py

@@ -0,0 +1,26 @@
+import networkx as nx
+import numpy as np
+import scipy.sparse.linalg
+from sklearn.cluster import KMeans
+import sklearn.preprocessing
+
+
+def k_way_spectral(G, k):
+        assert nx.is_connected(G), "the graph must be connnected"
+        clusters = []
+        if G.order() < k:
+                clusters = list(G.nodes())
+        else:
+            L = nx.laplacian_matrix(G)
+            _, eigenvecs = scipy.sparse.linalg.eigsh(L.asfptype(), k=k+1, which='SM')
+            eigenvecs = eigenvecs[:, 1:]
+            eigenvecs = sklearn.preprocessing.normalize(eigenvecs)
+            kmeans = KMeans(n_clusters=k).fit(eigenvecs)
+            cluster_labels = kmeans.labels_
+            clusters = [[] for _ in range(max(cluster_labels) + 1)]
+            for node_id, cluster_id in zip(G.nodes(), cluster_labels):
+                            clusters[cluster_id].append(node_id)
+        return clusters,cluster_labels
+
+def maximize_modularity(G):
+  return nx.community.greedy_modularity_communities(G)

CommunityRewiring/main.py

@@ -24,7 +24,7 @@
 
 from sklearn.metrics.cluster import normalized_mutual_info_score as NMI
 
-import nx_cugraph as nxcg
+# import nx_cugraph as nxcg
 
 ######### Hyperparams to use #############
 #Cora --> Dropout = 0.4130296 ; LR = 0.01 ; Hidden_Dimension = 32
@@ -254,8 +254,9 @@
 
 
 newG = to_networkx(data, to_undirected=True)
-nxcg_G = nxcg.from_networkx(newG) 
-communities_after = list(nx.community.louvain_communities(nxcg_G, seed=seed))
+# nxcg_G = nxcg.from_networkx(newG) 
+# communities_after = list(nx.community.louvain_communities(nxcg_G, seed=seed))
+communities_after = list(nx.community.louvain_communities(newG, seed=seed))
 cluster_dict_after = {node: i for i, cluster in enumerate(communities_after) for node in cluster}
 cluster_list_after = [cluster_dict_after[node] for node in range(len(data.y))]
 nmiscoremod_after = NMI(cluster_list_after, data.y.cpu().numpy())

CommunityRewiring/methods.py

@@ -12,7 +12,7 @@
 from torch_geometric.utils import to_networkx,from_networkx,homophily
 import random
 from clustering import *
-import nx_cugraph as nxcg
+# import nx_cugraph as nxcg
 from sklearn.metrics.cluster import normalized_mutual_info_score as NMI
 
 def proxydelmin(data, nxgraph,seed, max_iterations):
@@ -672,8 +672,9 @@ def community_rewiring(data, seed, comm_delete, comm_add):
     initial_edges = graph.number_of_edges()
 
     # Perform community detection
-    nxcg_G = nxcg.from_networkx(graph) 
-    communities = list(nx.community.louvain_communities(nxcg_G, seed=seed))
+    # nxcg_G = nxcg.from_networkx(graph) 
+    # communities = list(nx.community.louvain_communities(nxcg_G, seed=seed))
+    communities = list(nx.community.louvain_communities(graph, seed=seed))
 
     cluster_dict = {node: i for i, cluster in enumerate(communities) for node in cluster}
 
@@ -749,8 +750,9 @@ def inverse_community_rewiring(data, seed, intra_delete, inter_add):
     initial_edges = graph.number_of_edges()
 
     # Perform community detection
-    nxcg_G = nxcg.from_networkx(graph) 
-    communities = list(nx.community.louvain_communities(nxcg_G, seed=seed))
+    # nxcg_G = nxcg.from_networkx(graph) 
+    # communities = list(nx.community.louvain_communities(nxcg_G, seed=seed))
+    communities = list(nx.community.louvain_communities(graph, seed=seed))
 
     cluster_dict = {node: i for i, cluster in enumerate(communities) for node in cluster}
 

CommunityRewiring/results_comma.csv

@@ -0,0 +1,2 @@
+Method,Dataset,AvgValAcc,DeviationVal,AvgTestAcc,Deviation,ELI,AdjHom,NMI,EdgesAdded,EdgesDeleted,FinalGap,HiddenDim,LR,Dropout,RewiringTime,GCNTime
+community_rewiring,Citeseer,73.35,0.38,73.63,0.41,0.33304799654311035,0.5685390036963097,0.32185336244421514,2000,40,1.8829173342757362e-07,32,0.01,0.31,0.5870213508605957,118.51752805709839