libAGF

The library for [Adaptive Gaussian Filtering](Adaptive Gaussian Filtering) is a machine learning library based on variable bandwidth kernel density estimation using Gaussian filters. It is written in C++.

While the basic codes take all of the training data to make predictions, the option exists for pre-trained binary classifiers which store the location of the discrimination border as a series of points. Several binary classifiers can then be used in combination to perform multi-class classification. If the initial training data set is very large, these pre-trained classifiers will be much faster.

The multi-class classification routines are very general in the sense that there are many different ways in which to built up a set binary classifiers. The configuration is specified using a recursive control language. Since multi-class classifiers are not specific to AGF, the routines can also be applied to external binary classifiers, such as LIBSVM.