SHOGUN  5.0.0
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Modules
List of all members | Public Member Functions | Public Attributes | Protected Member Functions | Protected Attributes
SVRGMinimizer Class Reference

Detailed Description

The class implements the stochastic variance reduced gradient (SVRG) minimizer.

Reference: Johnson, Rie, and Tong Zhang. "Accelerating stochastic gradient descent using predictive variance reduction." Advances in Neural Information Processing Systems. 2013.

Definition at line 47 of file SVRGMinimizer.h.

Inheritance diagram for SVRGMinimizer:
[legend]

Public Member Functions

 SVRGMinimizer ()
 
 SVRGMinimizer (FirstOrderSAGCostFunction *fun)
 
virtual ~SVRGMinimizer ()
 
virtual float64_t minimize ()
 
virtual const char * get_name () const
 
virtual void set_sgd_number_passes (int32_t sgd_passes)
 
virtual void set_average_update_interval (int32_t interval)
 
virtual bool supports_batch_update () const
 
virtual void set_gradient_updater (DescendUpdater *gradient_updater)
 
virtual void set_number_passes (int32_t num_passes)
 
virtual void set_learning_rate (LearningRate *learning_rate)
 
virtual int32_t get_iteration_counter ()
 
virtual void set_cost_function (FirstOrderCostFunction *fun)
 
virtual void unset_cost_function (bool is_unref=true)
 
virtual void set_penalty_weight (float64_t penalty_weight)
 
virtual void set_penalty_type (Penalty *penalty_type)
 
virtual CSGObjectshallow_copy () const
 
virtual CSGObjectdeep_copy () const
 
virtual bool is_generic (EPrimitiveType *generic) const
 
template<class T >
void set_generic ()
 
template<>
void set_generic ()
 
template<>
void set_generic ()
 
template<>
void set_generic ()
 
template<>
void set_generic ()
 
template<>
void set_generic ()
 
template<>
void set_generic ()
 
template<>
void set_generic ()
 
template<>
void set_generic ()
 
template<>
void set_generic ()
 
template<>
void set_generic ()
 
template<>
void set_generic ()
 
template<>
void set_generic ()
 
template<>
void set_generic ()
 
template<>
void set_generic ()
 
template<>
void set_generic ()
 
void unset_generic ()
 
virtual void print_serializable (const char *prefix="")
 
virtual bool save_serializable (CSerializableFile *file, const char *prefix="")
 
virtual bool load_serializable (CSerializableFile *file, const char *prefix="")
 
void set_global_io (SGIO *io)
 
SGIOget_global_io ()
 
void set_global_parallel (Parallel *parallel)
 
Parallelget_global_parallel ()
 
void set_global_version (Version *version)
 
Versionget_global_version ()
 
SGStringList< char > get_modelsel_names ()
 
void print_modsel_params ()
 
char * get_modsel_param_descr (const char *param_name)
 
index_t get_modsel_param_index (const char *param_name)
 
void build_gradient_parameter_dictionary (CMap< TParameter *, CSGObject * > *dict)
 
bool has (const std::string &name) const
 
template<typename T >
bool has (const Tag< T > &tag) const
 
template<typename T , typename U = void>
bool has (const std::string &name) const
 
template<typename T >
void set (const Tag< T > &_tag, const T &value)
 
template<typename T , typename U = void>
void set (const std::string &name, const T &value)
 
template<typename T >
get (const Tag< T > &_tag) const
 
template<typename T , typename U = void>
get (const std::string &name) const
 
virtual void update_parameter_hash ()
 
virtual bool parameter_hash_changed ()
 
virtual bool equals (CSGObject *other, float64_t accuracy=0.0, bool tolerant=false)
 
virtual CSGObjectclone ()
 

