| Model Browser User's Guide | |
Model Evaluation Using Other Data
Prediction performance creates a resource dilemma for the experimenter. You could keep some data back to use later on to get a measure of the predictive capacity of the model, but why withhold data from the model building process if its addition is likely to improve the predictive capability of the model?
There is always a tradeoff in statistics when you decide how much of your data to use for model fitting and whether to leave some data to test your model against.
The Model Evaluation window is intended to help you validate your model against other data, although you can also evaluate the fit against the original data or without any data. The rest of the Model Evaluation dialog is only enabled when you choose Evaluate using other data, the last of three choices in the Data Selection dialog.
When you select the Evaluate using other data choice, if the input factors required to evaluate the model do not appear in the selected data set, another dialog appears to match signal names in the selected data set to those required to evaluate the model.
Match signal names from the data set (on the right) to the symbols required to evaluate the model (on the left) and click OK to accept this assignment.
In the example shown in Model Evaluation Window, the Data Selection dialog shows three data sets that you can use to evaluate this model: Data Object, Data Object_1, and New Data. Traceability information about the data set currently selected in the list box is always displayed in the Data info pane on the right.
In the lower Tests frame you can choose the tests from this data set to use. Certain test numbers might have been used to create the model (as in this case) and you can eliminate these test numbers from the evaluation data set if you want, by selecting the Filter out used log numbers check box.
Of the remaining tests, you can select some and reject others using the Selected Tests/Unselected Tests list boxes. You move tests between them using the arrow buttons. For the currently selected test, the mean test values of all variables in this data set are displayed on the right.
Click OK to use the selected tests to evaluate this model. The Model Evaluation window appears.
For information about the available views, see Selecting Models.
| | Model Evaluation Window | Exporting Models | |
