GVARCLUS Procedure
PROC GVARCLUS Statement
PROC GVARCLUS <options>;
The PROC GVARCLUS statement invokes the procedure. Table 1 summarizes the important options in the PROC GVARCLUS statement by function. The options are then described fully in alphabetical order.
Table 1: PROC GVARCLUS Statement Options
| Option | Description |
|---|---|
| Basic Options | |
| DATA= | Specifies the input data table |
| Options Related to Variable Clustering | |
| EXACT | Performs graphical variable clustering without preprocessing |
| MAXITER= | Specifies the maximum number of iterations for each estimation step |
| MAXMEMBER= | Specifies the maximum number of members within each final cluster |
| MAXSTEP= | Specifies the maximum number of clustering steps |
| MINCLUSTER= | Specifies the minimum number of clusters |
| RHO= | Specifies the initial value of the regularization parameter |
| SELECT= | Selects the optimal clustering on the basis of the specified criterion |
| STOP= | Requests that the clustering process stop if the clustering structure does not change in the previous number of specified steps |
| TOL= | Specifies the threshold of convergence |
| Options for Output Tables | |
| OUTCP= | Outputs the covariance matrix from the last step |
| OUTEDGE= | Outputs the edges for defining an undirected network |
| OUTTREE= | Outputs the hierarchical clustering results in a tree structure |
| OUTVERT= | Outputs the vertices and their size for defining an undirected network |
You can specify the following options:
- DATA=libref.data-table
-
names the input data table for PROC GVARCLUS to use. The default is the most recently created data table. libref.data-table is a two-level name, where
- libref
refers to a collection of information that is defined in the LIBNAME statement and includes the
library, which includes a path to the data, and a session identifier, which defaults to the active session but which can be explicitly defined in the LIBNAME statement. For more information about libref, see the section Using CAS Sessions and CAS Engine Librefs.- data-table
specifies the name of the input data table.
- EXACT
performs graphical variable clustering without preprocessing by thresholding the sample covariance into connected components. By default, the preprocessing step is performed.
-
MAXITER=n
MAXITERS=n specifies the maximum number of iterations for estimating the sparse precision matrix by using coordinate descent. By default, MAXITER=50.
-
MAXMEMBER=n
MAXMEMBERS=n stops PROC GVARCLUS when the number of members within any cluster is greater than or equal to n.
-
MAXSTEP=n
MAXSTEPS=n stops PROC GVARCLUS after it runs n steps. By default, MAXSTEP=3.
-
MINCLUSTER=n
MINCLUSTERS=n stops PROC GVARCLUS when the number of clusters is less than or equal to n. By default, MINCLUSTER=3.
- OUTCP=libref.data-table < / LIST <EPS=number>>
-
creates a data table that contains a covariance matrix that depicts the relationships among variables and also creates a set of statistics about the input data table and variables. libref.data-table is a two-level name, where libref refers to the library, and data-table specifies the name of the output data table. For more information about this two-level name, see the DATA= option and the section Using CAS Sessions and CAS Engine Librefs.
When you specify the LIST option, the symmetric matrix is output in the list-of-list (LIL) format. In this format, the matrix is represented as a set of tuples
, where x is an entry in the matrix and i and j denote its row and column indices, respectively. LIL format can be used when the output contains too many columns to fit in a data table. For example, in most database systems, the maximum number of columns in a table is usually limited to several thousand. If an output matrix contains more columns than the limit, you must use the LIST option in order to avoid errors that would arise from writing too many columns to the table. When LIL format is used, all entries in the matrix that have a value of 0 are ignored in the output. When you specify EPS=number in the LIST suboption, matrix entries whose absolute value is smaller than number are ignored in the output. This feature helps omit unreliable estimations and generate a compact representation of the matrix. When you do not specify the EPS= option, only the 0 entries in the matrix are ignored in the output.
- OUTEDGE=libref.data-table
creates a data table that contains the information that defines the edges in the network,
FROM, TO, and WEIGHT. libref.data-table is a two-level name, where libref refers to the library, and data-table specifies the name of the output data table. For more information about this two-level name, see the DATA= option and the section Using CAS Sessions and CAS Engine Librefs. - OUTTREE=libref.data-table
creates a data table that depicts a tree diagram to display the hierarchical clustering results. The tree diagram can be plotted using the DENDROGRAM statement in the Graph Template Language. libref.data-table is a two-level name, where libref refers to the library, and data-table specifies the name of the output data table. For more information about this two-level name, see the DATA= option and the section Using CAS Sessions and CAS Engine Librefs.
- OUTVERT=libref.data-table
creates a data table that contains the vertices in the network and their size. libref.data-table is a two-level name, where libref refers to the library, and data-table specifies the name of the output data table. For more information about this two-level name, see the DATA= option and the section Using CAS Sessions and CAS Engine Librefs.
- RHO=fraction
specifies the value of
that determines the sequence of regularization parameters, 
, that are used on sequential clustering steps. The higher the value, the more regularization and the sparser the inverse covariance. By default, RHO=0.8. - SELECT=NONE | PENALIZED
-
selects the optimal clustering on the basis of the specified criterion. You can specify the following values:
- NONE
performs no selection, and the output tables of the edges and vertices of the network reflect the clustering from the last step.
- PENALIZED
performs selection through minimizing the penalized log likelihood, as shown in the section Selection.
By default, SELECT=NONE.
- STOP=n
requests that the procedure stop if the clustering results do not change in the previous n consecutive steps. By default, STOP=3.
-
TOLERANCE=fraction
TOL=fraction specifies the minimal absolute tolerance at which an iteration stops. The tolerance number must be equal to or greater than 1.0E–12. By default, TOLERANCE=1.0E–3.