Public Attributes

SGIOio
 
Parallelparallel
 
Versionversion
 
Parameterm_parameters
 
Parameterm_model_selection_parameters
 
Parameterm_gradient_parameters
 
uint32_t m_hash
 

Protected Member Functions

virtual void init_minimization ()
 
virtual void do_proximal_operation (SGVector< float64_t >variable_reference)
 
virtual float64_t get_penalty (SGVector< float64_t > var)
 
virtual void update_gradient (SGVector< float64_t > gradient, SGVector< float64_t > var)
 
virtual void load_serializable_pre () throw (ShogunException)
 
virtual void load_serializable_post () throw (ShogunException)
 
virtual void save_serializable_pre () throw (ShogunException)
 
virtual void save_serializable_post () throw (ShogunException)
 
template<typename T >
void register_param (Tag< T > &_tag, const T &value)
 
template<typename T >
void register_param (const std::string &name, const T &value)
 

Protected Attributes

int32_t m_num_sgd_passes
 
int32_t m_svrg_interval
 
SGVector< float64_tm_average_gradient
 
SGVector< float64_tm_previous_variable
 
DescendUpdaterm_gradient_updater
 
int32_t m_num_passes
 
int32_t m_cur_passes
 
int32_t m_iter_counter
 
LearningRatem_learning_rate
 
FirstOrderCostFunctionm_fun
 
Penaltym_penalty_type
 
float64_t m_penalty_weight
 

Constructor & Destructor Documentation

Default constructor

Definition at line 36 of file SVRGMinimizer.cpp.

Constructor

Parameters
funstochastic cost function

Definition at line 46 of file SVRGMinimizer.cpp.

~SVRGMinimizer ( )
virtual

Destructor

Definition at line 42 of file SVRGMinimizer.cpp.

Member Function Documentation

void build_gradient_parameter_dictionary ( CMap< TParameter *, CSGObject * > *  dict)
inherited

Builds a dictionary of all parameters in SGObject as well of those of SGObjects that are parameters of this object. Dictionary maps parameters to the objects that own them.

Parameters
dictdictionary of parameters to be built.

Definition at line 630 of file SGObject.cpp.

CSGObject * clone ( )
virtualinherited

Creates a clone of the current object. This is done via recursively traversing all parameters, which corresponds to a deep copy. Calling equals on the cloned object always returns true although none of the memory of both objects overlaps.

Returns
an identical copy of the given object, which is disjoint in memory. NULL if the clone fails. Note that the returned object is SG_REF'ed

Definition at line 747 of file SGObject.cpp.

CSGObject * deep_copy ( ) const
virtualinherited

A deep copy. All the instance variables will also be copied.

Definition at line 231 of file SGObject.cpp.

void do_proximal_operation ( SGVector< float64_t variable_reference)
protectedvirtualinherited

Do proximal update in place

Parameters
variable_referencevariable_reference to be updated

Definition at line 71 of file FirstOrderStochasticMinimizer.cpp.

bool equals ( CSGObject other,
float64_t  accuracy = 0.0,
bool  tolerant = false 
)
virtualinherited

Recursively compares the current SGObject to another one. Compares all registered numerical parameters, recursion upon complex (SGObject) parameters. Does not compare pointers!

May be overwritten but please do with care! Should not be necessary in most cases.

Parameters
otherobject to compare with
accuracyaccuracy to use for comparison (optional)
tolerantallows linient check on float equality (within accuracy)
Returns
true if all parameters were equal, false if not

Definition at line 651 of file SGObject.cpp.

T get ( const Tag< T > &  _tag) const
inherited

Getter for a class parameter, identified by a Tag. Throws an exception if the class does not have such a parameter.

Parameters
_tagname and type information of parameter
Returns
value of the parameter identified by the input tag

Definition at line 367 of file SGObject.h.

T get ( const std::string &  name) const
inherited

Getter for a class parameter, identified by a name. Throws an exception if the class does not have such a parameter.

Parameters
namename of the parameter
Returns
value of the parameter corresponding to the input name and type

Definition at line 388 of file SGObject.h.

SGIO * get_global_io ( )
inherited

get the io object

Returns
io object

Definition at line 268 of file SGObject.cpp.

Parallel * get_global_parallel ( )
inherited

get the parallel object

Returns
parallel object

Definition at line 310 of file SGObject.cpp.

