CNum::Model::Tree::GBModel< TreeType > Class Template Reference

A gradient-boosting model for any child of the TreeBooster class. More...

#include <GBModel.h>

Public Member Functions
	GBModel (std::string lt="MSE", int n_learners=200, double lr=0.1, double ss=0.25, int md=5, int ms=3, SplitAlg sa=HIST, ::std::string activation_func="", double weight_decay=0.0, double rl=1.0, double gamma=0.0, SubsampleFunction ssf=default_subsample)
	Overloaded default constructor.
	GBModel (::CNum::Model::Loss::LossProfile loss_profile, int n_learners, double lr, double ss, int md, int ms, SplitAlg sa, ::CNum::Model::Activation::ActivationFunc activation_func, double weight_decay, double rl, double gamma, SubsampleFunction ssf=default_subsample)
	Parameterized constructor.
	GBModel (const GBModel &other) noexcept
	Copy constructor.
GBModel< TreeType > &	operator= (const GBModel &other) noexcept
	Copy assignment.
	GBModel (GBModel &&other) noexcept
	Move constructor.
GBModel< TreeType > &	operator= (GBModel &&other) noexcept
	Move assignment.
	~GBModel ()
	Destructor.
void	fit (::CNum::DataStructs::Matrix< double > &X, ::CNum::DataStructs::Matrix< double > &y, bool verbose=true)
	Train the model.
::CNum::DataStructs::Matrix< double >	predict (::CNum::DataStructs::Matrix< double > &data)
	Inference (making predictions).
void	save_model (std::string path)
	Save Model to JSON encoded ".cmod" file.

Static Public Member Functions
static GBModel< TreeType >	load_model (std::string path)
	Load Model from JSON encoded ".cmod" file.

Detailed Description

template<typename TreeType>
class CNum::Model::Tree::GBModel< TreeType >

A gradient-boosting model for any child of the TreeBooster class.

Template Parameters

TreeType

A child of the TreeBooster class

The GBModel is a robust gradient-boosting model that can be used for both regression and classification on tabular data; simply set the loss and activation parameters of the constructor accordingly (see Loss and Activation classes) for available Loss and Activation functions.

Constructor & Destructor Documentation

GBModel() [1/4]

template<typename TreeType>

CNum::Model::Tree::GBModel< TreeType >::GBModel	(	std::string	lt = "MSE",
		int	n_learners = 200,
		double	lr = 0.1,
		double	ss = 0.25,
		int	md = 5,
		int	ms = 3,
		SplitAlg	sa = HIST,
		::std::string	activation_func = "",
		double	weight_decay = 0.0,
		double	rl = 1.0,
		double	gamma = 0.0,
		SubsampleFunction	ssf = default_subsample )

Overloaded default constructor.

Parameters

lt	Loss Type (the name of the loss function)
n_learners	The number of TreeBooster learners to use
lr	The learning rate
ss	The amount of the dataset to sample for each tree
md	The max depth of each tree
ms	The minimum number of samples required to continue building a tree
sa	Split Algorithm; The method used to find the best splits at each node (GREEDY or HIST)
activation_func	The name of the activation function to apply after making predictions
weight_decay	A threshold used to detirmine whether a split is worth considering
rl	Reg Lambda; A regularization parameter used in gain calculations
gamma	A regularization parameter used in gain calculations
ssf	The function used to take subsamples of the data that a TreeBooster is trained on

In the tree building process a split is only taken if the sum of the hessians in the resultant data partitions are both greater than weight_decay.

See TreeBoosterNode to understand how the regularization parameters are used

GBModel() [2/4]

template<typename TreeType>

CNum::Model::Tree::GBModel< TreeType >::GBModel	(	::CNum::Model::Loss::LossProfile	loss_profile,
		int	n_learners,
		double	lr,
		double	ss,
		int	md,
		int	ms,
		SplitAlg	sa,
		::CNum::Model::Activation::ActivationFunc	activation_func,
		double	weight_decay,
		double	rl,
		double	gamma,
		SubsampleFunction	ssf = default_subsample )

Parameterized constructor.

Parameters

loss_profile	The LossProfile for the loss function
n_learners	The number of TreeBooster learners to use
lr	The learning rate
ss	The amount of the dataset to sample for each tree
md	The max depth of each tree
ms	The minimum number of samples required to continue building a tree
sa	Split Algorithm; The method used to find the best splits at each node (GREEDY or HIST)
activation_func	The activation function to apply after making predictions
weight_decay	A threshold used to detirmine whether a split is worth considering
rl	Reg Lambda; A regularization parameter used in gain calculations
gamma	A regularization parameter used in gain calculations
ssf	The function used to take subsamples of the data that a TreeBooster is trained on

In the tree building process a split is only taken if the sum of the hessians in the resultant data partitions are both greater than weight_decay.

See TreeBoosterNode to understand how the regularization parameters are used

GBModel() [3/4]

template<typename TreeType>

CNum::Model::Tree::GBModel< TreeType >::GBModel ( const GBModel< TreeType > & other )

noexcept

Copy constructor.

GBModel() [4/4]

template<typename TreeType>

CNum::Model::Tree::GBModel< TreeType >::GBModel ( GBModel< TreeType > && other )

noexcept

Move constructor.

~GBModel()

template<typename TreeType>

CNum::Model::Tree::GBModel< TreeType >::~GBModel

(

)

Destructor.

Member Function Documentation

fit()

template<typename TreeType>

void CNum::Model::Tree::GBModel< TreeType >::fit	(	::CNum::DataStructs::Matrix< double > &	X,
		::CNum::DataStructs::Matrix< double > &	y,
		bool	verbose = true )

Train the model.

Parameters

X	The tabular data used to train the GBModel
y	The labels for the data (the intended output of the model)

load_model()

template<typename TreeType>

GBModel< TreeType > CNum::Model::Tree::GBModel< TreeType >::load_model ( std::string path )

static

Load Model from JSON encoded ".cmod" file.

Parameters

path	The path to the ".cmod" file

Returns

The GBModel

operator=() [1/2]

template<typename TreeType>

GBModel< TreeType > & CNum::Model::Tree::GBModel< TreeType >::operator= ( const GBModel< TreeType > & other )

noexcept

Copy assignment.

operator=() [2/2]

template<typename TreeType>

GBModel< TreeType > & CNum::Model::Tree::GBModel< TreeType >::operator= ( GBModel< TreeType > && other )

noexcept

Move assignment.

predict()

template<typename TreeType>

CNum::DataStructs::Matrix< double > CNum::Model::Tree::GBModel< TreeType >::predict

(

::CNum::DataStructs::Matrix< double > &

data

)

Inference (making predictions).

Parameters

The	data to make predictions on

Returns

The predictions

save_model()

template<typename TreeType>

void CNum::Model::Tree::GBModel< TreeType >::save_model

(

std::string

path

)

Save Model to JSON encoded ".cmod" file.

Parameters

path	The path to save the file to

---

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

• include/CNum/Model/Tree/GBModel.h