通过dgl.graph()创建同构图 通过dgl.heterograph()创建异构图 借助dgl.*工具包和其他数据源生成图，如dgl.from_networkx等
1、两个节点数创建图
import dglimport torch#基于 tensor 创建src = torch.tensor([0, 1, 2, 3])dst = torch.tensor([1, 2, 3, 4])g = dgl.graph((src, dst))#基于 numpy 创建src = np.array([0, 1, 2, 3])dst = np.array([1, 2, 3, 4])g = dgl.graph((src, dst))#基于 ；list 创建src = [0, 1, 2, 3]dst = [1, 2, 3, 4]g = dgl.graph((src, dst))g
2、从networkx创建图
import dglimport networkx as nx# 创建一个 networkx 图nx_g = nx.karate_club_graph()# 转换为 DGL 图g = dgl.from_networkx(nx_g)import networkx as nxnx_g = nx.path_graph(5) # 一条链路0-1-2-3-4dgl.from_networkx(nx_g) # 来自NetworkXGraph(num_nodes=5, num_edges=8,ndata_schemes={}edata_schemes={})
nxg = nx.DiGraph([(2, 1), (1, 2), (2, 3), (0, 0)])dgl.from_networkx(nxg)=4, num_edges=4,ndata_schemes={}edata_schemes={})
3、使用稀疏矩阵创建图
import dglimport scipy.sparse as spimport torchadj = sp.coo_matrix([[],[],[],[],[]])g = dgl.graph((torch.tensor(adj.row), torch.tensor(adj.col)))spmat = sp.rand(100, 100, density=0.05)dgl.from_scipy(spmat)
4、从DataFrame创建图
import dglimport pandas as pd# 创建一个 dataframedf = pd.DataFrame({'source': [0, 1, 2, 3],'target': [1, 2, 3, 4]})# 转换为 DGL 图g = dgl.graph((df['source'].values, df['target'].values))g
5、 随机图生成
# 随机生产图g = dgl.rand_graph(100, 10)#转换为networkxnx_g = dgl.to_networkx(g)#可视化看fig = plt.figure(figsize=(4,3.5),dpi=1000)nx.draw(nx_graph,with_labels=True, width=0.2,node_color='skyblue',node_size=800, font_size=13)plt.show()
import dglg = dgl.rand_bipartite('user', 'buys', 'game', 50, 100, 10)gGraph(num_nodes={'game': 100, 'user': 50},num_edges={('user', 'buys', 'game'): 10},metagraph=[('user', 'game', 'buys')])g.edges()(tensor([38, 19, 16, 37, 10, 35, 7, 26, 3, 27]),tensor([50, 84, 53, 34, 75, 76, 69, 68, 80, 24]))
6、从其他图格式导入
import dglimport networkx as nx# 从 GraphML 文件读取图数据nx_g = nx.read_graphml('path_to_graph_file.graphml')g = dgl.from_networkx(nx_g)
7、边索引创建异构图
import dgldata_dict = {('user', 'follows', 'user'): ([0, 1], [1, 2]),('user', 'writes', 'post'): ([0, 1], [0, 1]),('post', 'mentions', 'user'): ([0, 1], [1, 2])}hetero_g = dgl.heterograph(data_dict)
8、二分图
# uni-directional bipartite举例graph_data1 = {('drug', 'interacts2', 'gene'): (th.tensor([0, 1]), th.tensor([2, 3]))}g1 = dgl.heterograph(graph_data1)# uni-bipartite举例graph_data2 = {('drug', 'interacts2', 'gene'): (th.tensor([0, 1]), th.tensor([2, 3])),('drug', 'treats', 'disease'): (th.tensor([1]), th.tensor([2]))}g2 = dgl.heterograph(graph_data2)
9、总 结
往期精彩：
SynchroTrap-基于松散行为相似度的欺诈账户检测算法
