Tutorial

This is a walk-through of steps to use JBoost. Complete documentation of features can be found in the documentation section. All the links are internal to this site and give examples/documentation on how to do the specified task.

• Download and install JBoost.
• Decide which features you are going to use for your dataset. Boosting, unlike many other machine learning methods, is fairly robust to "useless" features. In other words: include as many features as time permits.
• Create an input file. Start by creating a single file with all examples. Alternatively, you can use one of the datafile examples in the demo directory.
• Now that you have one file with all examples, you can do one of two things: 1) perform cross validation or 2) separate the data file into a training set and a test set. For preliminary analysis, a training and test set will be faster and may help you decide how to add more features and tweak boosting parameters.
• To assist you in tweaking parameters, JBoost comes with a variety of visualization tools. The two most important things to look at are: error and margin curves. In addition, you should also look at the pdf/png of the ADTree for a sanity check.
• You eventually have to do cross validation. This is one of the best pieces of evidence to show that your classifier works.
• At this point you have completed the above steps and you have a robust classification system as demonstrated through cross validation. Typically, classifying an instance that already had a label (as is the case in test/train sets and CV analysis) isn't really helpful. Thus, JBoost provides a mechanism for outputting the classifier to enable classification independent of JBoost. This classifier can incorporated into your own web server or software so that your classifier can be used by others.