Evolving Classification Rules¶
This tutorial gives a basic example of how to solve classification problems with RGP via Typed Genetic Programming. This tutorial assumes a basic understanding of Typed Genetic Programming, which can be acquired in the other Tutorials.
Typed Genetic Programming (GP) is a flexible and very general problem solving method that can also be applied to find classification rules via evolutionary search. Compared to specialized classification methods,
- GP has the potential to find compact human-readable classification rules, if the GP search space (function set, constant set, input variable set) is properly configured.
This capability comes at a price, as
- configuring the GP search space can be a complex task that requires domain knowledge. Oftentimes, the GP search space can become huge and extremely difficult, degrading evolutionary search to random search.
- On many problem domains, methods specialized for classification, like Learning Classifier Systems (LCS), will often give better results in much shorter time.
- If human-interpretable rules aren't very important for your application, you should try modern machine learning methods for classification, such as Support Vector Machines (SVM) and Random Forrest (RF). MLR is comprehensive, efficient and high-quality package of state-of-the-art machine learning methods that also includes comprehensive documentation. Modern machine learning methods will give you high quality classifiers in a very small fraction of the time of a typcial GP run (minutes instead of hours or days).
Step 0: Defining the Problem¶