Ordinary Least Squares trainer using Gradient Descent optimization. More...

#include <OlsGDTrainer.h>

Inheritance diagram for txeo::OlsGDTrainer< T >:

Collaboration diagram for txeo::OlsGDTrainer< T >:

Public Member Functions
	OlsGDTrainer (const OlsGDTrainer &)=delete

	OlsGDTrainer (OlsGDTrainer &&)=delete

OlsGDTrainer &	operator= (const OlsGDTrainer &)=delete

OlsGDTrainer &	operator= (OlsGDTrainer &&)=delete

	~OlsGDTrainer ()=default

	OlsGDTrainer (txeo::DataTable< T > &&data)
	Construct a new OlsGD Trainer object from a data table.

	OlsGDTrainer (const txeo::DataTable< T > &data)

txeo::Tensor< T >	predict (const txeo::Tensor< T > &input) const override
	Makes predictions using learned weights.

T	learning_rate () const
	Gets current learning rate.

void	set_learning_rate (T learning_rate)
	Sets learning rate for gradient descent.

void	enable_variable_lr ()
	Enables adaptive learning rate adjustment (Barzilai-Borwein Method). When enabled, learning rate automatically reduces when loss plateaus. For the majority of the cases, convergence drastically increases.

void	disable_variable_lr ()
	Disables adaptive learning rate adjustment (Barzilai-Borwein Method)

const txeo::Matrix< T > &	weight_bias () const
	Gets weight/bias matrix related to the minimum loss during fit.

T	tolerance () const
	Gets convergence tolerance.

void	set_tolerance (const T &tolerance)
	Sets convergence tolerance.

bool	is_converged () const
	Checks convergence status.

T	min_loss () const
	Gets the minimum loss during training.

Public Member Functions inherited from txeo::Trainer< T >
	Trainer (const Trainer &)=delete

	Trainer (Trainer &&)=delete

Trainer &	operator= (const Trainer &)=delete

Trainer &	operator= (Trainer &&)=delete

virtual	~Trainer ()=default

	Trainer (txeo::DataTable< T > &&data, txeo::Logger &logger=txeo::LoggerConsole::instance())
	Construct a new Trainer object from a data table.

	Trainer (const txeo::DataTable< T > &data)

virtual void	fit (size_t epochs, txeo::LossFunc metric)
	Trains the model for specified number of epochs.

virtual void	fit (size_t epochs, txeo::LossFunc metric, size_t patience)
	Trains with early stopping based on validation performance.

virtual void	fit (size_t epochs, txeo::LossFunc metric, size_t patience, txeo::NormalizationType type)
	Trains with early stopping based on validation performance and feature normalization.

T	compute_test_loss (txeo::LossFunc metric) const
	Evaluates test data based on a specified metric.

bool	is_trained () const
	Checks if model has been trained.

const txeo::DataTable< T > &	data_table () const
	Returns the data table of this trainer.

void	enable_feature_norm (txeo::NormalizationType type)
	Enable feature normalization.

void	disable_feature_norm ()
	Disable feature normalization.

Additional Inherited Members
Protected Member Functions inherited from txeo::Trainer< T >
	Trainer ()=default

Protected Attributes inherited from txeo::Trainer< T >
bool	_is_trained {false}

bool	_is_early_stop {false}

size_t	_patience {0}

std::unique_ptr< txeo::DataTable< T > >	_data_table

txeo::Logger *	_logger {nullptr}

txeo::DataTableNorm< T >	_data_table_norm

bool	_is_norm_enabled {false}

Detailed Description

template<typename T>
requires (std::floating_point<T>)
class txeo::OlsGDTrainer< T >

Ordinary Least Squares trainer using Gradient Descent optimization.

Template Parameters

T	Floating-point type for calculations (float, double, etc.)

Implements linear regression training through gradient descent with:

Configurable learning rate
Convergence tolerance
Variable learning rate support (Barzilai-Borwein Method)
Weight/bias matrix access

Inherits from txeo::Trainer<T> and implements required virtual methods.

Implements algorithms based on paper: Algarte, R.D., "Tensor-Based Foundations of Ordinary Least Squares and Neural Network Regression Models" (https://arxiv.org/abs/2411.12873)

Example Usage:

// Create training data (y = 2x + 1)
txeo::Matrix<double> X({{1.0}, {2.0}, {3.0}}); // 3x1
txeo::Matrix<double> y({{3.0}, {5.0}, {7.0}}); // 3x1
 
OlsGDTrainer<double> trainer(X, y);
trainer.set_tolerance(1e-5);
 
// Train with early stopping
trainer.fit(1000, LossFunc::MSE, 10);
 
if(trainer.is_converged()) {
    auto weights = trainer.weight_bias();
    std::cout << "Model: y = " << weights(0,0) << "x + " << weights(1,0) << std::endl;
 
    // Make prediction
    txeo::Matrix<double> test_input(1,1,{4.0});
    auto prediction = trainer.predict(test_input);
    std::cout << "Prediction for x=4: " << prediction(0,0) << std::endl;
}

Definition at line 60 of file OlsGDTrainer.h.

