fated / libvm

LibVM -- A Library for Venn Machine

LibVM -- A Library for Venn Machine

LibVM is a simple, easy-to-use, and efficient software for Venn Machine on classification, which gives label prediction together with it's probabilistic estimations. This library solves Venn prediction in both online and offline mode with k-nearest neighbors or support vector machines as the underlying algorithms. This document explains the use of LibVM.

Table of Contents

• Installation and Data Format
• "vm-offline" Usage
• "vm-online" Usage
• "vm-cv" Usage
• Parameters for Underlying Algorithms
• Tips on Practical Use
• Examples
• Library Usage
• Additional Information
• Acknowledgments

Installation and Data Format↩

On Unix systems, type make to build the vm-offline, vm-online and vm-cv programs. Run them without arguments to show the usage of them.

The format of training and testing data file is:

<label> <index1>:<value1> <index2>:<value2> ...
...
...
...

Each line contains an instance and is ended by a '\n' character (Unix line ending). For classification, <label> is an integer indicating the class label (multi-class is supported). For regression, <label> is the target value which can be any real number. The pair <index>:<value> gives a feature (attribute) value: <index> is an integer starting from 1 and <value> is the value of the attribute, which could be an integer number or real number. Indices must be in ASCENDING order. Labels in the testing file are only used to calculate accuracies and errors. If they are unknown, just fill the first column with any numbers.

A sample classification data set included in this package is iris_scale for training and iris_scale_t for testing.

Type vm-offline iris_scale iris_scale_t, and the program will read the training data and testing data and then output the result into iris_scale_t_output file by default. The model file iris_scale_model will not be saved by default, however, adding -s model_file_name to [option] will save the model to model_file_name. The output file contains the predicted labels and the lower and upper bounds of probabilities for each predicted label.

"vm-offline" Usage↩

Usage: vm-offline [options] train_file test_file [output_file]
options:
  -t taxonomy_type : set type of taxonomy (default 0)
    0 -- k-nearest neighbors (KNN)
    1 -- support vector machine with equal length (SVM_EL)
    2 -- support vector machine with equal size (SVM_ES)
    3 -- support vector machine with k-means clustering (SVM_KM)
    4 -- one-vs-all support vector machine (OVA_SVM)
    5 -- Crammer and Singer's multi-class support vector machine (MCSVM)
    6 -- Crammer and Singer's multi-class support vector machine with equal length (MCSVM_EL)
  -k num_neighbors : set number of neighbors in kNN (default 1)
  -c num_categories : set number of categories for Venn predictor (default 4)
  -s model_file_name : save model
  -l model_file_name : load model
  -b probability estimates : whether to output probability estimates for all labels, 0 or 1 (default 0)
  -q : quiet mode (no outputs)

train_file is the data you want to train with.
test_file is the data you want to predict.
vm-offline will produce outputs in the output_file by default.

"vm-online" Usage↩

Usage: vm-online [options] data_file [output_file]
options:
  -t taxonomy_type : set type of taxonomy (default 0)
    0 -- k-nearest neighbors (KNN)
    1 -- support vector machine with equal length (SVM_EL)
    2 -- support vector machine with equal size (SVM_ES)
    3 -- support vector machine with k-means clustering (SVM_KM)
    4 -- one-vs-all support vector machine (OVA_SVM)
    5 -- Crammer and Singer's multi-class support vector machine (MCSVM)
    6 -- Crammer and Singer's multi-class support vector machine with equal length (MCSVM_EL)
  -k num_neighbors : set number of neighbors in kNN (default 1)
  -c num_categories : set number of categories for Venn predictor (default 4)
  -q : turn off quiet mode (no outputs)

data_file is the data you want to run the online prediction on.
vm-online will produce outputs in the output_file by default.

"vm-cv" Usage↩

Usage: vm-cv [options] data_file [output_file]
options:
  -t taxonomy_type : set type of taxonomy (default 0)
    0 -- k-nearest neighbors (KNN)
    1 -- support vector machine with equal length (SVM_EL)
    2 -- support vector machine with equal size (SVM_ES)
    3 -- support vector machine with k-means clustering (SVM_KM)
    4 -- one-vs-all support vector machine (OVA_SVM)
    5 -- Crammer and Singer's multi-class support vector machine (MCSVM)
    6 -- Crammer and Singer's multi-class support vector machine with equal length (MCSVM_EL)
  -k num_neighbors : set number of neighbors in kNN (default 1)
  -c num_categories : set number of categories for Venn predictor (default 4)
  -v num_folds : set number of folders in cross validation (default 5)
  -q : turn off quiet mode (no outputs)

data_file is the data you want to run the cross validation on.
vm-cv will produce outputs in the output_file by default.

