One page tutorial
Previous  Top  Next

This One page tutorial illustrates how to use the PNC2 Rule Induction System to learn a model and validate it using a separate test data set. The learn and test data files used are wine1.dat and wine2.dat. They can be found in the same directory as the executable - in most cases this wille be C:\Programs\Pnc2\ .


tick
Loading the learn data file



Start the program. You will see just a single tab sheet named Basic. Hit the Load button and select the file wine1.dat. You find this file in the subdirectory examples. The file will be loaded, you will see the number of data tuples and variables, you will see that column 1 is selected as output and that the given problem is a classification task. More Info The tab sheet Basic



tick
Automatic parameter tuning



The PNC2 Cluster Algorithm contains several free parameters that allow an adjustment of the algorithms behavior to the actual learn data sample. However, one needs some experience to optimally chose the parameter values. Thus an automated parameter tuning component has been integrated into the PNC2 Rule Induction System.

First, you have got to define some parameter sets. Hit the Edit Sets button. The Parameter Sets dialog will pop up. Hit the button Go in the lower left corner to use the task wizard to automatically generate suitable parameter sets. Close the dialog by hitting Ok.

Now the button
 Tune is enabled. Hit it. This will start the tuning process and open the new tab sheet Tuning Results. The prediction accuracy of each parameter set is estimated using cross-validation or a similar approach within the learn data sample. Results are displayed as soon as they are available. Wait until the tuning is finished. Now click on the header of the column Error to sort the parameter sets with respect to the corresponding prediction accuracy. Select the best parameter set, as long as it is not colored red or yellow (light yellow is ok) by double clicking it. Now the program changes back to the tab sheet Basic. The check box Skip Tuning is checked now, the group box Parameters is enabled and the values of the selected parameter set have been copied to the corresponding edit fields. More Info The tab sheet Tuning Results



icon_tip
Alternative: Skipping the parameter tuning


You may skip the parameter tuning by just checking the check box Skip Tuning. This will immediately enable the group box Parameters and initialize the different values with suitable defaults.



tick
Learning a model



Start the rule induction process by hitting the Learn button. The tab sheet Model appears and after the learning is finished some model characteristics, like the number of rules induced, are shown. Also the tab sheet Use Model is visible now. More Info The tab sheet Model



tick
Use the model



Now we are going to use the learned model to predict the output values for some test data tuples based on the corresponding input space positions.

Change to the tab sheet Use Model and hit the button Load to load wine2.dat as test data file. You may modify the parameters used for testing by checking the Modify check boxes in the corresponding group box. Also, you may specify an output file, to which the predicted output values are written, by hitting the Set Output button. Hit the Test button to start the prediction process. The test data prediction accuracy is displayed afterwards, if the test data file contains the real output values. More Info The tab sheet Use Model



icon_tip
A short note on the test data file



The test data file must either have the same number of columns as the learn data file or it must have exactly one less. In the first case the test data file is with output column, in the latter case it is without. The input and - if present - the output columns must be ordered in the same way as in the learn data file. More Info