Constructor & Destructor Documentation

◆ OlsGDTrainer() [1/4]

template<typename T >

txeo::OlsGDTrainer< T >::OlsGDTrainer ( const OlsGDTrainer< T > & )

delete

◆ OlsGDTrainer() [2/4]

template<typename T >

txeo::OlsGDTrainer< T >::OlsGDTrainer ( OlsGDTrainer< T > && )

delete

◆ ~OlsGDTrainer()

template<typename T >

txeo::OlsGDTrainer< T >::~OlsGDTrainer ( )

default

◆ OlsGDTrainer() [3/4]

template<typename T >

txeo::OlsGDTrainer< T >::OlsGDTrainer ( txeo::DataTable< T > && data )

inline

Construct a new OlsGD Trainer object from a data table.

Parameters

data	Training/Evaluation/Test data

Definition at line 73 of file OlsGDTrainer.h.

73: txeo::Trainer<T>(std::move(data)) {};

◆ OlsGDTrainer() [4/4]

template<typename T >

txeo::OlsGDTrainer< T >::OlsGDTrainer ( const txeo::DataTable< T > & data )

inline

Definition at line 75 of file OlsGDTrainer.h.

75: txeo::Trainer<T>(data) {};

Member Function Documentation

◆ disable_variable_lr()

template<typename T >

void txeo::OlsGDTrainer< T >::disable_variable_lr ( )

inline

Disables adaptive learning rate adjustment (Barzilai-Borwein Method)

Definition at line 113 of file OlsGDTrainer.h.

113{ _variable_lr = false; }

◆ enable_variable_lr()

template<typename T >

void txeo::OlsGDTrainer< T >::enable_variable_lr ( )

inline

Enables adaptive learning rate adjustment (Barzilai-Borwein Method). When enabled, learning rate automatically reduces when loss plateaus. For the majority of the cases, convergence drastically increases.

Definition at line 108 of file OlsGDTrainer.h.

108{ _variable_lr = true; }

◆ is_converged()

template<typename T >

bool txeo::OlsGDTrainer< T >::is_converged ( ) const

inline

Checks convergence status.

Returns: true if training converged before max epochs

Definition at line 143 of file OlsGDTrainer.h.

143{ return _is_converged; }

◆ learning_rate()

template<typename T >

T txeo::OlsGDTrainer< T >::learning_rate ( ) const

Gets current learning rate.

Returns: Current learning rate value

◆ min_loss()

template<typename T >

T txeo::OlsGDTrainer< T >::min_loss ( ) const

Gets the minimum loss during training.

Returns: Value of the minimum loss

◆ operator=() [1/2]

template<typename T >

OlsGDTrainer & txeo::OlsGDTrainer< T >::operator= ( const OlsGDTrainer< T > & )

delete

◆ operator=() [2/2]

template<typename T >

OlsGDTrainer & txeo::OlsGDTrainer< T >::operator= ( OlsGDTrainer< T > && )

delete

◆ predict()

template<typename T >

txeo::Tensor< T > txeo::OlsGDTrainer< T >::predict ( const txeo::Tensor< T > & input ) const

overridevirtual

Makes predictions using learned weights.

Parameters

input Feature matrix (shape: [samples, features])

Returns: Prediction matrix (shape: [samples, outputs])

Exceptions

OlsGDTrainerError

Implements txeo::Trainer< T >.

◆ set_learning_rate()

template<typename T >

void txeo::OlsGDTrainer< T >::set_learning_rate ( T learning_rate )

Sets learning rate for gradient descent.

Parameters

learning_rate Must be > 0

Exceptions

OlsGDTrainerError for invalid values

◆ set_tolerance()

template<typename T >

void txeo::OlsGDTrainer< T >::set_tolerance ( const T & tolerance )

Sets convergence tolerance.

Parameters

tolerance Minimum loss difference to consider converged (>0)

◆ tolerance()

template<typename T >

T txeo::OlsGDTrainer< T >::tolerance ( ) const

inline

Gets convergence tolerance.

Returns: Current tolerance value

Definition at line 129 of file OlsGDTrainer.h.

129{ return _tolerance; }

◆ weight_bias()

template<typename T >

const txeo::Matrix< T > & txeo::OlsGDTrainer< T >::weight_bias ( ) const

Gets weight/bias matrix related to the minimum loss during fit.

Returns: Matrix containing model parameters (shape: [features+1, outputs])

Exceptions

OlsGDTrainerError

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

include/txeo/OlsGDTrainer.h

Public Member Functions

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ OlsGDTrainer() [1/4]

◆ OlsGDTrainer() [2/4]

◆ ~OlsGDTrainer()

◆ OlsGDTrainer() [3/4]

◆ OlsGDTrainer() [4/4]

Member Function Documentation

◆ disable_variable_lr()

◆ enable_variable_lr()

◆ is_converged()

◆ learning_rate()

◆ min_loss()

◆ operator=() [1/2]

◆ operator=() [2/2]

◆ predict()

◆ set_learning_rate()

◆ set_tolerance()

◆ tolerance()

◆ weight_bias()