PLearn 0.1
Public Member Functions | Static Public Member Functions | Public Attributes | Static Public Attributes | Static Protected Member Functions | Protected Attributes | Private Types | Private Member Functions
PLearn::SummationKernel Class Reference

Kernel computing the sum of other kernels. More...

#include <SummationKernel.h>

Inheritance diagram for PLearn::SummationKernel:
Inheritance graph
[legend]
Collaboration diagram for PLearn::SummationKernel:
Collaboration graph
[legend]

List of all members.

Public Member Functions

 SummationKernel ()
 Default constructor.
virtual void setDataForKernelMatrix (VMat the_data)
 Distribute to terms (sub-kernels) in the summation, subsetting if required.
virtual void addDataForKernelMatrix (const Vec &newRow)
 Distribute to terms (sub-kernels) in the summation, subsetting if required.
virtual real evaluate (const Vec &x1, const Vec &x2) const
 Compute K(x1,x2).
virtual real evaluate_i_x (int i, const Vec &x, real) const
 Evaluate a test example x against a train example given by its index.
virtual void evaluate_all_i_x (const Vec &x, const Vec &k_xi_x, real squared_norm_of_x=-1, int istart=0) const
 Fill k_xi_x with K(x_i, x), for all i from istart to istart + k_xi_x.length() - 1.
virtual void computeGramMatrix (Mat K) const
 Compute the Gram Matrix by calling subkernels computeGramMatrix.
virtual void computeGramMatrixDerivative (Mat &KD, const string &kernel_param, real epsilon=1e-6) const
 Directly compute the derivative with respect to hyperparameters (Faster than finite differences...)
virtual string classname () const
virtual OptionListgetOptionList () const
virtual OptionMapgetOptionMap () const
virtual RemoteMethodMapgetRemoteMethodMap () const
virtual SummationKerneldeepCopy (CopiesMap &copies) const
virtual void build ()
 Post-constructor.
virtual void makeDeepCopyFromShallowCopy (CopiesMap &copies)
 Transforms a shallow copy into a deep copy.

Static Public Member Functions

static string _classname_ ()
static OptionList_getOptionList_ ()
static RemoteMethodMap_getRemoteMethodMap_ ()
static Object_new_instance_for_typemap_ ()
static bool _isa_ (const Object *o)
static void _static_initialize_ ()
static const PPathdeclaringFile ()

Public Attributes

TVec< Kerm_terms
 Individual kernels to add to produce the final result.
TVec< TVec< int > > m_input_indexes
 Optionally, one can specify which of individual input variables should be routed to each kernel.

Static Public Attributes

static StaticInitializer _static_initializer_

Static Protected Member Functions

static void declareOptions (OptionList &ol)
 Declares the class options.

Protected Attributes

TVec< Vecm_input_buf1
 Input buffers for kernel evaluation in cases where subsetting is needed.
TVec< Vecm_input_buf2
Vec m_eval_buf
 Temporary buffer for kernel evaluation on all training dataset.
Mat m_gram_buf
 Temporary buffer for Gram matrix accumulation.

Private Types

typedef Kernel inherited

Private Member Functions

void build_ ()
 This does the actual building.

Detailed Description

Kernel computing the sum of other kernels.

This kernel computes the summation of several subkernel objects. It can also chop up parts of its input vector and send it to each kernel (so that each kernel can operate on a subset of the variables).

Definition at line 54 of file SummationKernel.h.


Member Typedef Documentation

Reimplemented from PLearn::Kernel.

Definition at line 56 of file SummationKernel.h.


Constructor & Destructor Documentation

PLearn::SummationKernel::SummationKernel ( )

Default constructor.

Definition at line 58 of file SummationKernel.cc.

{ }

Member Function Documentation

string PLearn::SummationKernel::_classname_ ( ) [static]

Reimplemented from PLearn::Kernel.

Definition at line 53 of file SummationKernel.cc.

OptionList & PLearn::SummationKernel::_getOptionList_ ( ) [static]

Reimplemented from PLearn::Kernel.

Definition at line 53 of file SummationKernel.cc.

RemoteMethodMap & PLearn::SummationKernel::_getRemoteMethodMap_ ( ) [static]

Reimplemented from PLearn::Kernel.

Definition at line 53 of file SummationKernel.cc.

bool PLearn::SummationKernel::_isa_ ( const Object o) [static]

Reimplemented from PLearn::Kernel.

Definition at line 53 of file SummationKernel.cc.

