Added the svm_one_class_trainer object.

--HG--
extra : convert_revision : svn%3Afdd8eb12-d10e-0410-9acb-85c331704f74/trunk%404013

dlib/svm/svm_one_class_trainer.h

+// Copyright (C) 2010  Davis E. King (davis@dlib.net)
+// License: Boost Software License   See LICENSE.txt for the full license.
+#ifndef DLIB_SVm_ONE_CLASS_TRAINER_H__ 
+#define DLIB_SVm_ONE_CLASS_TRAINER_H__
+
+#include "svm_one_class_trainer_abstract.h"
+#include <cmath>
+#include <limits>
+#include <sstream>
+#include "../matrix.h"
+#include "../algs.h"
+
+#include "function.h"
+#include "kernel.h"
+#include "../optimization/optimization_solve_qp3_using_smo.h"
+
+namespace dlib 
+{
+
+// ----------------------------------------------------------------------------------------
+
+    template <
+        typename K 
+        >
+    class svm_one_class_trainer
+    {
+    public:
+        typedef K kernel_type;
+        typedef typename kernel_type::scalar_type scalar_type;
+        typedef typename kernel_type::sample_type sample_type;
+        typedef typename kernel_type::mem_manager_type mem_manager_type;
+        typedef decision_function<kernel_type> trained_function_type;
+
+        svm_one_class_trainer (
+        ) :
+            nu(0.1),
+            cache_size(200),
+            eps(0.001)
+        {
+        }
+
+        svm_one_class_trainer (
+            const kernel_type& kernel_, 
+            const scalar_type& nu_
+        ) :
+            kernel_function(kernel_),
+            nu(nu_),
+            cache_size(200),
+            eps(0.001)
+        {
+            // make sure requires clause is not broken
+            DLIB_ASSERT(0 < nu && nu <= 1,
+                "\tsvm_one_class_trainer::svm_one_class_trainer(kernel,nu)"
+                << "\n\t invalid inputs were given to this function"
+                << "\n\t nu: " << nu 
+                );
+        }
+
+        void set_cache_size (
+            long cache_size_
+        )
+        {
+            // make sure requires clause is not broken
+            DLIB_ASSERT(cache_size_ > 0,
+                "\tvoid svm_one_class_trainer::set_cache_size(cache_size_)"
+                << "\n\t invalid inputs were given to this function"
+                << "\n\t cache_size: " << cache_size_ 
+                );
+            cache_size = cache_size_;
+        }
+
+        long get_cache_size (
+        ) const
+        {
+            return cache_size;
+        }
+
+        void set_epsilon (
+            scalar_type eps_
+        )
+        {
+            // make sure requires clause is not broken
+            DLIB_ASSERT(eps_ > 0,
+                "\tvoid svm_one_class_trainer::set_epsilon(eps_)"
+                << "\n\t invalid inputs were given to this function"
+                << "\n\t eps: " << eps_ 
+                );
+            eps = eps_;
+        }
+
+        const scalar_type get_epsilon (
+        ) const
+        { 
+            return eps;
+        }
+
+        void set_kernel (
+            const kernel_type& k
+        )
+        {
+            kernel_function = k;
+        }
+
+        const kernel_type& get_kernel (
+        ) const
+        {
+            return kernel_function;
+        }
+
+        void set_nu (
+            scalar_type nu_
+        )
+        {
+            // make sure requires clause is not broken
+            DLIB_ASSERT(0 < nu_ && nu_ <= 1,
+                "\tvoid svm_one_class_trainer::set_nu(nu_)"
+                << "\n\t invalid inputs were given to this function"
+                << "\n\t nu: " << nu_ 
+                );
+            nu = nu_;
+        }
+
+        const scalar_type get_nu (
+        ) const
+        {
+            return nu;
+        }
+
+        template <
+            typename in_sample_vector_type
+            >
+        const decision_function<kernel_type> train (
+            const in_sample_vector_type& x
+        ) const
+        {
+            return do_train(vector_to_matrix(x));
+        }
+
+        void swap (
+            svm_one_class_trainer& item
+        )
+        {
+            exchange(kernel_function, item.kernel_function);
+            exchange(nu,              item.nu);
+            exchange(cache_size,      item.cache_size);
+            exchange(eps,             item.eps);
+        }
+
+    private:
+
