lemga Namespace Reference

Learning models and generic algorithms. More...

Classes
class	AdaBoost
	AdaBoost (adaptive boosting). More...
class	AdaBoost_ECOC
	AdaBoost.ECC with exponential cost and Hamming distance. More...
class	AdaBoost_ERP
	AdaBoost.ERP (AdaBoost.ECC with Re-Partitioning). More...
class	Aggregating
	An abstract class for aggregating. More...
class	Bagging
	Bagging (boostrap aggregating). More...
class	Boosting
	Boosting generates a linear combination of hypotheses. More...
struct	_boost_gd
class	Cascade
	Aggregate hypotheses in a cascade (sequential) way. More...
class	CGBoost
	CGBoost (Conjugate Gradient Boosting). More...
struct	_boost_cg
class	CrossVal
	A combination of cross-validation and model selection. More...
class	vFoldCrossVal
	v-fold cross validation. More...
class	HoldoutCrossVal
	Randomized holdout cross-validation. More...
class	DataFeeder
	Feed (random splitted) training and testing data. More...
class	dataset
	Class template for storing, retrieving, and manipulating a vector of input-output style data. More...
class	FeedForwardNN
class	LearnModel
	A unified interface for learning models. More...
class	LPBoost
	LPBoost (Linear-Programming Boosting). More...
struct	MgnCost
	Cost proxy used in MgnBoost. More...
class	MgnBoost
	MgnBoost (margin maximizing boosting). More...
struct	_mgn_gd
class	MultiClass_ECOC
	Multiclass classification using error-correcting output code. More...
class	NNLayer
	A layer in a neural network. More...
struct	_search
	Interface used in iterative optimization algorithms. More...
struct	_gradient_descent
	Gradient descent. More...
struct	_gd_weightdecay
	Gradient descent with weight decay. More...
struct	_gd_momentum
	Gradient descent with momentum. More...
struct	_gd_adaptive
struct	_line_search
struct	_conjugate_gradient
class	Ordinal_BLE
	Ordinal regression via binary learning on extended examples. More...
class	Perceptron
	Perceptron models a type of artificial neural network that consists of only one neuron, invented by Frank Rosenblatt in 1957. More...
class	Pulse
	Multi-transition pulse functions (step functions). More...
class	Stump
	Decision stump. More...
class	SVM
Namespaces
namespace	cost
	Cost functions and their derivatives.
namespace	details
namespace	kernel
	Kernels (inner product in some space).
namespace	op
	Operators used in optimization.
Typedefs
typedef std::vector< std::vector< REAL > >	WMAT
typedef std::vector< DataWgt >	JointWgt
typedef const_shared_ptr< JointWgt >	pJointWgt
typedef vFoldCrossVal	kFoldCrossVal
typedef std::vector< REAL >	Input
typedef std::vector< REAL >	Output
typedef dataset< Input, Output >	DataSet
typedef std::vector< REAL >	DataWgt
typedef const_shared_ptr< DataSet >	pDataSet
typedef const_shared_ptr< DataWgt >	pDataWgt
typedef var_shared_ptr< LearnModel >	pLearnModel
typedef const_shared_ptr< LearnModel >	pcLearnModel
typedef std::vector< int >	ECOC_VECTOR
typedef std::vector< ECOC_VECTOR >	ECOC_TABLE
typedef std::vector< std::vector< REAL > >	EXT_TABLE
typedef std::map< REAL, REAL >::iterator	MI
typedef svm_node *	p_svm_node
Enumerations
enum	ECOC_TYPE { NO_TYPE, ONE_VS_ONE, ONE_VS_ALL }
enum	BLE_TYPE { MULTI_THRESHOLD, BLE_DEFAULT = MULTI_THRESHOLD }
Functions
DataSet *	load_data (std::istream &, UINT, UINT, UINT)
	Load a data set from a stream.
DataSet *	load_data (std::istream &is, UINT n)
bool	lp_add_hypothesis (LPX *lp, int *ndx, double *val, const LearnModel &lm, const pDataWgt &pdw=0, REAL maxe=0)
REAL	lp_solve (LPX *lp, pDataWgt &pdw)
template<class SEARCH>
void	iterative_optimize (SEARCH s)
	Main search routine.
void	update_wgt (RVEC &wgt, const RVEC &dir, const RMAT &X, const RVEC &y)
void	dset_extract (const pDataSet &ptd, RMAT &X, RVEC &y)
void	dset_mult_wgt (const pDataWgt &ptw, RVEC &y)
UINT	randcdf (REAL r, const RVEC &cdf)
p_svm_node	fill_svm_node (const Input &x, struct svm_node *pool)

---

Detailed Description

Learning models and generic algorithms.

The idea is to separate the learning model and optimization techniques.

