PRT Product Overview
Contents
PRT Toolbox Overview
The PRT is a collection of MATLAB objects and utility functions for developing pattern recognition algorithms. The toolbox provides a command-line interface to let you rapidly develop classification algorithms and evaluate algorithm performance.
Installation
Please see: Installing the PRT
Using This Guide
If you are new to pattern recognition in general, start out by reading What is Pattern Recognition?. Otherwise, if you want to dive right into the PRT, see Some examples of using the PRT. For installation and documentation questions, please see Installing the PRT.
PRT Organization
The PRT provides several hundred MATLAB M-files implementing various pattern classification objects and utility functions. To prevent cluttering the MATLAB namespace, all PRT classes and functions begin with the string "prt". Typing "prt", then tab at the MATLAB command prompt will bring up a list of all the PRT M-files available (assuming the PRT's path has been set properly, see "Installation" below).
To simplify finding the M-file a user is looking for, the PRT implements a hierarchical M-file naming scheme, so that all PRT M-files of a particular type start with "prtType" where Type is replaced by a short mnemonic describing the object or function implemented in the M-file.
The following are some commonly used prt M-file prefixes:
- prtClass* - M-file classes for implementing classification algorithms. Examples include prtClassFld, prtClassKnn, and prtClassRvm.
- prtCluster* - M-file classes for implementing clustering algorithms. Examples include prtClusterGmm and prtClusterKmeans.
- prtDataGen* - M-file functions for creating standard data sets for experimentation and algorithm evaluation. Examples include prtDataGenUnimodal, prtDataGenBimodal, and prtDataGenIris.
- prtDataSet* - M-file classes implementing data storage and bookkeeping for collections of data, known as "data sets". The most commonly used prtDataSet classes are prtDataSetClass and prtDataSetRegress.
- prtDecision* - M-file classes implementing decision making. These are commonly used after prtClass objects to turn continuous outputs from classifiers into binary or M-ary decisions. Examples include prtDecisionBinaryMinPe and prtDecisionMap.
- prtEval* - M-file functions for evaluating the performance of prtAlgorithms. These are commonly used in prtFeatSel objects to define a metric to optimize over. Examples include prtEvalAuc and prtEvalPfAtPd.
- prtKernel* - M-file classes implementing Kernel functions. Kernels are often used in machine learning, and in particular are used in prtClassRvm and prtClassSvm. Example prtKernel objects are prtKernelRbf and prtKernelDc.
- prtPreProc* - M-file classes implementing data transformations often used in data pre-processing. Commonly used objects are prtPreProPca, prtPreProcZmuv, and prtPreProcLda.
- prtPlotUtil* - M-file functions used internally to the PRT. These are mostly undocumented and are subject to change. They should not be called directly.
- prtRegress* - M-file classes for implementing regression algorithms. Examples include prtRegressLslr and prtRegressRvm.
- prtRv* - M-file classes for implementing random variables. These classes encapsulate PDF and CDF calculation, and maximum likelihood parameter estimation. Examples include prtRvMvn and prtRvUnif.
- prtScore* - M-file functions for evaluating the performance of different PRT algorithms. Examples inclide prtScoreRoc and prtScoreConfusionMatrix.
- prtUtil* - M-file functions used internally to the PRT. These are mostly undocumented and are subject to change. They should not be called directly.