最小二乘支持向量机工具箱使用指南

【实例简介】
最小二乘支持向量机工具箱使用指南详细介绍了各种函数的使用方法，并带有分类和回归分析的程序例子。
Preface to Ls-SVMLab v18 LS-SVMLab v1 8 contains some bug fixes from the previous version When using the preprocessing option, class labels are not considered as real variables. This probleIn occurred when the nUnber of dimensions were larger thanl the nuinber of data point The error "matrix is not positive definite"in the crossvalidatelssvm command has been olved e The error in the robustlssym command with functional interface has been solved. robustlssvm now only works with the object oriented interface. This is also adapted in the manual at pages 33 and 99 The crror "Rcfcrcncc to non-cxistent ficld implementation " has bccn solved in the bay_optimize command The ls-sVMlab Team Heverlee, belgium June 2011 4 Preface to LS-SVMLab v17 We have added new functions to the toolbox and updated some of the existing commands with respect to the previous version v1.6. Because many readers are familiar with the layout of version 1.5 and version 1.6, we have tried to change it as little as possible. Here is a summary of the main changes The major diffcrencc with the previous vcrsion is the optimization routinc uscd to find the minimum of the cross-validation score function. The tuning procedure consists out of two steps: 1) Coupled Simulated Annealing determines suitable tuning parameters and 2) a simplex method uses these previous values as starting values in order to perform a fine tuning of the parameters. The major advantage is speed. The number of function evaluations needed to find optimal parameters reduces from +200 in v1.6 to 50 in this version The construction of bias-corrected approximate 100(1-a)% pointwise/ simulataneous con- fidence and prediction intervals have been added to this version Some bug-fixes are performed in the function roc. The class labels do not need to be +1 or but can also be 0 and 1. The conversion is automatically done The ls-SVMLab Team Heverlee. belgium September 2010 5 6 Preface to Ls-SVMlab v16 We have added new functions to the toolbox and updated some of the existing commands with respect to the previous version v1.5. Because many readers are familiar with the layout of version 1.5, we have tried to change it as little as possible. The major difference is the speed-up of several methods. Here is a summary of the main changes Chapter/solver/function What's new 1. A birds eye on LS-SVMLab 2. LS-SVMLab toolbox examples Roadmap to LS-SVM; Addition of more regres- sion and classification examples; Easier interface for multi-class classification; Changed implementation for robust LS-SVM 3. Matlab functions ossibility of regression or classification using only one command! The function validate has been deleted; Faster(robust )training and (robust )model selection criteria are provided; In case of robust re gression different weight functions are provided to be used with iteratively reweighted LS-SVM 4. LS-SVM solver All CMEX and/or C files havc bccn rcmovcd. Thc linear system is solved by using the Matlab com nd "backslash”(\) The Ls-SVMLab Team Heverlee, Belgium June 2010 8 Contents 1 Introduction 11 birds eye view on ls-syMlab 13 2.1 Classification and regression 13 2.1.1 Classification extension 14 14 2.1.3 12.2 NARX models and prediction 15 2.3 Unsupervised learnin 15 2.4 Solving large scale problems with fixed size Ls-svm 3 LS-SVMlab toolbox examples 17 I roadmap tO LS-SVM 17 3. 2 Classificatio 17 3.2.1 Hello world 17 3.2.2 Example 19 Using the obiect oriented interface: initIssvi 3.2.4 LS-SVM classification: only one command line away! 21 3.2.5 Bavesian inference for classification 13.2.6 Multi-class coding 24 3. 3 Regression 25 simple example 25 33.2 SVM regression only one command line away 3.3.3 Bayesian Inference for Regression 3.3.4 Using the obiect oriented model interfac 29 3.5 Confidence/Predition Intervals for Regression 3.3.6 Robust rcercssion 33 33 M ultiple output regression 35 3.3.8 A time-series example e laser data prediction 36 3.3.9 Fixed size LS-SVM 37 3.4 Unsupervised learning using kernel principal component analysis 40 A MATLAB function 41 A 1 Gcncral notation LnlctIOll况 Training and simula tion 42 A.2.2 Obiect oriented interfac 23T d simulating functie 44 I functi 45 Tuning sparseness and robustnes 46 A 2.6 Cl 47 2.7B framework 48 CONTENTS A. 2. 8 NARX models and prediction 49 A 2.9 Unsupervised learnlin 50 A 2.10 Fixed size IS-SVM 51 A 2.11 demos 52 A 3 Alphabetical list of function calls A3, 2 bay-errorban 54 A 3.3 bay_initlssvml 56 A 3.4 bay lssvm .......57 A.3.5 bay-1ssymard 59 A. 3. 6 bay modoutclas 61 A.3.7 bay _optimize 3. 8 bav rr A 3.9 cilssvml 67 A 3.10 code, codelssvm 68 A 3.11 crossvalidatel 71 A..12 deltabl A.3.13d k 75 ossvalidat 76 A 3.16 initlssym, changelssvm A. 3. 17 kentrop A.3.18 kernel, matrix 81 A.3.19k 8 84 A 3.21 leaveoneout In-kernel, poly-kerne 上 kerne A 3.23 inf, mae. medae, misclass a mse A.3,241 90 91 dl 94 pre上ssvm stess A 3.30 rcrossvalidatel A.3.31 ridgeregress 98 A.3. 32 robust lssvm 99 A 3.33 roc 100 102 103 A 3.36 tunelssvm, linesearch gridsearch ....,,.,,,.105 A.3.37 windowize &z windowizeNARX 110 【实例截图】
【核心代码】