Parameters for Underlying Algorithms↩

-p : prefix of options to set parameters for SVM
  -ps svm_type : set type of SVM (default 0)
    0 -- C-SVC    (multi-class classification)
    1 -- nu-SVC   (multi-class classification)
    2 -- OVA-SVC  (multi-class classification)
  -pt kernel_type : set type of kernel function (default 2)
    0 -- linear: u'*v
    1 -- polynomial: (gamma*u'*v + coef0)^degree
    2 -- radial basis function: exp(-gamma*|u-v|^2)
    3 -- sigmoid: tanh(gamma*u'*v + coef0)
    4 -- precomputed kernel (kernel values in training_set_file)
  -pd degree : set degree in kernel function (default 3)
  -pg gamma : set gamma in kernel function (default 1/num_features)
  -pr coef0 : set coef0 in kernel function (default 0)
  -pc cost : set the parameter C of C-SVC (default 1)
  -pn nu : set the parameter nu of nu-SVC (default 0.5)
  -pm cachesize : set cache memory size in MB (default 100)
  -pe epsilon : set tolerance of termination criterion (default 0.001)
  -ph shrinking : whether to use the shrinking heuristics, 0 or 1 (default 1)
  -pwi weights : set the parameter C of class i to weight*C, for C-SVC (default 1)
-m : prefix of options to set parameters for MCSVM
  -ms redopt_type : set type of reduced optimization (default 0)
    0 -- exact (EXACT)
    1 -- approximate (APPROX)
    2 -- binary (BINARY)
  -mt kernel_type : set type of kernel function (default 2)
    0 -- linear: u'*v
    1 -- polynomial: (gamma*u'*v + coef0)^degree
    2 -- radial basis function: exp(-gamma*|u-v|^2)
    3 -- sigmoid: tanh(gamma*u'*v + coef0)
    4 -- precomputed kernel (kernel values in training_set_file)
  -md degree : set degree in kernel function (default 3)
  -mg gamma : set gamma in kernel function (default 1/num_features)
  -mr coef0 : set coef0 in kernel function (default 0)
  -mb beta : set margin (default 1e-4)
  -mw delta : set approximation tolerance for approximate method (default 1e-4)
  -mm cachesize : set cache memory size in MB (default 100)
  -me epsilon : set tolerance of termination criterion (default 1e-3)
  -mz epsilon0 : set initialize margin (default 1-1e-6)

use different prefix for different parameters of underlying algorithms.

Tips on Practical Use↩

• Scale your data. For example, scale each attribute to [0,1] or [-1,+1].
• Try different taxonomies. Some data sets will not achieve good results on some data sets.
• Change parameters for better results especially when you are using SVM related taxonomies.

Examples↩

> vm-offline -k 3 train_file test_file output_file

Train a venn predictor with 3-nearest neighbors as underlying algorithm from train_file. Then conduct this classifier to test_file and output the results to output_file.

> vm-offline -t 1 -s model_file train_file test_file

Train a venn predictor using support vector machines with equal length intervals as taxonomy from train_file. Then conduct this classifier to test_file and output the results to the default output file, also the model will be saved to file model_file.

> vm-online -t 2 data_file

Train an online venn predictor classifier using support vector machine with equal size intervals as taxonomy from data_file. Then output the results to the default output file.

> vm-cv -t 3 -v 10 data_file

Do a 10-fold cross validation venn predictor using support vector machine with k-means clustering intervals as taxonomy from data_file. Then output the results to the default output file.

Library Usage↩

All functions and structures are declared in different header files. There are 7 parts in this library, which are utilities, knn, kernel, svm, mcsvm, vm and the other driver programs.

utilities.h and utilities.cpp

The structure Problem for storing the data sets (including the structure Node for storing the attributes pair of index and value) and all the constant variables are declared in utilities.h.

In this file, some utilizable function templates or functions are also declared.

