-------------------------------------------- --- MATLAB/OCTAVE interface of LIBLINEAR --- -------------------------------------------- Table of Contents ================= - Introduction - Installation - Usage - Returned Model Structure - Other Utilities - Examples - Additional Information Introduction ============ This tool provides a simple interface to LIBLINEAR, a library for large-scale regularized linear classification and regression (http://www.csie.ntu.edu.tw/~cjlin/liblinear). It is very easy to use as the usage and the way of specifying parameters are the same as that of LIBLINEAR. Installation ============ On Windows systems, pre-built binary files are already in the directory '..\windows', so no need to conduct installation. Now we provide binary files only for 64bit MATLAB on Windows. If you would like to re-build the package, please rely on the following steps. We recommend using make.m on both MATLAB and OCTAVE. Just type 'make' to build 'libsvmread.mex', 'libsvmwrite.mex', 'train.mex', and 'predict.mex'. On MATLAB or Octave: >> make If make.m does not work on MATLAB (especially for Windows), try 'mex -setup' to choose a suitable compiler for mex. Make sure your compiler is accessible and workable. Then type 'make' to start the installation. Example: matlab>> mex -setup (ps: MATLAB will show the following messages to setup default compiler.) Please choose your compiler for building external interface (MEX) files: Would you like mex to locate installed compilers [y]/n? y Select a compiler: [1] Microsoft Visual C/C++ version 7.1 in C:\Program Files\Microsoft Visual Studio [0] None Compiler: 1 Please verify your choices: Compiler: Microsoft Visual C/C++ 7.1 Location: C:\Program Files\Microsoft Visual Studio Are these correct?([y]/n): y matlab>> make On Unix systems, if neither make.m nor 'mex -setup' works, please use Makefile and type 'make' in a command window. Note that we assume your MATLAB is installed in '/usr/local/matlab'. If not, please change MATLABDIR in Makefile. Example: linux> make To use octave, type 'make octave': Example: linux> make octave For a list of supported/compatible compilers for MATLAB, please check the following page: http://www.mathworks.com/support/compilers/current_release/ Usage ===== matlab> model = train(training_label_vector, training_instance_matrix [,'liblinear_options', 'col']); -training_label_vector: An m by 1 vector of training labels. (type must be double) -training_instance_matrix: An m by n matrix of m training instances with n features. It must be a sparse matrix. (type must be double) -liblinear_options: A string of training options in the same format as that of LIBLINEAR. -col: if 'col' is set, each column of training_instance_matrix is a data instance. Otherwise each row is a data instance. matlab> [predicted_label, accuracy, decision_values/prob_estimates] = predict(testing_label_vector, testing_instance_matrix, model [, 'liblinear_options', 'col']); matlab> [predicted_label] = predict(testing_label_vector, testing_instance_matrix, model [, 'liblinear_options', 'col']); -testing_label_vector: An m by 1 vector of prediction labels. If labels of test data are unknown, simply use any random values. (type must be double) -testing_instance_matrix: An m by n matrix of m testing instances with n features. It must be a sparse matrix. (type must be double) -model: The output of train. -liblinear_options: A string of testing options in the same format as that of LIBLINEAR. -col: if 'col' is set, each column of testing_instance_matrix is a data instance. Otherwise each row is a data instance. Returned Model Structure ======================== The 'train' function returns a model which can be used for future prediction. It is a structure and is organized as [Parameters, nr_class, nr_feature, bias, Label, w]: -Parameters: Parameters (now only solver type is provided) -nr_class: number of classes; = 2 for regression -nr_feature: number of features in training data (without including the bias term) -bias: If >= 0, we assume one additional feature is added to the end of each data instance. -Label: label of each class; empty for regression -w: a nr_w-by-n matrix for the weights, where n is nr_feature or nr_feature+1 depending on the existence of the bias term. nr_w is 1 if nr_class=2 and -s is not 4 (i.e., not multi-class svm by Crammer and Singer). It is nr_class otherwise. If the '-v' option is specified, cross validation is conducted and the returned model is just a scalar: cross-validation accuracy for classification and mean-squared error for regression. If the '-C' option is specified, best parameters are found by cross validation. The parameter selection utility is supported only by -s 0, -s 2 (for finding C) and -s 11 (for finding C, p). The returned model is a three dimensional vector with the best C, the best p, and the corresponding cross-validation accuracy or mean squared error. The returned best p for -s 0 and -s 2 is set to -1 because the p parameter is not used by classification models. Result of Prediction ==================== The function 'predict' has three outputs. The first one, predicted_label, is a vector of predicted labels. The second output, accuracy, is a vector including accuracy (for classification), mean squared error, and squared correlation coefficient (for regression). The third is a matrix containing decision values or probability estimates (if '-b 1' is specified). If k is the number of classes and k' is the number of classifiers (k'=1 if k=2, otherwise k'=k), for decision values, each row includes results of k' binary linear classifiers. For probabilities, each row contains k values indicating the probability that the testing instance is in each class. Note that the order of classes here is the same as 'Label' field in the model structure. Other Utilities =============== A matlab function libsvmread reads files in LIBSVM format: [label_vector, instance_matrix] = libsvmread('data.txt'); Two outputs are labels and instances, which can then be used as inputs of svmtrain or svmpredict. A matlab function libsvmwrite writes Matlab matrix to a file in LIBSVM format: libsvmwrite('data.txt', label_vector, instance_matrix] The instance_matrix must be a sparse matrix. (type must be double) For windows, `libsvmread.mexw64' and `libsvmwrite.mexw64' are ready in the directory `..\windows'. These codes are prepared by Rong-En Fan and Kai-Wei Chang from National Taiwan University. Examples ======== Train and test on the provided data heart_scale: matlab> [heart_scale_label, heart_scale_inst] = libsvmread('../heart_scale'); matlab> model = train(heart_scale_label, heart_scale_inst, '-c 1'); matlab> [predict_label, accuracy, dec_values] = predict(heart_scale_label, heart_scale_inst, model); % test the training data Note that for testing, you can put anything in the testing_label_vector. For probability estimates, you need '-b 1' only in the testing phase: matlab> [predict_label, accuracy, prob_estimates] = predict(heart_scale_label, heart_scale_inst, model, '-b 1'); Use the best parameter to train (C for -s 0, 2 and C, p for -s 11): matlab> best = train(heart_scale_label, heart_scale_inst, '-C -s 0'); matlab> model = train(heart_scale_label, heart_scale_inst, sprintf('-c %f -s 0', best(1))); % use the same solver: -s 0 Additional Information ====================== Please cite LIBLINEAR as follows R.-E. Fan, K.-W. Chang, C.-J. Hsieh, X.-R. Wang, and C.-J. Lin. LIBLINEAR: A Library for Large Linear Classification, Journal of Machine Learning Research 9(2008), 1871-1874.Software available at http://www.csie.ntu.edu.tw/~cjlin/liblinear For any question, please contact Chih-Jen Lin .