How Graph Data is Stored in R and Aster
Graph Data in Aster
Teradata Aster is relational data store that uses relational tables for all data structures. By convention, every graph should be represented using two tables:
- the vertices table must contain unique vertex key and optional columns with vertex attributes
- the edges table must contain source and target columns (referencing vertex keys) and optional columns with edge attributes
In turn, every SQL-GR analytical function expects at minimum the following specification:
- the vertex table, view or SQL query with the vertex key
- the edges table, view or SQL query with target and source keys
- if graph is directed or not
- if edges have weight attribute
toaster Graph Metadata Object
To simplify and streamline graph operations in Aster toaster provides specialized object toagraph that holds metadata about Aster graph. The graph metadata consists of the following:
- vertices the name of table or view, or SQL query representing the vertices
- edges the name of table or view, or SQL query representing the edges
- directed logical flag indicating if graph is directed or not
- key the name of a unique column (key) in the vertex table or query
- source the name of a source column in the edge table or query
- target the name of a target column in the edge table or query
- vertexAttrnames the names of vertex attributes (optional)
- edgeAttrnames the names of edge attributes (optional)
- vertexWhere a SQL expression to filter vertices inside WHERE clause
- edgeWhere a SQL expression to filter edges inside WHERE clause
Examples of two graphs defined on Open Dallas Police data and used in all examples that follow:
# undirected graph of police officers
policeGraphUn = toaGraph(vertices = "dallaspolice_officer_vertices",
edges = "dallaspolice_officer_edges_un",
directed = FALSE,
key = "officer", source = "officer1", target = "officer2",
vertexAttrnames = c("offense_count"),
edgeAttrnames = c("weight"))
# directed graph of police officers
policeGraphDi = toaGraph(vertices = "dallaspolice_officer_vertices",
edges = "dallaspolice_officer_edges_di",
directed = TRUE,
key = "officer", source = "officer1", target = "officer2",
vertexAttrnames = c("offense_count"),
edgeAttrnames = c("weight"))Note, that toagraph stores only metadata and requires no ODBC connectivity.
R Network Object and toaster
In toaster we always use de facto R standards to hold in R results of computations in Aster. Examples of such approach were using kmeans, lm, or DocumentTermMatrix in corresponding computation functions. The same applied to graph with the choice between igraph and network objects from the same packages. The choice of network object is not obvious and you can find details here. Moreover, using package intergraph one can always use them interchangeably (more examples later).
Function computeGraph transfers graph from the Aster database into R memory as network object where it can be used to visualize graph using ggplot2, igraph and other graph visualization functions. The simplest invocation of computeGraph may look like this:
netPoliceUn = computeGraph(conn, policeGraphUn)
GGally::ggnet2(netPoliceUn, node.label="vertex.names", node.size="offense_count",
legend.position="none", label.size=3)
To finish let’s visualize the same graph with igraph:
igraphPoliceUn = asIgraph(netPoliceUn)
plot(igraphPoliceUn)
