def louvain(G):
communities = community_louvain.best_partition(G)
for node, label in communities.iteritems():
G.node[node]['lbl'] = label
partition = {}
for node in G.nodes:
lbl = communities[node]
if partition.has_key(lbl):
partition[lbl] += [node]
else:
partition[lbl] = [node]
return partition.values()
return partition
import networkx as nx from community_louvain import best_partition g = nx.Graph() g.add_nodes_from(range(0, 17)) edgestr = ''' [(0, 0, 47), (0, 1, 2), (0, 5, 1), (0, 7, 1), (0, 13, 1), (0, 16, 1), (1, 1, 31), (1, 11, 1), (1, 14, 2), (2, 2, 53), (2, 11, 1), (2, 10, 1), (2, 15, 1), (3, 3, 32), (4, 16, 12), (4, 8, 1), (4, 10, 2), (4, 11, 1), (4, 4, 29), (5, 5, 29), (5, 8, 1), (5, 10, 12), (5, 13, 2), (5, 14, 1), (6, 6, 15), (7, 7, 77), (7, 9, 1), (7, 12, 1), (7, 13, 1), (7, 14, 1), (8, 8, 41), (8, 14, 1), (8, 13, 1), (9, 9, 22), (9, 12, 17), (9, 14, 1), (10, 10, 9), (11, 11, 52), (11, 14, 2), (12, 12, 38), (12, 13, 1), (12, 15, 1), (13, 13, 73), (13, 16, 1), (14, 14, 49), (15, 15, 10), (16, 16, 17)] ''' weightededgelist = eval(edgestr) unweightededgelist = [(e[0], e[1]) for e in weightededgelist] g.add_weighted_edges_from(weightededgelist) #g.add_edges_from(unweightededgelist) print(list(g.edges(data=True))) communities = best_partition(g) print(communities)
def louvain_partition(graph): partition = community_louvain.best_partition(graph) return partition
import community_louvain
import networkx as nx
import matplotlib.pyplot as plt
import pandas as pd
actor_edge = pd.read_csv('2.csv') #导入演员网络边的csv文件,导入后为Dataframe格式
print(actor_edge.head(3)) #显示表的前3行
weight_edge = []
for _, row in actor_edge.iterrows(): #把边及边的权重加入列表，数据格式为（节点，节点，权重）
weight_edge.append((row['A'], row['B'], row['rankaveragebox']))
G = nx.Graph() #初始化无向图
G.add_weighted_edges_from(weight_edge) #把带权重边的信息加入无向图中
#first compute the best partition
partition = community_louvain.best_partition(G)
print("0")
#drawing
size = float(len(set(partition.values())))
print("0")
pos = nx.spring_layout(G)
count = 0.
for com in set(partition.values()):
count = count + 1.
list_nodes = [
nodes for nodes in partition.keys() if partition[nodes] == com
]
nx.draw_networkx_nodes(G,
pos,
list_nodes,
def create_graph(self):
n = 28
G = nx.grid_2d_graph(n, n)
graph_list = []
self.fill_edges(G, n)
mapping = dict()
for node in G.nodes():
mapping[node] = len(mapping)
G = nx.relabel_nodes(G, mapping)
for x in range(10000):
H = nx.Graph()
image, graph_label = mnist_trainset[x]
node_attributes,position = self.process_image(image, n)
#print(node_attributes)
#print(G.number_of_edges())
#self.compute_weights(G,n,node_attributes)
#print(G.edges.data())
#print(G.number_of_edges())
partition = best_partition(G)
print(partition)
no_clusters = max(partition.values()) + 1
print(no_clusters)
new_attributes = self.compute_atrributes(H,no_clusters,partition,node_attributes,position)
#print(new_attributes)
#print(x)
#print(H.number_of_nodes())
#print(H.number_of_edges())
nodes = list(G.nodes())
#print(nodes)
edge_matrix = np.zeros(shape=(no_clusters,no_clusters))
#count_edges = 0
for i in range(len(nodes)):
for j in range(i+1, len(nodes)):
if G.has_edge(nodes[i], nodes[j]):
if(partition[i]!=partition[j]):
H.add_edge(partition[i],partition[j])
edge_matrix[partition[i],partition[j]] = edge_matrix[partition[i],partition[j]] + 1
#count_edges = count_edges + 1
#print(count_edges)
edge_attributes = np.zeros(H.number_of_edges())
counter = 0
for i in range(no_clusters):
for j in range(no_clusters):
if(edge_matrix[i,j] != 0 or edge_matrix[j,i]!=0):
edge_attributes[counter] = edge_matrix[i,j] + edge_matrix[j,i]
edge_matrix[i,j] = 0
edge_matrix[j,i] = 0
counter = counter + 1
edges = np.array(H.edges()).transpose()
graph = Data(
x=torch.tensor(new_attributes),
edge_index=torch.tensor(edges),
edge_attr=torch.tensor(edge_attributes),
y=graph_label)
graph_list.append(graph)
for x in range(1000):
H = nx.Graph()
image, graph_label = mnist_testset[x]
node_attributes,position = self.process_image(image, n)
self.compute_weights(G,n,node_attributes)
partition = best_partition(G)
no_clusters = max(partition.values()) + 1
new_attributes = self.compute_atrributes(H,no_clusters,partition,node_attributes,position)
#print(new_attributes)
nodes = list(G.nodes())
edge_matrix = np.zeros(shape=(no_clusters,no_clusters))
#count_edges = 0
for i in range(len(nodes)):
for j in range(i+1, len(nodes)):
if G.has_edge(nodes[i], nodes[j]):
if(partition[i]!=partition[j]):
H.add_edge(partition[i],partition[j])
edge_matrix[partition[i],partition[j]] = edge_matrix[partition[i],partition[j]] + 1
edge_attributes = np.zeros(H.number_of_edges())
counter = 0
for i in range(no_clusters):
for j in range(no_clusters):
if(edge_matrix[i,j] != 0 or edge_matrix[j,i]!=0):
edge_attributes[counter] = edge_matrix[i,j] + edge_matrix[j,i]
edge_matrix[i,j] = 0
edge_matrix[j,i] = 0
counter = counter + 1
edges = np.array(H.edges()).transpose()
graph = Data(
x=torch.tensor(new_attributes),
edge_index=torch.tensor(edges),
edge_attr=torch.tensor(edge_attributes),
y=graph_label)
graph_list.append(graph)
return graph_list