The OPTNET Procedure

The DATA_LINKS= option in the PROC OPTNET statement defines the data set that contains the list of links in the graph. A link is represented as a pair of nodes, which are defined by using either numeric or character labels. The links data set is expected to contain some combination of the following possible variables:

As described in the GRAPH_DIRECTION= option, if the graph is undirected, the from and to labels are interchangeable. If the weights are not given for algorithms that call for link weights, they are all assumed to be 1.

The data set variable names can have any values that you want. If you use nonstandard names, you must identify the variables by using the DATA_LINKS_VAR statement, as described in the section DATA_LINKS_VAR Statement.

For example, the following two data sets identify the same graph:

data LinkSetInA;
   input from $ to $ weight;
   datalines;
A B 1
A C 2
A D 4
;

data LinkSetInB;
   input source_node $ sink_node $ value;
   datalines;
A B 1
A C 2
A D 4
;

These data sets can be presented to PROC OPTNET by using the following equivalent statements:

proc optnet
   data_links = LinkSetInA;
run;

proc optnet
   data_links = LinkSetInB;
   data_links_var
      from    = source_node
      to      = sink_node
      weight  = value;
run;

The directed graph G shown in Figure 2.5 can be represented by the following links data set LinkSetIn:

data LinkSetIn;
   input from $ to $ weight @@;
   datalines;
A B 1  A C 2  A D 4  B C 1  B E 2
B F 5  C E 1  D E 1  E D 1  E F 2
F G 6  G H 1  G I 1  H G 2  H I 3
;

The following statements read in this graph, declare it as a directed graph, and output the resulting links and nodes data sets. These statements do not run any algorithms, so the resulting output contains only the input graph.

proc optnet
   graph_direction = directed
   data_links      = LinkSetIn
   out_nodes       = NodeSetOut
   out_links       = LinkSetOut;
run;

The data set NodeSetOut, shown in Figure 2.6, now contains the nodes that are read from the input link data set. The variable node shows the label associated with each node.

The data set LinkSetOut, shown in Figure 2.7, contains the links that were read from the input link data set. The variables from and to show the associated node labels.

If you define this graph as undirected, then reciprocal links (for example, and ) are treated as the same link, and duplicates are removed. PROC OPTNET takes the first occurrence of the link and ignores the others. By default, GRAPH_DIRECTION=UNDIRECTED, so you can just remove this option to declare the graph as undirected.

proc optnet
   data_links = LinkSetIn
   out_nodes  = NodeSetOut
   out_links  = LinkSetOut;
run;

The progress of the procedure is shown in Figure 2.8. The log now shows the links (and their observation identifiers) that were declared as duplicates and removed.

The data set NodeSetOut is equivalent to the one shown in Figure 2.6. However, the new links data set LinkSetOut shown in Figure 2.9 contains two fewer links than before, because duplicates are removed.

Certain algorithms can perform more efficiently when you specify GRAPH_INTERNAL_FORMAT=THIN in the PROC OPTNET statement. However, when you specify this option, PROC OPTNET does not remove duplicate links. Instead, you should use appropriate DATA steps to clean your data before calling PROC OPTNET.

The DATA_NODES= option in the PROC OPTNET statement defines the data set that contains the list of nodes in the graph. This data set is used to assign node weights.

The nodes data set is expected to contain some combination of the following possible variables:

You can specify any values that you want for the data set variable names. If you use nonstandard names, you must identify the variables by using the DATA_NODES_VAR statement, as described in the section DATA_NODES_VAR Statement.

The data set that is specified in the DATA_LINKS= option defines the set of nodes that are incident to some link. If the graph contains a node that has no links (called a singleton node), then this node must be defined in the DATA_NODES data set. The following is an example of a graph with three links but four nodes, including a singleton node D:

data NodeSetIn;
   input label $ @@;
   datalines;
A B C D
;

data LinkSetInS;
   input from $ to $ weight;
   datalines;
A B 1
A C 2
B C 1
;

If you specify duplicate entries in the node data set, PROC OPTNET takes the first occurrence of the node and ignores the others. A warning is printed to the log.

For some algorithms you might want to process only a subset of the nodes that appear in the input graph. You can accomplish this by using the DATA_NODES_SUB= option in the PROC OPTNET statement. You can use the node subset data set in conjunction with the SHORTPATH statement (see the section Shortest Path. The node subset data set is expected to contain some combination of the following variables:

The values in the node subset data set determine how to process nodes when the SHORTPATH statement is processed. A value of 0 for the source variable designates that the node is not to be processed as a source; a value of 1 designates that the node is to be processed as a source. The same values can be used for the sink variable to designate whether the node is to be processed as a sink. The missing indicator (.) can also be used in place of 0 to designate that a node is not to be processed.

A representative example of a node subset data set that might be used with the graph in Figure 2.5 is as follows:

data NodeSubSetIn;
   input node $ source sink;
   datalines;
A 1 .
F . 1
E 1 .
;

The data set NodeSubSetIn indicates that you want to process the shortest paths for the source-sink pairs in .

For large-scale graphs, the processing stage that reads the nodes and links into memory can be time-consuming. Under the following assumptions, you can use the STANDARDIZED_LABELS option in the PROC OPTNET statement to greatly speed up this stage:

Consider the following links data set that uses numeric labels:

data LinkSetIn;
   input from to weight;
   datalines;
0 1 1
3 0 2
1 5 1
;

Using default settings, the following statements echo back link and node data sets that contain three links and four nodes, respectively:

proc optnet
   data_links = LinkSetIn
   out_nodes  = NodeSetOut
   out_links  = LinkSetOut;
run;

The log is shown in Figure 2.10.

The data set NodeSetOut, shown in Figure 2.11, contains the unique numeric node labels, .

Using standardized labels, the same input data set defines a graph that has six (not four) nodes:

proc optnet
   standardized_labels
   data_links = LinkSetIn
   out_nodes  = NodeSetOut
   out_links  = LinkSetOut;
run;

The log that results from using standardized labels is shown in Figure 2.12.

The data set NodeSetOut, shown in Figure 2.13, now contains all node labels from 0 to 5, based on the assumptions when you use the STANDARDIZED_LABELS option.

When you use standardized labels, the DATA_NODES= input order (which can be arbitrary) is not preserved in the OUT_NODES= output data set. Instead, the order is ascending, starting from zero.