Version * get_global_version ( )
inherited

get the version object

Returns
version object

Definition at line 323 of file SGObject.cpp.

virtual int32_t get_iteration_counter ( )
virtualinherited

How many samples/mini-batch does the minimizer use?

Returns
the number of samples/mini-batches used in optimization

Definition at line 136 of file FirstOrderStochasticMinimizer.h.

SGStringList< char > get_modelsel_names ( )
inherited
Returns
vector of names of all parameters which are registered for model selection

Definition at line 531 of file SGObject.cpp.

char * get_modsel_param_descr ( const char *  param_name)
inherited

Returns description of a given parameter string, if it exists. SG_ERROR otherwise

Parameters
param_namename of the parameter
Returns
description of the parameter

Definition at line 555 of file SGObject.cpp.

index_t get_modsel_param_index ( const char *  param_name)
inherited

Returns index of model selection parameter with provided index

Parameters
param_namename of model selection parameter
Returns
index of model selection parameter with provided name, -1 if there is no such

Definition at line 568 of file SGObject.cpp.

virtual const char* get_name ( ) const
virtual

returns the name of the class

Returns
name SVRGMinimizer

Reimplemented from FirstOrderStochasticMinimizer.

Definition at line 71 of file SVRGMinimizer.h.

float64_t get_penalty ( SGVector< float64_t var)
protectedvirtualinherited

Get the penalty given target variables For L2 penalty, the target variable is \(w\) and the value of penalty is \(\lambda \frac{w^t w}{2}\), where \(\lambda\) is the weight of penalty

Parameters
varthe variable used in regularization

Definition at line 69 of file FirstOrderMinimizer.cpp.

bool has ( const std::string &  name) const
inherited

Checks if object has a class parameter identified by a name.

Parameters
namename of the parameter
Returns
true if the parameter exists with the input name

Definition at line 289 of file SGObject.h.

bool has ( const Tag< T > &  tag) const
inherited

Checks if object has a class parameter identified by a Tag.

Parameters
tagtag of the parameter containing name and type information
Returns
true if the parameter exists with the input tag

Definition at line 301 of file SGObject.h.

bool has ( const std::string &  name) const
inherited

Checks if a type exists for a class parameter identified by a name.

Parameters
namename of the parameter
Returns
true if the parameter exists with the input name and type

Definition at line 312 of file SGObject.h.

void init_minimization ( )
protectedvirtual

init the minimization process

Reimplemented from FirstOrderStochasticMinimizer.

Definition at line 69 of file SVRGMinimizer.cpp.

bool is_generic ( EPrimitiveType *  generic) const
virtualinherited

If the SGSerializable is a class template then TRUE will be returned and GENERIC is set to the type of the generic.

Parameters
genericset to the type of the generic if returning TRUE
Returns
TRUE if a class template.

Definition at line 329 of file SGObject.cpp.

bool load_serializable ( CSerializableFile file,
const char *  prefix = "" 
)
virtualinherited

Load this object from file. If it will fail (returning FALSE) then this object will contain inconsistent data and should not be used!

Parameters
filewhere to load from
prefixprefix for members
Returns
TRUE if done, otherwise FALSE

Definition at line 402 of file SGObject.cpp.

void load_serializable_post ( )
throw (ShogunException
)
protectedvirtualinherited

Can (optionally) be overridden to post-initialize some member variables which are not PARAMETER::ADD'ed. Make sure that at first the overridden method BASE_CLASS::LOAD_SERIALIZABLE_POST is called.

Exceptions
ShogunExceptionwill be thrown if an error occurs.

Reimplemented in CKernel, CWeightedDegreePositionStringKernel, CList, CAlphabet, CLinearHMM, CGaussianKernel, CInverseMultiQuadricKernel, CCircularKernel, and CExponentialKernel.

Definition at line 459 of file SGObject.cpp.

void load_serializable_pre ( )
throw (ShogunException
)
protectedvirtualinherited

Can (optionally) be overridden to pre-initialize some member variables which are not PARAMETER::ADD'ed. Make sure that at first the overridden method BASE_CLASS::LOAD_SERIALIZABLE_PRE is called.

