Description:

Implementation of the Probabilistic Classification Vector Machine (PCVM)in acc. to the paper of Huanhuan Chen et al. The code contains a full library implementation including test code which can be similarly used like the libsvm. Thanks to the used armadillo / boost framework it supports multicore calculations and runs very fast using the numerical lapack libraries.

The PCVM is similar to the Relevance Vector Machine (RVM) of Tipping but with a more appropriate probabilistic model (see paper). The classifier can be used for potentially indefinite input kernels so is directly (and valid) applicable for non-metric input similarities e.g. as obtained from sequence alignment data or using shape measurements. The code provides a recent extension for a Nystroem approximated PCVM such that it also scales to larger scale problems and has finally a linear runtime complexity (if we fix the number of landmarks).

The code contains some simple examples shown in the Readme and comes with a command line interface, model input / output and a crossvalidation. It can also be used for own input kernel matrices. Default implemented kernels are a linear kernel and an extreme learning machine (elm) kernel.

Changes to previous version:

Works now also with Matlab (see Readme - really - how it must be used)

BibTeX Entry: Download

Corresponding Paper BibTeX Entry: Download

Supported Operating Systems: Linux

Data Formats: Csv

Tags: Kernel, Classification, Kernel Methods, Machine Learning, Lapack, Sparse Learning, Algorithm, Sparse, Rvm, Multicore, Bayesian, Nonlinear Classification, Indefinite Kernels, Probabilistic Model, Structu

Archive: download here

Comments

No one has posted any comments yet. Perhaps you'd like to be the first?

Leave a comment

You must be logged in to post comments.