• void PrintCout(const char *s)
• void PrintNull(const char *s)
• void Info(const char *format, ...)
• void SetPrintNull()
• void SetPrintCout()
  This 5 functions are related to the printing of intermediate process. The contents printing by Info(...) will be redirected to output stream cout or empty stream null. SetPrintNull() will print the output to nowhere (except the warning and error messages and the final results). SetPrintCout() will print the output to the standard output stream.
• T FindMostFrequent(T *array, int size)
  This function is used to find the most frequent category in _k_NN taxonomy.
• static inline void clone(T *&dest, S *src, int size)
  This static function is used to clone an array from src to dest.
• void QuickSortIndex(T array[], size_t index[], size_t left, size_t right)
  This function is used to quicksort an array and preserve the original indices.
• Problem *ReadProblem(const char *file_name)
  This function is used to read in a data set from a file named file_name.
• void FreeProblem(struct Problem *problem)
  This function is used to free a problem stored in the memory.
• void GroupClasses(const Problem *prob, int *num_classes_ret, int **labels_ret, int **start_ret, int **count_ret, int *perm)
  This function is used in Cross Validation and other predictions using SVM related taxonomies. This function will group the examples with same label together. The last 5 parameters are using to return corresponding values. num_classes_ret is used to store the number of classes in the problem. labels_ret is an array used to store the actual label in the order of appearance. start_ret is an array used to store the starting index of each group of examples. count_ret is an array used to store the count number of each group of examples. perm is an array used to store the permutation of the permuted index of the problem.
• int *GetLabels(const Problem *prob, int *num_classes_ret) This function is used to get label list of prob. The label list will store in an integer array as the return value, and the number of classes num_classes_ret will also be returned.

knn.h and knn.cpp

The structure KNNParameter for storing the _k_NN related parameters and the structure KNNModel for storing the _k_NN related model are declared in knn.h.

In this file, some utilizable function templates or functions are also declared.

• static inline void InsertLabel(T *labels, T label, int num_neighbors, int index)
  This static function will insert label into the index-th location of the array labels of which the size is num_neighbors.
• KNNModel *TrainKNN(const struct Problem *prob, const struct KNNParameter *param)
  This function is used to train a _k_NN model from a problem prob and the parameter param, it will return a model of the structure KNNModel.
• double PredictKNN(struct Problem *train, struct Node *x, const int num_neighbors)
  This function is used to predict the label for object x using _k_NN classifier.
• double CalcDist(const struct Node *x1, const struct Node *x2)
  This function is used to calculate the distance between two objects x1 and x2, which will be used in _k_NN.
• int CompareDist(double *neighbors, double dist, int num_neighbors)
  This function is used to compare a distance dist with the nearest neighbors' distances stored in an array neighbors, it will return the position of dist, if it is greater than all the distances in neighbors, it gives num_neighbors.
• int SaveKNNModel(std::ofstream &model_file, const struct KNNModel *model)
• KNNModel *LoadKNNModel(std::ifstream &model_file)
• void FreeKNNModel(struct KNNModel *model)
  These three functions are used to manipulate the _k_NN model file, including "save to file", "load from file" and "free the model".
• void FreeKNNParam(struct KNNParameter *param)
• void InitKNNParam(struct KNNParameter *param)
• const char *CheckKNNParameter(const struct KNNParameter *param)
  These three functions are used to manipulate the _k_NN parameter file, including "free the param", "initial the param" and "check the param".

kernel.h and kernel.cpp

The structure KernelParameter for storing kernel related parameters and the class Cache, QMatrix and Kernel for storing kernel related model are declared in knn.h.

In this file, some utilizable function templates or functions are also declared.

• static double KernelFunction(const Node *x, const Node *y, const KernelParameter *param)
  This static method in class Kernel is used to doing single kernel evaluation.
• void InitKernelParam(struct KernelParameter *param)
• const char *CheckKernelParameter(const struct KernelParameter *param)
  These two functions are used to manipulate the KernelParameter variable, including "initial the param" and "check the param". We don't have a function for free structure KernelParameter, since we don't allocate memery blocks for the parameter.

svm.h and svm.cpp

The structure SVMParameter for storing the SVM related parameters and the structure SVMModel for storing the SVM related model are declared in svm.h.

In this file, some utilizable function templates or functions are also declared.

• SVMModel *TrainSVM(const struct Problem *prob, const struct SVMParameter *param)
  This function is used to train a SVM model from a problem prob and the parameter param, it will return a model of the structure SVMModel.
• double PredictSVMValues(const struct SVMModel *model, const struct Node *x, double* decision_values)
  This function is used to predict the label for object x using SVM classifier. The decision values for object x will be returned in decision_values.
• double PredictSVM(const struct SVMModel *model, const struct Node *x)
  This function is an interface for PredictSVMValues() to predict label.
• int SaveSVMModel(std::ofstream &model_file, const struct SVMModel *model)
• SVMModel *LoadSVMModel(std::ifstream &model_file)
• void FreeSVMModel(struct SVMModel **model)
  These three functions are used to manipulate the SVM model file, including "save to file", "load from file" and "free the model".
• void FreeSVMParam(struct SVMParameter *param)
• void InitSVMParam(struct SVMParameter *param)
• const char *CheckSVMParameter(const struct SVMParameter *param)
  These three functions are used to manipulate the SVM parameter file, including "free the param", "initial the param" and "check the param".

mcsvm.h and mcsvm.cpp

The structure MCSVMParameter for storing the MCSVM related parameters and the structure MCSVMModel for storing the MCSVM related model are declared in mcsvm.h.

