Home

Welcome to the svm-embed wiki!

Introduction

svm_embed is a support vector machine,especially for embedded computing. svm_embed is implemented in pure C,so it is very convenient to use in your projects.

For sake of the computing capabilities(embedded system),I just implement the basic properties of SVM:

Two labels come into considered Just for Linear kernel(I will explain the reasons in the wiki page) Automatically produce the exact file,which you can directly use in you project.The file includes the learned model(precisely,a constant float array,a float bias,and a predicting function)

Usage

When you "make" the project,simple command line "./main -f data". Here "data" is your data sets file. After you run the main.exe,there will be an additional file"model.h",which includes a function ,named "predict".

Format

Data sets file. The format is Label:value1:velue2:....:valuen

Attention: every line end up with a enter symbol,and the last line does not have.

Useful Scene

When you design a tinny embedded system,probably you will use a micro-controller say MSP430,C51...The system source is very limited. And you want to use it to do some simple classification. Our svm-embed is a good choice.After you do the training procedure,the only thing you will do is copy the model.h file to your project and use the "predic" function in your program. The variable,in the "predic" function,we use the keywords "const" to program it in the flash or rom. So it will do add little burden to your system.

PS: I will be highly appreciated if you can give me some wise advice.

email to me:[email protected]