libtinysvm Reference Manual

Last update: $Date: 2001/09/04 16:10:56 $

List of Contents

• Class Reference Manual
  • TinySVM::Param class
  • TinySVM::BaseExample class
  • TinySVM::Example class
  • TinySVM::Model class
• Sample Programming
• How to compile

Class Reference Manual

namespace TinySVM {
   class Param
   {
    public:
     int set (int,  char **);
     int set (const char *);
   };
};

TinySVM::Param::set (int argc, char **argv)

Set parameter with argc, argv style, --- arguments for main() function.
This function accepts same parameter used for svm_learn.

TinySVM::Param::set (const char *s)

Set parameter with string s, e.g."-t 1 -d 3 -c 1".
This function accepts same parameter used for svm_learn.

namespace TinySVM {
   class BaseExample
   {
   public:
     int add (const double, feature_node *);
     int add (const double, const char *);
     int add (const char *);
     int set (int i, const double, feature_node *);
     int set (int i, const double, const char *);
     int set (int i, const char *);
     int remove (int i);
     int clear ();
     int size ();

     virtual int read  (const char *, const char *mode = "r", const int offset = 0) = 0;
     virtual int write (const char *, const char *mode = "w", const int offset = 0) = 0;
     int readSVindex  (const char *, const char *mode = "r", const int offset = 0);
     int writeSVindex (const char *, const char *mode = "w", const int offset = 0);
   };
};

TinySVM::BaseExample is a base class for TinySVM::Example and TinySVM::Model. You need not use this class directly.

TinySVM::BaseExample::add (const double y, feature_node *x)

Add new example y and x into current instance.
Return 1 if success, return 0 if fail.

TinySVM::BaseExample::add (const double y, char *x)

Add new example y and x into current instance. x is represented with a string , e.g. "1:2 3:4 5:6".
Return 1 if success, return 0 if fail.

TinySVM::BaseExample::add (char *s)

Add new example represented with single string, e.g. "+1 1:2 3:4 5:1" into current instance.
Return 1 if success, return 0 if fail.

TinySVM::BaseExample::set(int i const double y, feature_node *x)

Replace current i-th example with new example y and x.
Return 1 if success, return 0 if fail.

TinySVM::BaseExample::set (int i const double y, char *x)

Replace current i-th example with new example y and x .
x is represented with a string, (e.g. "1:2 3:4 5:6").
Return 1 if success, return 0 if fail.

TinySVM::BaseExample::set (int i char *s)

Replace current i-th example with new example represented with single string s, (e.g. "+1 1:2 3:4 5:1").
Return 1 if success, return 0 if fail.

TinySVM::BaseExample::remove (int i)

Remove i-th example (pair of y and x) from current instance.

TinySVM::BaseExample::clear ()

Clear all examples (pair of y and x) stored in current instance.

TinySVM::BaseExample::size ()

Get the size of examples (pair of y and x).

TinySVM::BaseExample::read (const char *file_name, const char *mode ="r", const int offset = 0)

Read all examples from file_name. mode is used for calling fopen (3).
if offset is given, seek offset byte using fseek (3) before reading examples.
Return 1 if success, return 0 if fail.

TinySVM::BaseExample::write (const char *file_name, const char *mode ="w", const int offset = 0)

Save all examples into file_name. mode is used for calling fopen(3).
By setting "a" to mode, you can append examples into file_name.
offset is reserved argument for the future.
Return 1 if success, return 0 if fail.

TinySVM::BaseExample::writeSVindex (const char *file_name const char *mode = "w", const int offset = 0)

Read pairs of alpha and gradiant from file_name. mode is used for calling fopen (3).
if offset is given, seek offset byte using fseek (3) before reading examples.
Return 1 if success, return 0 if fail.