+    // ------------------------------------------------------------------------------------
+
+        template <
+            typename in_sample_vector_type
+            >
+        const decision_function<kernel_type> do_train (
+            const in_sample_vector_type& x
+        ) const
+        {
+            typedef typename K::scalar_type scalar_type;
+            typedef typename decision_function<K>::sample_vector_type sample_vector_type;
+            typedef typename decision_function<K>::scalar_vector_type scalar_vector_type;
+
+            // make sure requires clause is not broken
+            DLIB_ASSERT(is_col_vector(x) && x.size() > 0,
+                "\tdecision_function svm_one_class_trainer::train(x)"
+                << "\n\t invalid inputs were given to this function"
+                << "\n\t x.nr(): " << x.nr() 
+                << "\n\t x.nc(): " << x.nc() 
+                );
+
+
+            scalar_vector_type alpha;
+
+            solve_qp3_using_smo<scalar_vector_type> solver;
+
+            solver(symmetric_matrix_cache<float>(kernel_matrix(kernel_function,x), cache_size), 
+                   zeros_matrix<scalar_type>(x.size(),1),
+                   ones_matrix<scalar_type>(x.size(),1), 
+                   nu*x.size(),
+                   1,
+                   1,
+                   alpha,
+                   eps);
+
+            scalar_type rho;
+            calculate_rho(alpha,solver.get_gradient(),rho);
+
+
+            // count the number of support vectors
+            const long sv_count = (long)sum(alpha != 0);
+
+            scalar_vector_type sv_alpha;
+            sample_vector_type support_vectors;
+
+            // size these column vectors so that they have an entry for each support vector
+            sv_alpha.set_size(sv_count);
+            support_vectors.set_size(sv_count);
+
+            // load the support vectors and their alpha values into these new column matrices
+            long idx = 0;
+            for (long i = 0; i < alpha.nr(); ++i)
+            {
+                if (alpha(i) != 0)
+                {
+                    sv_alpha(idx) = alpha(i);
+                    support_vectors(idx) = x(i);
+                    ++idx;
+                }
+            }
+
+            // now return the decision function
+            return decision_function<K> (sv_alpha, rho, kernel_function, support_vectors);
+        }
+
+    // ------------------------------------------------------------------------------------
+
+        template <
+            typename scalar_vector_type
+            >
+        void calculate_rho(
+            const scalar_vector_type& alpha,
+            const scalar_vector_type& df,
+            scalar_type& rho
+        ) const
+        {
+            using namespace std;
+            long num_p_free = 0;
+            scalar_type sum_p_free = 0;
+
+
+            scalar_type upper_bound_p;
+            scalar_type lower_bound_p;
+
+            find_min_and_max(df, upper_bound_p, lower_bound_p);
+
+            for(long i = 0; i < alpha.nr(); ++i)
+            {
+                if(alpha(i) == 1)
+                {
+                    if (df(i) > upper_bound_p)
+                        upper_bound_p = df(i);
+                }
+                else if(alpha(i) == 0)
+                {
+                    if (df(i) < lower_bound_p)
+                        lower_bound_p = df(i);
+                }
+                else
+                {
+                    ++num_p_free;
+                    sum_p_free += df(i);
+                }
+            }
+
+            scalar_type r1;
+            if(num_p_free > 0)
+                r1 = sum_p_free/num_p_free;
+            else
+                r1 = (upper_bound_p+lower_bound_p)/2;
+
+            rho = r1;
+        } 
+
+        kernel_type kernel_function;
+        scalar_type nu;
+        long cache_size;
+        scalar_type eps;
+    }; // end of class svm_one_class_trainer
+
+// ----------------------------------------------------------------------------------------
+
+    template <typename K>
+    void swap (
+        svm_one_class_trainer<K>& a,
+        svm_one_class_trainer<K>& b
+    ) { a.swap(b); }
+
+// ----------------------------------------------------------------------------------------
+
+}
+
+#endif // DLIB_SVm_ONE_CLASS_TRAINER_H__
+

