Report Options¶
Response Options¶
To access features specific to an individual response, click the red triangle menu icon in the SVEM for <response> outline box, where <response> represents the response name.
The red triangle menu options are described in the following sections.
Save Columns -> Prediction Formula¶
To save a column containing the prediction formula for the ensemble model to your data table, select Save Columns then Prediction Formula from the dropdown.
The new column is named Pred Formula <response>. It is assigned the Formula column property. You can view the formula by clicking the plus sign icon to the right of the column name in the columns panel, or by right-clicking in the column header to display a menu and selecting Formula.
For a continous response, the ensemble model is the average of the number of bootstrapped models that you specified in the Number of Bootstrapped Models field. For a categorical response, a probability column is saved for each level of the response. A Most likely <response> column is also saved.
The probability columns are averages of the probability columns for the simulated models. The Most Likely <response> column gives the level with the highest probability across the probability columns for the ensemble model.
Ensemble model formulas for Neural fits are large because they are averages of the symbolic prediction formulas for the bootstrapped models. The image below shows only a portion of the formula for a SVEM Neural fit for a continuous response. Ensemble model formulas for Generalized Regression fits are more manageable because they reduce to linear models.
Two other column properties are assigned to the prediction formula column: Notes and Predicting.
The Notes column property displays a list of settings that are used for your SVEM fit. The following example shows the Notes property settings for a SVEM Neural fit.
The Predicting column property indicates that the prediction formula was created using SVEM and gives the SVEM pathway, Neural (Neural with SVEM) or Generalized Regression (Fit Generalized with SVEM).
| Example for a Neural fit | Example for a Generalized Regression fit |
|---|---|
 |
 |
Save Columns -> Standard Error¶
To save a column containing estimates of the prediction error to the data table, select Save Columns then Standard Error from the dropdown. This option is not available for categorical responses or for Generalized Regression fits if you select a Distribution for the response other than Normal.
The new column is named PredSE <response>. It is assigned the Formula column property. You can view the formula by clicking the plus sign icon to the right of the column name in the columns panel, or by right-clicking in the column header to display a menu and selecting Formula.
The formula for the PredSE <response> column is an empirically based estimate of prediction error:
where
-
represents the average of the RASE values for all bootstrapped models
- If out-of-bag rows were used, this is the average of all out-of-bag RASEs, otherwise it is the average of all validation RASE
- for each row, represents the standard deviation of the predicted values across all bootstrapped models.
Save Columns -> Residuals¶
To save a column containing the residuals in the data table select Save Columns then Residuals from the dropdown menu. This option is only available for continuous responses.
Prediction Profiler¶
To display the Profiler outline box, which contains the Prediction Profiler, select Profiler. For continuous predictors, prediction bands that are based on the PredSE <response> column are shown. These prediction bands provide approximately 95% bounds on individual response values.
Note that all of the usual options are available from Profiler's red triangle menu.
Important Note
Do not save the Profiler script to the data table because in JMP 17 and later versions, the script cannot be saved and in earlier versions, the script will not run. If you want to save work that has been generated using the Profiler, it's strongly recommended that you save the prediction formula to the data table, and then access the Profiler from the Graph menu > Profiler. This approach enables you to save the Profiler script to your data table for future use.
When you select Profiler from the SVEM red triangle menu, a warning appears to alert you not to save the Profiler script to the data table. If you want to disable this warning, turn on the corresponding checkbox.
Show Bootstrap Results Table¶
To display a data table that contains parameter estimates, intermediate values, and RASE values for all bootstrapped models, select Show Bootstrap Results Table. Each row of the data table contains values for a given bootstrapped model. The first column, named Boot#, gives the number of the bootstrapped model. The second column, Included In Model, indicates if the bootstrap run was excluded from the model based on Proportion of Runs to Exclude. These columns are followed by columns that give the parameters of the bootstrap model, the Validation RASE for that model, and the OOB RASE if Leave K Out was selected. Also, if Leave K Out is selected, the final column(s) show the Test Rows used for the given bootstrap run.
The table below shows results for a SVEM Generalized Regression fit based on 50 bootstrapped models (50 rows).
Save a SVEM Script to a Data Table¶
To save an encrypted script to your data table that will rerun SVEM with the specifications that produced the given SVEM report for this response, select Save Script to Data Table. The script is named SVEM <response>. Running this script reproduces your results because the script contains the random seed that was used in your original calculation.
Feature Options at the Parent Outline Box¶
If multiple responses were chosen or a number range of hidden nodes was specified, the SVEM report will have a parent outline box containing individual reports for each of the SVEM fits. To access additional features that apply to all SVEM fits contained in the parent outline box, click the red triangle menu icon.
Save All -> Prediction and Standard Error Formulas¶
Saves interleaved prediction formula and standard error columns to the data table for all SVEM fits in the outline box.
Save All -> Prediction Formulas¶
Saves prediction formula columns to the data table for all SVEM fits in the parent outline box.
Save All -> Standard Errors¶
Saves columns containing estimates of the prediction error to the data table for all SVEM fits in the parent outline box.
Save All -> Residuals¶
Saves residual columns to the data table for all responses in the report.
Save a SVEM Script to a Data Table¶
Saves a script to the data table that will reproduce all the SVEM fits in the parent outline box.