SANN - Results - Custom Predictions Tab

Neural Networks

Select the Custom predictions tab of the SANN - Results dialog box to access the options described here. You can produce predictions from the list of selected (active) neural networks for any kind of task - including regression, classification, time series, and cluster analysis.

Using the options on this tab, you can define new cases (that are not drawn from the data set) and execute the selected models against them. You may also select individual cases from the data set, modify, and then execute them, allowing you to perform some ad hoc analysis. The predictions of the neural network models are accumulated in a data grid, which you can transfer to a separate results spreadsheet.

Number of cases to predict. To produce custom predictions, first decide how many cases you want to create, and then type the total into this box. Note that you can have a maximum of 100 custom predictions at a time. If the number of custom predictions you already have plus the number of custom predictions you are about to make is larger than a 100, you will be notified to that effect. Any attempt to make custom predictions beyond the maximum limit will erase the existing ones. In that case, you can use the Custom predictions button (see below) to save your existing results in a spreadsheet.

Clear previous predictions. Select this check box to delete previous custom predictions and show only new predictions. Clear this check box to keep previous predictions displayed in the data grid in addition to new predictions.

Custom inputs. Click the Custom inputs button to display a user-defined input spreadsheet where you can enter the specifics of your custom predictions. Click the OK button in this dialog to close it and return to the SANN - Results dialog box - Custom predictions tab. The custom prediction results will be displayed in the data grid on this tab (see below).

Note: With the user-defined input spreadsheet, you can enter the input values for making custom predictions. For continuous inputs, the values must be real numbers. For categorical inputs, they can be either the code values of their corresponding text. For example, for a categorical variable with two categorical levels (Male, 101) and (Female, 102), you can enter either Male or 101 for the first category and Female or 102 for the second. Please note that entering Male or male is the same, since text entries are case insensitive.

Custom predictions. Click the Custom predictions button to transfer the current custom predictions to a spreadsheet. The information in this spreadsheet is identical to the one displayed in the Custom predictions data grid.

Custom predictions data grid. This data grid is located directly below the option above. Its purpose is to display full details of the existing custom predictions for all active networks.

For regression, classification, and time series tasks, the order of the displayed information is as follows:

  • Column 1:  Custom prediction number

  • Column 2-N:  Predictions of first, second, …, Nth networks

  • Column N+1:  Custom inputs

For cluster analysis task, the arrangement of information in the custom data grid is slightly different due to the nature of the Kohonen networks:

  • Column 1:  Custom prediction number

  • Column 2-3xN:  ID, position and activations of the first, second, …, Nth Kohonen network

  • Column N+1…:  Custom inputs

Note: The Custom predictions tab allows you to make predictions with completely user specified values for the inputs. For categorical variables, these values (i.e. categorical levels) must have the right code or text. If a wrong value is entered, the prediction cells in the data grid will remain empty indicating that no predictions were possible. On the other hand, for continuous inputs you may use any range of values for the variables. Be aware that using custom values outside the (minimum and maximum) training data limits may yield unreliable predictions.