Exceptions
ShogunExceptionwill be thrown if an error occurs.

Reimplemented in CDynamicArray< T >, CDynamicArray< float64_t >, CDynamicArray< float32_t >, CDynamicArray< int32_t >, CDynamicArray< char >, CDynamicArray< bool >, and CDynamicObjectArray.

Definition at line 454 of file SGObject.cpp.

float64_t minimize ( )
virtual

Do minimization and get the optimal value

Returns
optimal value

Implements FirstOrderStochasticMinimizer.

Definition at line 89 of file SVRGMinimizer.cpp.

bool parameter_hash_changed ( )
virtualinherited
Returns
whether parameter combination has changed since last update

Definition at line 295 of file SGObject.cpp.

void print_modsel_params ( )
inherited

prints all parameter registered for model selection and their type

Definition at line 507 of file SGObject.cpp.

void print_serializable ( const char *  prefix = "")
virtualinherited

prints registered parameters out

Parameters
prefixprefix for members

Definition at line 341 of file SGObject.cpp.

void register_param ( Tag< T > &  _tag,
const T &  value 
)
protectedinherited

Registers a class parameter which is identified by a tag. This enables the parameter to be modified by set() and retrieved by get(). Parameters can be registered in the constructor of the class.

Parameters
_tagname and type information of parameter
valuevalue of the parameter

Definition at line 439 of file SGObject.h.

void register_param ( const std::string &  name,
const T &  value 
)
protectedinherited

Registers a class parameter which is identified by a name. This enables the parameter to be modified by set() and retrieved by get(). Parameters can be registered in the constructor of the class.

Parameters
namename of the parameter
valuevalue of the parameter along with type information

Definition at line 452 of file SGObject.h.

bool save_serializable ( CSerializableFile file,
const char *  prefix = "" 
)
virtualinherited

Save this object to file.

Parameters
filewhere to save the object; will be closed during returning if PREFIX is an empty string.
prefixprefix for members
Returns
TRUE if done, otherwise FALSE

Definition at line 347 of file SGObject.cpp.

void save_serializable_post ( )
throw (ShogunException
)
protectedvirtualinherited

Can (optionally) be overridden to post-initialize some member variables which are not PARAMETER::ADD'ed. Make sure that at first the overridden method BASE_CLASS::SAVE_SERIALIZABLE_POST is called.

Exceptions
ShogunExceptionwill be thrown if an error occurs.

Reimplemented in CKernel.

Definition at line 469 of file SGObject.cpp.

void save_serializable_pre ( )
throw (ShogunException
)
protectedvirtualinherited

Can (optionally) be overridden to pre-initialize some member variables which are not PARAMETER::ADD'ed. Make sure that at first the overridden method BASE_CLASS::SAVE_SERIALIZABLE_PRE is called.

Exceptions
ShogunExceptionwill be thrown if an error occurs.

Reimplemented in CKernel, CDynamicArray< T >, CDynamicArray< float64_t >, CDynamicArray< float32_t >, CDynamicArray< int32_t >, CDynamicArray< char >, CDynamicArray< bool >, and CDynamicObjectArray.

Definition at line 464 of file SGObject.cpp.

void set ( const Tag< T > &  _tag,
const T &  value 
)
inherited

Setter for a class parameter, identified by a Tag. Throws an exception if the class does not have such a parameter.

Parameters
_tagname and type information of parameter
valuevalue of the parameter

Definition at line 328 of file SGObject.h.

void set ( const std::string &  name,
const T &  value 
)
inherited

Setter for a class parameter, identified by a name. Throws an exception if the class does not have such a parameter.

