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::ARDBaseKernel Class Reference

Base class for kernels that carry out Automatic Relevance Determination (ARD) More...

#include <ARDBaseKernel.h>

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

List of all members.

Public Member Functions

 ARDBaseKernel ()
 Default constructor.
virtual string classname () const
virtual OptionListgetOptionList () const
virtual OptionMapgetOptionMap () const
virtual RemoteMethodMapgetRemoteMethodMap () const
virtual ARDBaseKerneldeepCopy (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

real m_isp_signal_sigma
 Inverse softplus of the global signal variance. Default value=0.0.
real m_isp_global_sigma
 Inverse softplus of the global length-scale.
Vec m_isp_input_sigma
 If specified, contain input-specific length-scales that can be individually optimized for (these are the ARD hyperparameters).

Static Public Attributes

static StaticInitializer _static_initializer_

Static Protected Member Functions

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

Protected Attributes

Vec m_input_sigma
 Buffer for final input sigma (add both global and per-input terms); Can be used by derived class.

Private Types

typedef KroneckerBaseKernel inherited

Private Member Functions

void build_ ()
 This does the actual building.

Detailed Description

Base class for kernels that carry out Automatic Relevance Determination (ARD)

The purpose of this Kernel is to introduce utility options that are generally shared by Kernels that perform Automatic Relevance Determination (ARD). It does not introduce any specific behavior related to those options (since exactly where the ARD hyperparameters are used is very kernel-specific, this is left to derived classes).

Note that to make its operations more robust when used with unconstrained optimization of hyperparameters, all hyperparameters of this kernel are specified in the inverse softplus domain, hence the 'isp' prefix. This is used in preference to the log-domain used by Rasmussen and Williams in their implementation of gaussian processes, due to numerical stability. (It may happen that the optimizer jumps 'too far' along one hyperparameter and this causes the Gram matrix to become extremely ill-conditioned.)

Definition at line 64 of file ARDBaseKernel.h.


Member Typedef Documentation


Constructor & Destructor Documentation

PLearn::ARDBaseKernel::ARDBaseKernel ( )

Default constructor.

Definition at line 60 of file ARDBaseKernel.cc.


Member Function Documentation

string PLearn::ARDBaseKernel::_classname_ ( ) [static]
OptionList & PLearn::ARDBaseKernel::_getOptionList_ ( ) [static]
RemoteMethodMap & PLearn::ARDBaseKernel::_getRemoteMethodMap_ ( ) [static]
bool PLearn::ARDBaseKernel::_isa_ ( const Object o) [static]
Object * PLearn::ARDBaseKernel::_new_instance_for_typemap_ ( ) [static]
StaticInitializer ARDBaseKernel::_static_initializer_ & PLearn::ARDBaseKernel::_static_initialize_ ( ) [static]
void PLearn::ARDBaseKernel::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::KroneckerBaseKernel.

Reimplemented in PLearn::LinearARDKernel, PLearn::Matern1ARDKernel, PLearn::NeuralNetworkARDKernel, PLearn::RationalQuadraticARDKernel, and PLearn::SquaredExponentialARDKernel.

Definition at line 97 of file ARDBaseKernel.cc.

References PLearn::KroneckerBaseKernel::build(), and build_().

Referenced by PLearn::RationalQuadraticARDKernel::build(), and PLearn::Matern1ARDKernel::build().

{
    // ### Nothing to add here, simply calls build_
    inherited::build();
    build_();
}

Here is the call graph for this function:

Here is the caller graph for this function:

void PLearn::ARDBaseKernel::build_ ( ) [private]

This does the actual building.

Reimplemented from PLearn::KroneckerBaseKernel.

Reimplemented in PLearn::LinearARDKernel, PLearn::Matern1ARDKernel, PLearn::NeuralNetworkARDKernel, PLearn::RationalQuadraticARDKernel, and PLearn::SquaredExponentialARDKernel.

Definition at line 105 of file ARDBaseKernel.cc.

Referenced by build().

{ }

Here is the caller graph for this function:

string PLearn::ARDBaseKernel::classname ( ) const [virtual]
void PLearn::ARDBaseKernel::declareOptions ( OptionList ol) [static, protected]

Declares the class options.

Reimplemented from PLearn::KroneckerBaseKernel.

Reimplemented in PLearn::LinearARDKernel, PLearn::Matern1ARDKernel, PLearn::NeuralNetworkARDKernel, PLearn::RationalQuadraticARDKernel, and PLearn::SquaredExponentialARDKernel.

Definition at line 68 of file ARDBaseKernel.cc.

References PLearn::OptionBase::buildoption, PLearn::declareOption(), PLearn::KroneckerBaseKernel::declareOptions(), m_isp_global_sigma, m_isp_input_sigma, and m_isp_signal_sigma.

Referenced by PLearn::RationalQuadraticARDKernel::declareOptions(), and PLearn::Matern1ARDKernel::declareOptions().

{
    declareOption(
        ol, "isp_signal_sigma",
        &ARDBaseKernel::m_isp_signal_sigma,
        OptionBase::buildoption,
        "Inverse softplus of the global signal variance.  Default value=0.0");

    declareOption(
        ol, "isp_global_sigma",
        &ARDBaseKernel::m_isp_global_sigma,
        OptionBase::buildoption,
        "Inverse softplus of the global length-scale.  Note that if ARD is\n"
        "performed on input-specific sigmas, this hyperparameter should have a\n"
        "fixed value (and not be varied during the optimization).  Default\n"
        "value=0.0.\n");

    declareOption(
        ol, "isp_input_sigma",
        &ARDBaseKernel::m_isp_input_sigma,
        OptionBase::buildoption,
        "If specified, contain input-specific length-scales that can be\n"
        "individually optimized for (these are the ARD hyperparameters).\n");

    // Now call the parent class' declareOptions
    inherited::declareOptions(ol);
}

Here is the call graph for this function:

Here is the caller graph for this function:

static const PPath& PLearn::ARDBaseKernel::declaringFile ( ) [inline, static]
ARDBaseKernel * PLearn::ARDBaseKernel::deepCopy ( CopiesMap copies) const [virtual]
OptionList & PLearn::ARDBaseKernel::getOptionList ( ) const [virtual]
OptionMap & PLearn::ARDBaseKernel::getOptionMap ( ) const [virtual]
RemoteMethodMap & PLearn::ARDBaseKernel::getRemoteMethodMap ( ) const [virtual]
void PLearn::ARDBaseKernel::makeDeepCopyFromShallowCopy ( CopiesMap copies) [virtual]

Member Data Documentation


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