Object * PLearn::SummationKernel::_new_instance_for_typemap_ ( ) [static]

Reimplemented from PLearn::Object.

Definition at line 53 of file SummationKernel.cc.

StaticInitializer SummationKernel::_static_initializer_ & PLearn::SummationKernel::_static_initialize_ ( ) [static]

Reimplemented from PLearn::Kernel.

Definition at line 53 of file SummationKernel.cc.

void PLearn::SummationKernel::addDataForKernelMatrix ( const Vec newRow) [virtual]

Distribute to terms (sub-kernels) in the summation, subsetting if required.

Reimplemented from PLearn::Kernel.

Definition at line 145 of file SummationKernel.cc.

References i, n, and PLearn::selectElements().

{
    inherited::addDataForKernelMatrix(newRow);
    bool split_inputs = m_input_indexes.size() > 0;
    for (int i=0, n=m_terms.size() ; i<n ; ++i) {
        if (split_inputs && m_input_indexes[i].size() > 0) {
            selectElements(newRow, m_input_indexes[i], m_input_buf1[i]);
            m_terms[i]->addDataForKernelMatrix(m_input_buf1[i]);
        }
        else
            m_terms[i]->addDataForKernelMatrix(newRow);
    }
}

Here is the call graph for this function:

void PLearn::SummationKernel::build ( ) [virtual]

Post-constructor.

The normal implementation should call simply inherited::build(), then this class's build_(). This method should be callable again at later times, after modifying some option fields to change the "architecture" of the object.

Reimplemented from PLearn::Kernel.

Definition at line 89 of file SummationKernel.cc.

{
    // ### Nothing to add here, simply calls build_
    inherited::build();
    build_();
}
void PLearn::SummationKernel::build_ ( ) [private]

This does the actual building.

Reimplemented from PLearn::Kernel.

Definition at line 99 of file SummationKernel.cc.

References i, n, N, and PLERROR.

{
    // Preallocate buffers for kernel evaluation
    const int N = m_input_indexes.size();
    m_input_buf1.resize(N);
    m_input_buf2.resize(N);
    for (int i=0 ; i<N ; ++i) {
        const int M = m_input_indexes[i].size();
        m_input_buf1[i].resize(M);
        m_input_buf2[i].resize(M);
    }

    // Kernel is symmetric only if all terms are
    is_symmetric = true;
    for (int i=0, n=m_terms.size() ; i<n ; ++i) {
        if (! m_terms[i])
            PLERROR("SummationKernel::build_: kernel for term[%d] is not specified",i);
        is_symmetric = is_symmetric && m_terms[i]->is_symmetric;
    }

    if (m_input_indexes.size() > 0 && m_terms.size() != m_input_indexes.size())
        PLERROR("SummationKernel::build_: if 'input_indexes' is specified "
                "it must have the same size (%d) as 'terms'; found %d elements",
                m_terms.size(), m_input_indexes.size());
}
string PLearn::SummationKernel::classname ( ) const [virtual]

Reimplemented from PLearn::Object.

Definition at line 53 of file SummationKernel.cc.

void PLearn::SummationKernel::computeGramMatrix ( Mat  K) const [virtual]

Compute the Gram Matrix by calling subkernels computeGramMatrix.

Reimplemented from PLearn::Kernel.

Definition at line 221 of file SummationKernel.cc.

References i, PLearn::TMat< T >::length(), n, and PLearn::TMat< T >::width().

{
    // Assume that K has the right size; will have error in subkernels
    // evaluation if not the right size in any case.
    m_gram_buf.resize(K.width(), K.length());

    for (int i=0, n=m_terms.size() ; i<n ; ++i) {
        if (i==0)
            m_terms[i]->computeGramMatrix(K);
        else {
            m_terms[i]->computeGramMatrix(m_gram_buf);
            K += m_gram_buf;
        }
    }
}

Here is the call graph for this function:

void PLearn::SummationKernel::computeGramMatrixDerivative ( Mat KD,
const string &  kernel_param,
real  epsilon = 1e-6 
) const [virtual]

Directly compute the derivative with respect to hyperparameters (Faster than finite differences...)

Reimplemented from PLearn::Kernel.

Definition at line 239 of file SummationKernel.cc.

References i, PLearn::lexical_cast(), PLERROR, and PLearn::string_begins_with().

