The LoopyBeliefPropagation function has two required input tables, vertices and edges, and one optional input table, observation.
The following table describes the vertices table columns that you must or can specify in the function call. The table can have additional columns, but the function ignores them.
Column Name | Data Type | Description |
---|---|---|
vertex_key_column | INTEGER | Column that is, or is part of, the vertex key. Every column in the key appears in this table. Every variable is represented in the graph by a vertex. |
accumulate_column | Any | Column to be copied to the output table. |
The computational cost for each vertex is exponential in terms of in-degree. If any vertex has an in-degree greater than 20, the function may be very slow.
If variables are conditionally dependent on each other, their representative vertices are connected by edges. The edges table contains the columns that comprise the keys of the source and target vertices of the edges, and optionally, a column that contains the weights of the edges.
Column Name | Data Type | Description |
---|---|---|
source_vertex_key_column | INTEGER | Column that is, or is part of, the key of the source vertex. Every column in the key appears in this table. |
target_vertex_key_column | INTEGER | Column that is, or is part of, the key of the target vertex. Every column in the key appears in this table. |
edge_weight | DOUBLE PRECISION | Optional column that contains the edge weights. The function uses only positive edge weights. The sum of the edge weights that the function uses must be 1. |
The observations table contains the vertices (which represent variables) and the observations for observed variables.
Column Name | Data Type | Description |
---|---|---|
vertex | INTEGER | Column that is, or is part of, the vertex key. Every column in the key appears in this table. |
observation | INTEGER | For an observed variable: 1 or 0 For an unobserved variable: NULL |