Changes from 1.2 to 2.0 *********************** 03-04-2005 - upgrade LIBSVM to version 2.71 03-13-2005 - add Pearson correlation as feature selection criterion 03-13-2005 - remove clustering. There are better programs for clustering elsewhere. 03-13-2005 - remove distance selection menu. 03-15-2005 - support individual class costs for C-SVM learning 03-17-2005 - eliminate FORTRAN code 03-28-2005 - add Model Selection menu for Support Vector Machine algorithm 04-19-2005 - add forward selection and backward elimination feature subset selection algorithms 04-19-2005 - add inter-intra distance, 1-NN error rate and Bayes error rate as criteria for feature selection 06-15-2005 - use GNU autoconf to build PCP Changes from 2.0 to 2.1 *********************** 07-13-2005 - use -O3 flags and dynamic building in the distributed Makefiles 08-09-2005 - SVM model selection now automatically creates the prediction file pcp.rcl 08-10-2005 - handle empty classes in the test set for named data format 08-11-2005 - create prediction file pcp.rcl for MLP prediction 09-02-2005 - implement MLP model selection. Rename model selection file pcp.svp to pcp.msl. 09-03-2005 - implement k-NN model selection 11-18-2005 - add additional information in class prediction file pcp.rcl (correct classification flag, TP, FN, FP, TN flags for two-class case) 11-22-2005 - remove major memory handling defect in forward selection algorithm, leading to poor (computational) performance of model selection involving forward selection 11-25-2005 - enforce the feasible region for nu in NU-SVM 02-02-2006 - change default number of cross-validation experiments from 10 to 1 Changes from 2.1 to 2.2 *********************** 02-12-2006 - support model selection for linear SVM kernel 04-17-2006 - introduce an option to build SVD transform using training+test datasets (as opposed to just training data) 05-09-2006 - report p-error in SVM model selection 05-21-2006 - simplify build process