{
    // Find which term we want to compute the derivative for
    if (string_begins_with(kernel_param, "terms[")) {
        string::size_type rest = kernel_param.find("].");
        if (rest == string::npos)
            PLERROR("%s: malformed hyperparameter name for computing derivative '%s'",
                    __FUNCTION__, kernel_param.c_str());

        string sub_param  = kernel_param.substr(rest+2);
        string term_index = kernel_param.substr(6,rest-6); // len("terms[") == 6
        int i = lexical_cast<int>(term_index);
        if (i < 0 || i >= m_terms.size())
            PLERROR("%s: out of bounds access to term %d when computing derivative\n"
                    "for kernel parameter '%s'; only %d terms (0..%d) are available\n"
                    "in the SummationKernel", __FUNCTION__, i, kernel_param.c_str(),
                    m_terms.size(), m_terms.size()-1);
        
        m_terms[i]->computeGramMatrixDerivative(KD, sub_param, epsilon);
    }
    else
        inherited::computeGramMatrixDerivative(KD, kernel_param, epsilon);

    // Compare against finite differences
    // Mat KD1;
    // Kernel::computeGramMatrixDerivative(KD1, kernel_param, epsilon);
    // cerr << "Kernel hyperparameter: " << kernel_param << endl;
    // cerr << "Analytic derivative (15th row):" << endl
    //      << KD(15) << endl
    //      << "Finite differences:" << endl
    //      << KD1(15) << endl;
}

Here is the call graph for this function:

void PLearn::SummationKernel::declareOptions ( OptionList ol) [static, protected]

Declares the class options.

Reimplemented from PLearn::Kernel.

Definition at line 64 of file SummationKernel.cc.

References PLearn::OptionBase::buildoption, PLearn::declareOption(), m_input_indexes, and m_terms.

{
    declareOption(
        ol, "terms", &SummationKernel::m_terms, OptionBase::buildoption,
        "Individual kernels to add to produce the final result.  The\n"
        "hyperparameters of kernel i can be accesed under the option names\n"
        "'terms[i].hyperparam' for, e.g. GaussianProcessRegressor.\n");

    declareOption(
        ol, "input_indexes", &SummationKernel::m_input_indexes,
        OptionBase::buildoption,
        "Optionally, one can specify which of individual input variables should\n"
        "be routed to each kernel.  The format is as a vector of vectors: for\n"
        "each kernel in 'terms', one must list the INDEXES in the original input\n"
        "vector(zero-based) that should be passed to that kernel.  If a list of\n"
        "indexes is empty for a given kernel, it means that the COMPLETE input\n"
        "vector should be passed to the kernel.\n");
    
    // Now call the parent class' declareOptions
    inherited::declareOptions(ol);
}

Here is the call graph for this function:

static const PPath& PLearn::SummationKernel::declaringFile ( ) [inline, static]

Reimplemented from PLearn::Kernel.

Definition at line 116 of file SummationKernel.h.

:
SummationKernel * PLearn::SummationKernel::deepCopy ( CopiesMap copies) const [virtual]

Reimplemented from PLearn::Kernel.

Definition at line 53 of file SummationKernel.cc.

real PLearn::SummationKernel::evaluate ( const Vec x1,
const Vec x2 
) const [virtual]

Compute K(x1,x2).

Implements PLearn::Kernel.

Definition at line 162 of file SummationKernel.cc.

References i, n, and PLearn::selectElements().

{
    real kernel_value = 0.0;
    bool split_inputs = m_input_indexes.size() > 0;
    for (int i=0, n=m_terms.size() ; i<n ; ++i) {
        if (split_inputs && m_input_indexes[i].size() > 0) {
            selectElements(x1, m_input_indexes[i], m_input_buf1[i]);
            selectElements(x2, m_input_indexes[i], m_input_buf2[i]);
            kernel_value += m_terms[i]->evaluate(m_input_buf1[i],
                                                 m_input_buf2[i]);
        }
        else
            kernel_value += m_terms[i]->evaluate(x1,x2);
    }
    return kernel_value;
}

Here is the call graph for this function:

void PLearn::SummationKernel::evaluate_all_i_x ( const Vec x,
const Vec k_xi_x,
real  squared_norm_of_x = -1,
int  istart = 0 
) const [virtual]

Fill k_xi_x with K(x_i, x), for all i from istart to istart + k_xi_x.length() - 1.

Reimplemented from PLearn::Kernel.

Definition at line 200 of file SummationKernel.cc.

References PLearn::TVec< T >::fill(), i, n, PLearn::selectElements(), and PLearn::TVec< T >::size().