Parameters
namename of the parameter
valuevalue of the parameter along with type information

Definition at line 354 of file SGObject.h.

virtual void set_average_update_interval ( int32_t  interval)
virtual

Set the number of interval to average stochastic sample gradients

If we have \((n-g)\) passes to go through data and the interval is \(k\), we will average stochastic sample gradients at the 0-th, k-th, 2k-th, 3k-th, ... pass

Note that \(n\) is the total number to go through data and \(g\) is the number of using SGDMinimizer to initialize variables,

Parameters
intervalhow often to average stochastic sample gradients

Definition at line 101 of file SVRGMinimizer.h.

void set_cost_function ( FirstOrderCostFunction fun)
virtualinherited

Set cost function used in the minimizer

Parameters
funthe cost function

Definition at line 42 of file FirstOrderMinimizer.cpp.

void set_generic ( )
inherited

Definition at line 74 of file SGObject.cpp.

void set_generic ( )
inherited

Definition at line 79 of file SGObject.cpp.

void set_generic ( )
inherited

Definition at line 84 of file SGObject.cpp.

void set_generic ( )
inherited

Definition at line 89 of file SGObject.cpp.

void set_generic ( )
inherited

Definition at line 94 of file SGObject.cpp.

void set_generic ( )
inherited

Definition at line 99 of file SGObject.cpp.

void set_generic ( )
inherited

Definition at line 104 of file SGObject.cpp.

void set_generic ( )
inherited

Definition at line 109 of file SGObject.cpp.

void set_generic ( )
inherited

Definition at line 114 of file SGObject.cpp.

void set_generic ( )
inherited

Definition at line 119 of file SGObject.cpp.

void set_generic ( )
inherited

Definition at line 124 of file SGObject.cpp.

void set_generic ( )
inherited

Definition at line 129 of file SGObject.cpp.

void set_generic ( )
inherited

Definition at line 134 of file SGObject.cpp.

void set_generic ( )
inherited

Definition at line 139 of file SGObject.cpp.

void set_generic ( )
inherited

Definition at line 144 of file SGObject.cpp.

void set_generic ( )
inherited

set generic type to T

void set_global_io ( SGIO io)
inherited

set the io object

Parameters
ioio object to use

Definition at line 261 of file SGObject.cpp.

void set_global_parallel ( Parallel parallel)
inherited

set the parallel object

Parameters
parallelparallel object to use

Definition at line 274 of file SGObject.cpp.

void set_global_version ( Version version)
inherited

set the version object

Parameters
versionversion object to use

Definition at line 316 of file SGObject.cpp.

void set_gradient_updater ( DescendUpdater gradient_updater)
virtualinherited

Set a gradient updater

Parameters
gradient_updaterthe gradient_updater

Definition at line 38 of file FirstOrderStochasticMinimizer.cpp.

void set_learning_rate ( LearningRate learning_rate)
virtualinherited

Set the learning rate of a minimizer

Parameters
learning_ratelearn rate or step size

Definition at line 61 of file FirstOrderStochasticMinimizer.cpp.

void set_number_passes ( int32_t  num_passes)
virtualinherited

Set the number of times to go through all data points (samples) For example, num_passes=1 means go through all data points once.

Recall that a stochastic cost function \(f(w)\) can be written as \(\sum_i{ f_i(w) }\), where \(f_i(w)\) is the differentiable function for the i-th sample.

Parameters
num_passesthe number of times

Definition at line 55 of file FirstOrderStochasticMinimizer.cpp.

void set_penalty_type ( Penalty penalty_type)
virtualinherited

Set the type of penalty For example, L2 penalty

Parameters
penalty_typethe type of penalty. If NULL is given, regularization is not enabled.

Definition at line 53 of file FirstOrderMinimizer.cpp.

void set_penalty_weight ( float64_t  penalty_weight)
virtualinherited

Set the weight of penalty

Parameters
penalty_weightthe weight of penalty, which is positive

Definition at line 63 of file FirstOrderMinimizer.cpp.

virtual void set_sgd_number_passes ( int32_t  sgd_passes)
virtual

Set the number to go through data using SGDMinimizer to initialize variables before SVRG minimization

Parameters
sgd_passesthe number to go through data using SGDMinimizer

Definition at line 79 of file SVRGMinimizer.h.

CSGObject * shallow_copy ( ) const
virtualinherited

A shallow copy. All the SGObject instance variables will be simply assigned and SG_REF-ed.