In this file, some utilizable function templates or functions are also declared.

• MCSVMModel *TrainMCSVM(const struct Problem *prob, const struct MCSVMParameter *param)
  This function is used to train a MCSVM model from a problem prob and the parameter param, it will return a model of the structure MCSVMModel.
• double *PredictMCSVMValues(const struct MCSVMModel *model, const struct Node *x)
  This function is used to get all similarity score for object x using MCSVM classifier. The similarity score will be returned as a double array.
• int PredictMCSVM(const struct MCSVMModel *model, const struct Node *x, int *num_max_sim_score_ret)
  This function is an interface for PredictMCSVMValues() to predict label. The varible num_max_sim_score_ret will be used to store the number of maximal similarity score which will be used to detect errors.
• double PredictMCSVMMaxValue(const struct MCSVMModel *model, const struct Node *x)
  This function is an interface for PredictMCSVMValues() to predict the largest similarity score.
• int SaveMCSVMModel(std::ofstream &model_file, const struct MCSVMModel *model)
• MCSVMModel *LoadMCSVMModel(std::ifstream &model_file)
• void FreeMCSVMModel(struct MCSVMModel *model)
  These three functions are used to manipulate the MCSVM model file, including "save to file", "load from file" and "free the model".
• void FreeMCSVMParam(struct MCSVMParameter *param)
• void InitMCSVMParam(struct MCSVMParameter *param)
• const char *CheckMCSVMParameter(const struct MCSVMParameter *param)
  These three functions are used to manipulate the MCSVM parameter file, including "free the param", "initial the param" and "check the param".

vm.h and vm.cpp

The structure Parameter for storing the Venn Machine related parameters and the structure Model for storing the Venn Machine related model are declared in vm.h. You need to #include "vm.h" in your C/C++ source files and link your program with vm.cpp. You can see vm-offline.cpp, vm-online.cpp and vm-cv.cpp for examples showing how to use them.

In this file, some utilizable function templates or functions are also declared.

• Model *TrainVM(const struct Problem *train, const struct Parameter *param) This function is used to train a venn predictor from the problem train and the parameter param.
• double PredictVM(const struct Problem *train, const struct Model *model, const struct Node *x, double &lower, double &upper, double **avg_prob)
  This function is used to predict a new object x from the problem train and the model. It will return the predicted label, lower for lower bound of the probability, upper for upper bound and avg_prob for calculate performance measures are also returned.
• void CrossValidation(const struct Problem *prob, const struct Parameter *param, double *predict_labels, double *lower_bounds, double *upper_bounds, double *brier, double *logloss)
  This function is used to do a cross validation on the problem prob and the parameter param. The other 5 parameters are used to return the corresponding values.
• void OnlinePredict(const struct Problem *prob, const struct Parameter *param, double *predict_labels, int *indices, double *lower_bounds, double *upper_bounds, double *brier, double *logloss)
  This function is used to do a online prediction on the problem prob and the parameter param. The other 6 parameters are used to return the corresponding values.
• int SaveModel(const char *model_file_name, const struct Model *model)
• Model *LoadModel(const char *model_file_name)
• void FreeModel(struct Model *model)
  These three functions are used to manipulate the model file, including "save to file", "load from file" and "free the model".
• void FreeParam(struct Parameter *param)
• const char *CheckParameter(const struct Parameter *param)
  These two functions are used to manipulate the parameter file, including "free the param" and "check the param".

vm-offline.cpp, vm-online.cpp and vm-cv.cpp

These three files are the driver programs for LibVM. vm-offline.cpp is for training and testing data sets in offline setting. vm-online.cpp is for doing online prediction on data sets. vm-cv.cpp is for doing cross validation on data sets.

The structure of these files are similar. In these programs, the command-line inputs will be parsed, the data sets will be read into the memory, the train and predict process will be called, the performance measure process will be carried out and finally the memories it claimed will be cleaned up. It includes the following functions.

• void ExitWithHelp()
  This function is used to print out the usage of the executable file.
• void ParseCommandLine(int argc, char *argv[], ...)
  This function is used to parse the options from the command-line input, and return the values like file names to the other parameters which is represented by ....

Additional Information↩

For any questions and comments, please email [e-mail]

Acknowledgments↩

Special thanks to Chih-Chung Chang and Chih-Jen Lin, which are the authors of LibSVM.