{
    k_xi_x.fill(0.0);
    m_eval_buf.resize(k_xi_x.size());
    bool split_inputs = m_input_indexes.size() > 0;
    for (int i=0, n=m_terms.size() ; i<n ; ++i) {
        // Note: if we slice x, we cannot rely on sq_norm_of_x any more...
        if (split_inputs && m_input_indexes[i].size() > 0) {
            selectElements(x, m_input_indexes[i], m_input_buf1[i]);
            m_terms[i]->evaluate_all_i_x(m_input_buf1[i], m_eval_buf, -1, istart);
        }
        else
            m_terms[i]->evaluate_all_i_x(x, m_eval_buf, sq_norm_of_x, istart);

        k_xi_x += m_eval_buf;
    }
}

Here is the call graph for this function:

real PLearn::SummationKernel::evaluate_i_x ( int  i,
const Vec x,
real   
) const [virtual]

Evaluate a test example x against a train example given by its index.

Reimplemented from PLearn::Kernel.

Definition at line 182 of file SummationKernel.cc.

References i, n, and PLearn::selectElements().

{
    real kernel_value = 0.0;
    bool split_inputs = m_input_indexes.size() > 0;
    for (int i=0, n=m_terms.size() ; i<n ; ++i) {
        if (split_inputs && m_input_indexes[i].size() > 0) {
            selectElements(x, m_input_indexes[i], m_input_buf1[i]);
            kernel_value += m_terms[i]->evaluate_i_x(j, m_input_buf1[i]);
        }
        else
            kernel_value += m_terms[i]->evaluate_i_x(j, x);
    }
    return kernel_value;
}

Here is the call graph for this function:

OptionList & PLearn::SummationKernel::getOptionList ( ) const [virtual]

Reimplemented from PLearn::Object.

Definition at line 53 of file SummationKernel.cc.

OptionMap & PLearn::SummationKernel::getOptionMap ( ) const [virtual]

Reimplemented from PLearn::Object.

Definition at line 53 of file SummationKernel.cc.

RemoteMethodMap & PLearn::SummationKernel::getRemoteMethodMap ( ) const [virtual]

Reimplemented from PLearn::Object.

Definition at line 53 of file SummationKernel.cc.

void PLearn::SummationKernel::makeDeepCopyFromShallowCopy ( CopiesMap copies) [virtual]

Transforms a shallow copy into a deep copy.

Reimplemented from PLearn::Kernel.

Definition at line 276 of file SummationKernel.cc.

References PLearn::deepCopyField().

Here is the call graph for this function:

void PLearn::SummationKernel::setDataForKernelMatrix ( VMat  the_data) [virtual]

Distribute to terms (sub-kernels) in the summation, subsetting if required.

Reimplemented from PLearn::Kernel.

Definition at line 128 of file SummationKernel.cc.

References i, and n.

{
    inherited::setDataForKernelMatrix(the_data);
    bool split_inputs = m_input_indexes.size() > 0;
    for (int i=0, n=m_terms.size() ; i<n ; ++i) {
        if (split_inputs && m_input_indexes[i].size() > 0) {
            VMat sub_inputs = new SelectColumnsVMatrix(the_data, m_input_indexes[i]);
            m_terms[i]->setDataForKernelMatrix(sub_inputs);
        }
        else
            m_terms[i]->setDataForKernelMatrix(the_data);
    }
}

Member Data Documentation

Reimplemented from PLearn::Kernel.

Definition at line 116 of file SummationKernel.h.

Temporary buffer for kernel evaluation on all training dataset.

Definition at line 130 of file SummationKernel.h.

Temporary buffer for Gram matrix accumulation.

Definition at line 133 of file SummationKernel.h.

Input buffers for kernel evaluation in cases where subsetting is needed.

Definition at line 126 of file SummationKernel.h.

Definition at line 127 of file SummationKernel.h.

Optionally, one can specify which of individual input variables should be routed to each kernel.

The format is as a vector of vectors: for each kernel in 'terms', one must list the INDEXES in the original input vector(zero-based) that should be passed to that kernel. If a list of indexes is empty for a given kernel, it means that the COMPLETE input vector should be passed to the kernel.

Definition at line 76 of file SummationKernel.h.

Referenced by declareOptions().

Individual kernels to add to produce the final result.

The hyperparameters of kernel i can be accesed under the option names 'terms[i].hyperparam' for, e.g. GaussianProcessRegressor.

Definition at line 66 of file SummationKernel.h.

Referenced by declareOptions().


The documentation for this class was generated from the following files:
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Defines