Scale-free Graphs

• Description: This dataset is generated as a directed scalefree network, which is a network whose degree distribution follows power-law property (Bollobas´ et al., 2003). It fits the "one-to-many" mapping graph translation problem. There are no node features in this dataset, and the goal is to learn the mapping from the input graphs' topology to the target graph's topology. To generate a target graph, a node will by selected as target node with probability proportional to its in-degree, which will be linked to a new source node with probability of 0.41. Similarly, a node will by selected as source node with probability proportional to its out-degree, which will be linked to a new target node with probability of 0.54. Then, a corresponding target graph is generated by adding m (m equals the number of nodes of the input graph) edges between two nodes. Thus, both input and target graphs are directed scale-free graphs.
• Statistics:

Name	Type	#Graphs	#Nodes	#Edges	Attributed	Directed	Weighted	Signed	Homogeneous	Spatial	Temporal	Labels
Scale-free Graphs	Synthetic Networks	10,000	10/20/50/100/150	~20/~40/~100/~200/~320	NO	YES	NO	NO	YES	NO	NO	NO

• Acknowlegement:
  Guo, X., Wu, L., Zhao, L. (2018). Deep graph translation. arXiv preprint arXiv:1805.09980.

Erdos-Renyi Graphs

• Description: This dataset is generated by the Erdos Renyi model with the edge probability of 0.2. It fits the problem of "one-to-one" mapping problem of graph translation. It contains pairs of (input, target) graphs. The target graph topology is the 2-hop connection of the input graph, where each edge in the target graph refers to the 2-hop reachability in the input graph (e.g. if node i is 2-hop reachable to node j in the input graph, then they are connected in the target graph). There are edge and node attributes for graphs in this dataset: the edge attributes E_(i,j) denotes the existence of the edge and the node attributes are continuous values computed following the polynomial function: f(x) : y = ax^2 + bx + c(a =0; b =1; c= 5), where x is the node degree and f(x) is the node attribute.
• Statistics:

Name	Type	#Graphs	#Nodes	#Edges	Attributed	Directed	Weighted	Signed	Homogeneous	Spatial	Temporal	Labels
Erdos-Renyi Graphs	Synthetic Networks	1,000	20/40/60	~100/~200/~400	YES	NO	NO	NO	YES	NO	NO	NO

• Acknowlegement:
  Guo X, Zhao L, Nowzari C, Rafatirad S, Homayoun H, Dinakarrao SM. Deep Multi-attributed Graph Translation with Node-Edge Co-evolution. Inhe 19th International Conference on Data Mining (ICDM 2019).

Barab´asi-Albert Graphs

• Description: This dataset is generated by the Barab´asi-Albert model. It fits the problem of "one-to-one" mapping problem of graph translation. It contains pairs of (input, target) graphs. The target graph topology is the 2-hop connection of the input graph, where each edge in the target graph refers to the 3-hop reachability in the input graph (e.g. if node i is 3-hop reachable to node j in the input graph, then they are connected in the target graph). There are edge and node attributes for graphs in this dataset: the edge attributes E_(i,j) denotes the existence of the edge and the node attributes are continuous values computed following the polynomial function: f(x) : y = ax^2 + bx + c(a =0; b =1; c= 5), where x is the node degree and f(x) is the node attribute. Here we provide the datasets with three different node sizes.
• Statistics:

Name	Type	#Graphs	#Nodes	#Edges	Attributed	Directed	Weighted	Signed	Homogeneous	Spatial	Temporal	Labels
Barab´asi-Albert Graphs	Synthetic Networks	1,000	20/40/60	~60/~190/~300	YES	NO	NO	NO	YES	NO	NO	NO

• Acknowlegement:
  Guo X, Zhao L, Nowzari C, Rafatirad S, Homayoun H, Dinakarrao SM. Deep Multi-attributed Graph Translation with Node-Edge Co-evolution. Inhe 19th International Conference on Data Mining (ICDM 2019).

Waxman Graphs

• Description: The Waxman random graph model places n nodes uniformly at random in a rectangular domain. Each pair of nodes at distance d is joined by an edge with the probability of βe−d/αL, where L is the maximum distance between any pair of nodes, α and β are model parameters, β is set to 0.4 and α is set to 0.1. The coordinates of the four vertexes of rectangular domain for generating the spatial location of nodes are (p,p),(p,n×s+p),(n×s+p,p),(n×s+p, n×s+p), respectively. There are 8,000 samples for training and 1,600 for testing.
• Statistics:

Name	Type	#Graphs	#Nodes	#Edges	Attributed	Directed	Weighted	Signed	Homogeneous	Spatial	Temporal	Labels
Waxman Graphs	Synthetic Networks	9,600	25	~250	YES	NO	NO	NO	YES	YES	2D	YES

• Acknowlegement:
  Guo X, Du Y, Zhao L. Deep Generative Model for Spatial Networks. 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2021), to appear, virtual, Aug 2021.

Random Geometric Graphs

• Description: The random geometric graph model places $n$ nodes uniformly at random in a rectangular domain. Two nodes are joined by an edge if their distance is larger than a threshold β = 12. The coordinates of the four vertexes of rectangular domain is defined as the same as that in Waxman graph. The node attributes among a graph are generated in the same rule as that for generating Waxman graphs. There are 8,000 samples for training and 1,600 for testing in this dataset.
• Statistics:

Name	Type	#Graphs	#Nodes	#Edges	Attributed	Directed	Weighted	Signed	Homogeneous	Spatial	Temporal	Labels
Random Geometric Graphs	Synthetic Networks	9,600	25	~350	YES	NO	NO	NO	YES	YES	2D	YES

• Acknowlegement:
  Guo X, Du Y, Zhao L. Deep Generative Model for Spatial Networks. 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2021), to appear, virtual, Aug 2021.