Weka

Output

The more_options menu on the Classify panel can be used to customized the output. Depending on the setup, Weka will generate one or more of the following sections:

• Run information (is always present)
• Stratified cross-validation. The “Folds” parameter is used for stratified cross-validation.
• Evaluation on test set (is present when a separate test set is used)
• Evaluation on training set
• Summary
• Classifier model (full training set)
• Predictions on test data
• Detailed accuracy by class
• Confusion matrix
• Source code

When using cross-validation, Weka prints a model built on the full dataset. The statistics, however, are calculated from the various train/test splits. This can be confusing, because the model stay the same regardless of the number of folds or the value of the random seed.

More options

The more_options menu on the Classify panel gives the following options:

• 1 Output model. The “Output model” option toggles whether or not the model built on the full dataset is printed.
• 2 Output per-class stats
• 3 Output entropy evaluation measures
• 4 Output confusion matrix
• 5 Store predictions for visualization: This option makes Weka save the classifier's predictions on the test data, and if the model is a tree it saves them at the appropriate leaves. With a very large test set, you might want to turn this off. If you plan to visualize the decision tree produced by J48, this option should you enable to see the classifier's errors on the tree
• 6 Output predictions: This prints a table with actual and predicted values for each instance of the test data. should be set in order to see how the learned classifier deals with each instance in a supplied test set.
• 7 Output additional attributes option should you configure if you want the classifier's predictions to show values of attributes other than the class. Here you can specify a comma-separated range of attribute indices whose values will be included along with the actual and predicted class values. For example, the specification “first-3,10,12-14” would include attributes 1, 2, 3, 10, 12, 13, 14.
• 8 Cost-sensitive evaluation
• 9 Random seed for XCal / % Split
• 10 Preserve order for % Split
• 11 Output source code: will generate Java code that represents the model produced by the classifier. This section prints Java code for the model built on the full training set. This code can be embedded in other Java applications.