Using vectorop.h for default vector operation

Todo:

documentation: basic idea, ...

---

Typedef Documentation

typedef dataset<Input,Output> DataSet

Definition at line 26 of file learnmodel.h.

typedef std::vector<REAL> DataWgt

Definition at line 27 of file learnmodel.h.

typedef std::vector<ECOC_VECTOR> ECOC_TABLE

Definition at line 21 of file multiclass_ecoc.h.

typedef std::vector<int> ECOC_VECTOR

Definition at line 20 of file multiclass_ecoc.h.

typedef std::vector<std::vector<REAL> > EXT_TABLE

Definition at line 15 of file ordinal_ble.h.

typedef std::vector<REAL> Input

Definition at line 23 of file learnmodel.h.

typedef std::vector<DataWgt> JointWgt

Definition at line 16 of file adaboost_ecoc.h.

typedef vFoldCrossVal kFoldCrossVal

Definition at line 115 of file crossval.h.

typedef std::map<REAL,REAL>::iterator MI

Definition at line 41 of file stump.cpp.

typedef std::vector<REAL> Output

Definition at line 24 of file learnmodel.h.

typedef struct svm_node* p_svm_node

Definition at line 31 of file svm.cpp.

typedef const_shared_ptr<LearnModel> pcLearnModel

Definition at line 196 of file learnmodel.h.

typedef const_shared_ptr<DataSet> pDataSet

Definition at line 28 of file learnmodel.h.

typedef const_shared_ptr<DataWgt> pDataWgt

Definition at line 29 of file learnmodel.h.

typedef const_shared_ptr<JointWgt> pJointWgt

Definition at line 17 of file adaboost_ecoc.h.

typedef var_shared_ptr<LearnModel> pLearnModel

Definition at line 195 of file learnmodel.h.

typedef std::vector<std::vector<REAL> > WMAT

Definition at line 42 of file adaboost_ecoc.cpp.

---

Enumeration Type Documentation

enum BLE_TYPE

Enumerator: MULTI_THRESHOLD  BLE_DEFAULT  Definition at line 16 of file ordinal_ble.h.

enum ECOC_TYPE

Enumerator: NO_TYPE  ONE_VS_ONE  ONE_VS_ALL  Definition at line 22 of file multiclass_ecoc.h.

---

Function Documentation

void lemga::dset_extract (  const pDataSet &  ptd, RMAT &  X, RVEC &  y )

Definition at line 127 of file perceptron.cpp. Referenced by Perceptron::train().

void lemga::dset_mult_wgt (  const pDataWgt &  ptw, RVEC &  y )  [inline]

Definition at line 137 of file perceptron.cpp.

p_svm_node lemga::fill_svm_node (  const Input &  x, struct svm_node *  pool )

Definition at line 92 of file svm.cpp. Referenced by SVM::kernel(), SVM::operator()(), and SVM::signed_margin().

void lemga::iterative_optimize (  SEARCH  s  )

Main search routine. Definition at line 74 of file optimize.h. Referenced by FeedForwardNN::train(), MgnBoost::train_gd(), CGBoost::train_gd(), and Boosting::train_gd().

DataSet * load_data (  std::istream &  is, UINT  n )

An easier-to-use version, where the output dimension is fixed at 1, and the input dimension is auto-detected. This version requires that each row of stream is should be a sample. Definition at line 46 of file learnmodel.cpp. References dataset::append(), and dataset::size().

DataSet * load_data (  std::istream &  is, UINT  n, UINT  in, UINT  out )

Load a data set from a stream. Each sample consists of first the input and then the output. Numbers are separated by spaces. Parameters: is  the input stream n  gives the number of samples in  is the dimension of input out  is the dimension of output Todo: documentation: why separate function Definition at line 37 of file learnmodel.cpp. Referenced by DataFeeder::DataFeeder().

bool lemga::lp_add_hypothesis (  LPX *  lp, int *  ndx, double *  val, const LearnModel &  lm, const pDataWgt &  pdw = 0, REAL  maxe = 0 )

Definition at line 22 of file lpboost.cpp. References LearnModel::c_error(), LearnModel::get_output(), LearnModel::train_data(), and U.

REAL lemga::lp_solve (  LPX *  lp, pDataWgt &  pdw )

Definition at line 49 of file lpboost.cpp. References EPSILON, and U.

UINT lemga::randcdf (  REAL  r, const RVEC &  cdf )  [inline]

Definition at line 271 of file perceptron.cpp.

void lemga::update_wgt (  RVEC &  wgt, const RVEC &  dir, const RMAT &  X, const RVEC &  y )

Update the weight wgt along the direction dir. If necessary, the whole wgt will be negated. Definition at line 79 of file perceptron.cpp. References DOTPROD.

---