dlib/svm/svm_one_class_trainer_abstract.h

+// Copyright (C) 2010  Davis E. King (davis@dlib.net)
+// License: Boost Software License   See LICENSE.txt for the full license.
+#undef DLIB_SVm_ONE_CLASS_TRAINER_ABSTRACT_
+#ifdef DLIB_SVm_ONE_CLASS_TRAINER_ABSTRACT_
+
+#include <cmath>
+#include <limits>
+#include "../matrix/matrix_abstract.h"
+#include "../algs.h"
+#include "function_abstract.h"
+#include "kernel_abstract.h"
+#include "../optimization/optimization_solve_qp3_using_smo_abstract.h"
+
+namespace dlib
+{
+
+// ----------------------------------------------------------------------------------------
+
+    template <
+        typename K 
+        >
+    class svm_one_class_trainer
+    {
+        /*!
+            REQUIREMENTS ON K 
+                is a kernel function object as defined in dlib/svm/kernel_abstract.h 
+
+            WHAT THIS OBJECT REPRESENTS
+                This object implements a trainer for a support vector machine for 
+                solving one-class classification problems.  It is implemented using the SMO
+                algorithm.
+
+                The implementation of the training algorithm used by this object is based
+                on the following excellent paper:
+                    - Chih-Chung Chang and Chih-Jen Lin, LIBSVM : a library for support vector 
+                      machines, 2001. Software available at http://www.csie.ntu.edu.tw/~cjlin/libsvm
+
+        !*/
+
+    public:
+        typedef K kernel_type;
+        typedef typename kernel_type::scalar_type scalar_type;
+        typedef typename kernel_type::sample_type sample_type;
+        typedef typename kernel_type::mem_manager_type mem_manager_type;
+        typedef decision_function<kernel_type> trained_function_type;
+
+        svm_one_class_trainer (
+        );
+        /*!
+            ensures
+                - This object is properly initialized and ready to be used
+                  to train a support vector machine.
+                - #get_nu() == 0.1 
+                - #get_cache_size() == 200
+                - #get_epsilon() == 0.001
+        !*/
+
+        svm_one_class_trainer (
+            const kernel_type& kernel, 
+            const scalar_type& nu
+        );
+        /*!
+            requires
+                - 0 < nu <= 1
+            ensures
+                - This object is properly initialized and ready to be used
+                  to train a support vector machine.
+                - #get_kernel() == kernel
+                - #get_nu() == nu
+                - #get_cache_size() == 200
+                - #get_epsilon() == 0.001
+        !*/
+
+        void set_cache_size (
+            long cache_size
+        );
+        /*!
+            requires
+                - cache_size > 0
+            ensures
+                - #get_cache_size() == cache_size 
+        !*/
+
+        const long get_cache_size (
+        ) const;
+        /*!
+            ensures
+                - returns the number of megabytes of cache this object will use
+                  when it performs training via the this->train() function.
+                  (bigger values of this may make training go faster but won't affect 
+                  the result.  However, too big a value will cause you to run out of 
+                  memory, obviously.)
+        !*/
+
+        void set_epsilon (
+            scalar_type eps
+        );
+        /*!
+            requires
+                - eps > 0
+            ensures
+                - #get_epsilon() == eps 
+        !*/
+
+        const scalar_type get_epsilon (
+        ) const;
+        /*!
+            ensures
+                - returns the error epsilon that determines when training should stop.
+                  Generally a good value for this is 0.001.  Smaller values may result
+                  in a more accurate solution but take longer to execute.
+        !*/
+
+        void set_kernel (
+            const kernel_type& k
+        );
+        /*!
+            ensures
+                - #get_kernel() == k 
+        !*/
+
+        const kernel_type& get_kernel (
+        ) const;
+        /*!
+            ensures
+                - returns a copy of the kernel function in use by this object
+        !*/
+
+        void set_nu (
+            scalar_type nu
+        );
+        /*!
+            requires
+                - 0 < nu <= 1
+            ensures
+                - #get_nu() == nu
+        !*/
+
+        const scalar_type get_nu (
+        ) const;
+        /*!
+            ensures
+                - returns the nu svm parameter.  This is a value between 0 and
+                  1.  It is the parameter that determines the trade off between
+                  trying to fit the training data exactly or allowing more errors 
+                  but hopefully improving the generalization ability of the 
+                  resulting classifier.  Smaller values encourage exact fitting 
+                  while larger values of nu may encourage better generalization. 
+                  For more information you should consult the papers referenced 
+                  above.
+        !*/
+
+        template <
+            typename in_sample_vector_type
+            >
+        const decision_function<kernel_type> train (
+            const in_sample_vector_type& x
+        ) const;
+        /*!
+            requires
+                - x.size() > 0
+                - is_col_vector(x) == true
+                - x == a matrix or something convertible to a matrix via vector_to_matrix().
+                  Also, x should contain sample_type objects.
+            ensures
+                - trains a one-class support vector classifier given the training samples in x.
+                  Training is done when the error is less than get_epsilon().
+                - returns a decision function F with the following properties:
+                    - if (new_x is a sample predicted to arise from the distribution
+                      which generated the training samples) then
+                        - F(new_x) >= 0
+                    - else
+                        - F(new_x) < 0
+        !*/
+
+        void swap (
+            svm_one_class_trainer& item
+        );
+        /*!
+            ensures
+                - swaps *this and item
+        !*/
+    }; 
+
+    template <typename K>
+    void swap (
+        svm_one_class_trainer<K>& a,
+        svm_one_class_trainer<K>& b
+    ) { a.swap(b); }
+    /*!
+        provides a global swap
+    !*/
+
+// ----------------------------------------------------------------------------------------
+
+}
+
+#endif // DLIB_SVm_ONE_CLASS_TRAINER_ABSTRACT_
+
+
+