Reimplemented in CGaussianKernel.

Definition at line 225 of file SGObject.cpp.

virtual bool supports_batch_update ( ) const
virtualinherited

Does minimizer support batch update

Returns
whether minimizer supports batch update

Implements FirstOrderMinimizer.

Definition at line 102 of file FirstOrderStochasticMinimizer.h.

virtual void unset_cost_function ( bool  is_unref = true)
virtualinherited

Unset cost function used in the minimizer

Definition at line 94 of file FirstOrderMinimizer.h.

void unset_generic ( )
inherited

unset generic type

this has to be called in classes specializing a template class

Definition at line 336 of file SGObject.cpp.

void update_gradient ( SGVector< float64_t gradient,
SGVector< float64_t var 
)
protectedvirtualinherited

Add gradient of the penalty wrt target variables to unpenalized gradient For least sqaure with L2 penalty,

\[ L2f(w)=f(w) + L2(w) \]

where \( f(w)=\sum_i{(y_i-w^T x_i)^2}\) is the least sqaure cost function and \(L2(w)=\lambda \frac{w^t w}{2}\) is the L2 penalty

Target variables is \(w\) Unpenalized gradient is \(\frac{\partial f(w) }{\partial w}\) Gradient of the penalty wrt target variables is \(\frac{\partial L2(w) }{\partial w}\)

Parameters
gradientunpenalized gradient wrt its target variable
varthe target variable

Definition at line 81 of file FirstOrderMinimizer.cpp.

void update_parameter_hash ( )
virtualinherited

Updates the hash of current parameter combination

Definition at line 281 of file SGObject.cpp.

Member Data Documentation

SGIO* io
inherited

io

Definition at line 537 of file SGObject.h.

SGVector<float64_t> m_average_gradient
protected

used to store average gradient

Definition at line 126 of file SVRGMinimizer.h.

int32_t m_cur_passes
protectedinherited

current iteration to go through data

Definition at line 156 of file FirstOrderStochasticMinimizer.h.

FirstOrderCostFunction* m_fun
protectedinherited

Cost function

Definition at line 146 of file FirstOrderMinimizer.h.

Parameter* m_gradient_parameters
inherited

parameters wrt which we can compute gradients

Definition at line 552 of file SGObject.h.

DescendUpdater* m_gradient_updater
protectedinherited

the gradient update step

Definition at line 150 of file FirstOrderStochasticMinimizer.h.

uint32_t m_hash
inherited

Hash of parameter values

Definition at line 555 of file SGObject.h.

int32_t m_iter_counter
protectedinherited

number of used samples/mini-batches

Definition at line 159 of file FirstOrderStochasticMinimizer.h.

LearningRate* m_learning_rate
protectedinherited

learning_rate object

Definition at line 162 of file FirstOrderStochasticMinimizer.h.

Parameter* m_model_selection_parameters
inherited

model selection parameters

Definition at line 549 of file SGObject.h.

int32_t m_num_passes
protectedinherited

iteration to go through data

Definition at line 153 of file FirstOrderStochasticMinimizer.h.

int32_t m_num_sgd_passes
protected

the number to go through data using SGD before SVRG update

Definition at line 120 of file SVRGMinimizer.h.

Parameter* m_parameters
inherited

parameters

Definition at line 546 of file SGObject.h.

Penalty* m_penalty_type
protectedinherited

the type of penalty

Definition at line 149 of file FirstOrderMinimizer.h.

float64_t m_penalty_weight
protectedinherited

the weight of penalty

Definition at line 152 of file FirstOrderMinimizer.h.

SGVector<float64_t> m_previous_variable
protected

used to store previous result

Definition at line 129 of file SVRGMinimizer.h.

int32_t m_svrg_interval
protected

interval to average gradient

Definition at line 123 of file SVRGMinimizer.h.

Parallel* parallel
inherited

parallel

Definition at line 540 of file SGObject.h.

Version* version
inherited

version

Definition at line 543 of file SGObject.h.


The documentation for this class was generated from the following files:

SHOGUN Machine Learning Toolbox - Documentation