TinySVM::BaseExample::readSVindex (const char *file_name const char *mode = "w", const int offset = 0)

Save pairs of alpha and gradiant into file_name. mode is used for calling fopen (3).
if offset is given, seek offset byte using fseek (3) before reading examples.
Return 1 if success, return 0 if fail.

namespace TinySVM {
   class Example: public BaseExample
   {
   public:
     Model *learn (const Param &);
     Model *learn (const char *);
     int rebuildSVindex (Model *);
   };
};

TinySVM::Example class inherits it's interface from TinySVM::BaseExample class. You can use all member functions defined in TinySVM::BaseExample class.

TinySVM::Example::learn (const TinySVM::Param &p)

Execute learning with TinySVM::Param object p.
Return new instance (pointer) of TinySVM::Model object.

TinySVM::Example::learn (const char *s)

Execute learning with Parameter represented with single string s, e.g. "-t 1 -d 2 -c 1".
Return new instance (pointer) of TinySVM::Model object.

TinySVM::Example::rebuldSVindex (const TinySVM::Model *m)

Rebuild alpha and gradient for all examples using Model m.

namespace TinySVM {
class Model: public BaseExample
   {
    public:
     Model ();
     Model (const Param &);

     double classify (const char *);
     double classify (const feature_node *);
     double estimateMargin ();
     double estimateVC ();
     double estimateXA (const double rho = 2.0);
     double estimateSphere ();
     double getLoss ();
     int compress (); 

     int getSVnum ();
     int getBSVnum();
     int getTrainingDataSize();
   };
};

TinySVM::Model class inherits it's interface from TinySVM::BaseExample class. You can use all member functions defined in TinySVM::BaseExample class.

TinySVM::Model::Model (const TinySVM::Param &p)

Create new TinySVM::Model instance with TinySVM::Param object p.

TinySVM::Model::classify (const char *s)

Classify test example represented with single string s, e.g. "1:1 2:1 3:1".
Return distance from the decision hyperplane.

TinySVM::Model::classify (const feature_node *f)

Classify test example represented with feature_node "f".
Return distance from the decision hyper plane.

TinySVM::Model::estimateMargin ()

Return the estimated margin value

TinySVM::Model::estimateVC ()

Return the estimated VC dimension. This value is roughly approximated.

TinySVM::Model::estimateXA(const double rho = 2.0 )

Return the Xi-Alpha estimation value, which is known as a tighter bound for Leave-One-Out bound, with parameter rho.

TinySVM::Model::estimateSphere ()

Return the minimum diameter of sphere which encloses the all examples stored in current model. This value is roughly approximated.

TinySVM::Model::getSVnum ()

Return the number of Support Vectors.

TinySVM::Model::getBSVnum ()

Return the number of bounded Support Vectors, -- the number of examples whose alpha is C.

TinySVM::Model::getTrainingDataSize ()

Return the size of training data.

TinySVM::Model::getLoss ()

Return the 1-norm loss evaluated with training data. This value is useally mentioned as "Empirical Risk". The case of C-SVR, this function returns the "1-norm epsilon-insenstive loss" evaluated with given training data.

TinySVM::Model::compress ()

Replace the current representation of model instance (lists of alpha and training data) with single vector w (w * x + b).

Sample Programming

• Learn with file-name, and store the result into model-file.

  #include <tinysvm.h>
  
     TinySVM::Example e;
     TinySVM::Param p;
  
     if (! e.read(file-name)) throw;
     if (! p.set("-t 1 -d 2 -c 1")) throw;
  
     TinySVM::Model *m = e.learn(p);
  
     m->write(model-file);
  
     delete m;
  

• Classify using stored model-file

  #include <tinysvm.h>
  
     TinySVM::Model m;
  
     if (! m.read(model-file)) throw;
  
     cout << m.classify("1:1 2:1 3:1 4:1") << endl;
  
  

• svm_learn, svm_classify and svm_model, all of these included program are written with libtinysvm.
  Tese sources would also help you to build your own programs.

How to compile

c++ -I/usr/local/include -L/usr/local/lib sample.cc -o sample -ltinysvm

$Id: libtinysvm.html,v 1.11 2001/09/04 16:10:56 taku-ku Exp $;