docNumSelect

Centre for Vision, Speech & Signal Processing

Feature Selection Class Structure and Usage

In order to perform feature selection it is usually necessary to be able to evaluate the performance of classification with any subset of the features and then perform a search of feature space in order to find the subset which produces the best results. Performance evaluation can be achieved using classes derived from NumPerformanceC and NumErrorC.

With any class derived from NumSelectC this is possible by calling the Design member function with the appropriate parameters (sample data, classifier, error counting function and performance evaluator). After this, since NumSelectC is derived from NumFuncC, any feature vector can be transformed to the best subset using the Evaluate or operator() member functions.

Author: Robert Crida, Created: 12/6/1998, Generated by DocCreate: March 20, 2001