IntroductionThe current version of malibu is 1.1 triarii. The triarii were the veteran soldiers in the Roman legion forming the third standard line of infantry; malibu has undergone three major releases and triarii represents the first release to reach a level of maturity such that it can be easily used by members of the wider scientific community. The primary users of malibu will be researchers or students with a solid background in computing (e.g. while it runs under windows, it was primarily developed for Linux) and some understanding of machine learning. malibu is organized around a core set of binary classifiers and was originally developed to support binary classification. As research projects progressed, other interesting learning problems cropped up and a set of wrappers were added to allow the binary classifiers to solve such problems. For example, malibu includes the Costing, which allows any binary classifier to handle both importance weighted and cost-sensitive classification problems. This approach to wrapping binary classifiers (as opposed to using only those that can naturally support the problem) has two primary advantages. First, no one classifiers performs well on every problem; each classifier has a particular bias that allows it to perform well on some problems while performing poorly on others. Thus, using wrappers gives the user access to a larger number of algorithms from which to choose the one with the best bias. Second, the wrappers can be layered to create any number of algorithms handling a wide variety of problems. For example, the a calibration algorithm can be layered on the Costing algorithm (on a classifier) to solve a cost-sensitive learning problem requiring accurate probability estimates of confidence in a prediction. Features- Learning problems
- Learning utilities
- Cluster computing (MPI)
- ANSI C++: Tested on Windows and Linux
Learning Problems- Binary classification
- Cost-sensitive classification
- Multiclass classification
- Multiple-instance learning
- Probabilistic regression
- Regression
Utilities- Dataset partition
- Attribute transform
- Prediction evaluation
|
Sign in to add a comment
