The Pattern Recognition Toolbox Users Guide
The Pattern Recognition Toolbox Users Guide begins with an introduction to the various dataset objects that are necessary to perform classification or regression.
The next topic is the Pattern Recognition Toolbox Engine, which discusses the syntax for prtAction objects. This syntax is common for most of the functionality in the Pattern Recognition Toolbox.
Following, Classifiers and Pre-processing techniques are presented.
Algorithm objects allow multiple prtActions to be connected together, and use the same training, running and evaluation syntax as a single prtAction.
Regression is similar to classification, but maps observations to a continuum of numbers, as opposed to a discrete set of class labels.
Clustering is also closely related to classification, however, in clustering, class labels are not used during training.
Scoring and evaulation provide methods to analyze the results of your classification, regression or clustering. Scoring functions work on prtDataSets, while evaluation functions work on prtAction objects.
Decision objects take the outputs of a prtAction object and make decisions according to particular criteria, such as the minimum probability of error.
Kernels are useful in many nonlinear classification and regression problems. The Pattern Recognition Toolbox provides a suite of common kernels to be used in conjunction with prtClass and prtRegress objects.
The Pattern Recognition Toolbox also provides a set of distance functions, such as Euclidean, Mahalonobis and other distance metrics, to be used in conjunction with prtActions.
Random variables often form important components of prtActions, and a set of random variable objects is provided.
Feature selection is a technique that helps select the features that have the greatest effect on the performance of prtActions.
Often, data collected from real world applications may contain outliers that are not relevant, and that may skew results. The Pattern Recognition Toolbox provides a set of outlier removal objects.
The Pattern Recognition Toolbox provides a set of data generation functions, which can be useful for creating example or prototying.
Finally, the Patern Recognition Toolbox provides support for developing and implementing your own features.
Copyright 2013 New Folder Consulting L.L.C.