The HPSVM Procedure
ODS Tables
Each table that the HPSVM procedure creates has a name associated with it. You must use this name to refer to the table in ODS statements. These table names are listed in Table 1.
Table 1: ODS Tables Produced by PROC HPSVM
| Table Name | Description | Statement, Option |
|---|---|---|
| CLASSIFICATIONMATRIX | Classification matrix table | Default |
| FITSTATISTICS | Accuracy information of the training | Default |
| ITERHISTORY | Iteration history | Default |
| MODELINFORMATION | Basic model information for the training | Default |
| NOBS | Observation information about the input data | Default |
| PENALTYSELECTION | List of cross validation statistics for each penalty | SELECT, PENALTY |
| PERFORMANCEINFO | Information about the high-performance environment | Default |
| TIMING | Time used during the training | PERFORMANCE, DETAILS |
| TRAININGRESULT | Displays the training results | Default |
The following list provides more information about these tables:
- CLASSIFICATIONMATRIX
The "Classification Matrix" table contains the target information, both observed and predicted. The columns include observed target, predicted event, predicted non-event, and total numbers of events or non-events for the training data. If the PARTITION statement is specified, the validating information is also displayed.
- FITSTATISTICS
The "Fit Statistics" table contains the model accuracy information, which includes accuracy, error, sensitivity, and specificity. The statistics are calculated from the "Classification Matrix" table.
- ITERHISTORY
The "Iteration History" table contains the number of iterations, the complementarity, and the feasibility. The complementarity is controlled by the TOLERANCE statement and the MAXITER option, which controls the number of iterations. This table is produced only by the interior point method.
- MODELINFORMATION
The "Model Information" table contains the initial training settings, such as task type, optimization technique, and kernel function type. If the kernel function type is polynomial, then the kernel degree is also displayed.
- NOBS
The table contains the number of observations and the number of observations used.
- PENALTYSELECTION
The "Penalty Selection" table shows the cross validation statistics. It includes Penalty, True Positive, True Negative, False Negative, False Positive, Misclassification, and Accuracy. The table is generated if either the PENALTY statement or the SELECT statement is specified. The best penalty C is the first penalty that has the best accuracy value. If you specify only the PENALTY statement, the statistics in the table are based on the entire input data set.
- PERFORMANCEINFO
The "Performance Information" table is produced by default. It displays the execution mode and the number of threads used during the computation.
- TIMING
The "Procedure Task Timing" table shows the time consumed during the process, which includes the times for data processing and model training.
- TRAININGRESULT
The "Training Results" table shows the model information. It includes but is not limited to the inner product of weights, bias, and the number of support vectors.