CNum::Model::Loss Namespace Reference

Loss functions. More...

Classes
struct	LossProfile
	The loss, gradient, and hessian functions associated with a loss function. More...

Typedefs
using	GHFunction = std::function< double(double, double) >
using	LossFunction

Functions
double	MSE_loss (const CNum::DataStructs::Matrix< double > &y, const CNum::DataStructs::Matrix< double > &y_pred)
	Mean squared error.
double	MSE_gradient (double y, double y_pred)
	Calculate the gradient value of the mean squared error.
double	MSE_hessian (double y, double y_pred)
	Calculate the hessian (second order derivative) value of the mean squared error.
double	RMSE_loss (const CNum::DataStructs::Matrix< double > &y, const CNum::DataStructs::Matrix< double > &y_pred)
	Root mean squared error.
double	binary_crossentropy_loss (const CNum::DataStructs::Matrix< double > &y, const CNum::DataStructs::Matrix< double > &y_pred)
	Binary crossentropy (log loss).
double	binary_crossentropy_gradient (double y, double y_pred)
	Calculate the gradient of the log loss.
double	binary_crossentropy_hessian (double y, double y_pred)
	Calculate the hessian (second order derivative) of the log loss.
void	get_gradients_hessians (const CNum::DataStructs::Matrix< double > &y, const CNum::DataStructs::Matrix< double > &y_pred, arena_view_t &g_out, arena_view_t &h_out, const arena_view_t &position_array, GHFunction &grad_func, GHFunction &hess_func)
	Get the Gradients and Hessians of a Matrix.
double	get_loss (const CNum::DataStructs::Matrix< double > &y, const CNum::DataStructs::Matrix< double > &y_pred, LossFunction &loss_func)
	Get the loss of a matrix of values.
LossProfile	get_loss_profile (::std::string loss)
	Get the LossProfile associated with a string (i.e. "MSE" -> mean squared error function).

Detailed Description

Loss functions.

Typedef Documentation

GHFunction

using CNum::Model::Loss::GHFunction = std::function< double(double, double) >

LossFunction

using CNum::Model::Loss::LossFunction

Initial value:

 std::function< double(const CNum::DataStructs::Matrix<double> &,
                         const CNum::DataStructs::Matrix<double> &) >

Function Documentation

binary_crossentropy_gradient()

double CNum::Model::Loss::binary_crossentropy_gradient	(	double	y,
		double	y_pred )

Calculate the gradient of the log loss.

Parameters

y	A true y sample
y_pred	A predicted y sample

Returns

The gradient value

binary_crossentropy_hessian()

double CNum::Model::Loss::binary_crossentropy_hessian	(	double	y,
		double	y_pred )

Calculate the hessian (second order derivative) of the log loss.

Parameters

y	A true y sample
y_pred	A predicted y sample

Returns

The hessian value

binary_crossentropy_loss()

double CNum::Model::Loss::binary_crossentropy_loss	(	const CNum::DataStructs::Matrix< double > &	y,
		const CNum::DataStructs::Matrix< double > &	y_pred )

Binary crossentropy (log loss).

Parameters

y	List of true y values (shape=(n,1))
y_pred	List of predicted values (shape=(n,1))

Returns

The loss

get_gradients_hessians()

void CNum::Model::Loss::get_gradients_hessians	(	const CNum::DataStructs::Matrix< double > &	y,
		const CNum::DataStructs::Matrix< double > &	y_pred,
		arena_view_t &	g_out,
		arena_view_t &	h_out,
		const arena_view_t &	position_array,
		GHFunction &	grad_func,
		GHFunction &	hess_func )

Get the Gradients and Hessians of a Matrix.

Parameters

y	List of true y values (shape=(n,1))
y_pred	List of predicted values (shape=(n,1))
g_out	The arena_view_t to output the gradient values to
h_out	The arena_view_t to output the hessian values to
position_array	The arena_view_t containing the partitioned indeces of the data (partitions are for keeping track of which samples tree nodes have to work with)
grad_func	The GHFunction for the gradient of the loss
hess_func	The GHFunction for the hessian of the loss

get_loss()

double CNum::Model::Loss::get_loss	(	const CNum::DataStructs::Matrix< double > &	y,
		const CNum::DataStructs::Matrix< double > &	y_pred,
		LossFunction &	loss_func )

Get the loss of a matrix of values.

Parameters

y	List of true y values (shape=(n,1))
y_pred	List of predicted values (shape=(n,1))
loss_func	The LossFunction to use

Returns

The loss

get_loss_profile()

LossProfile CNum::Model::Loss::get_loss_profile

(

::std::string

loss

)

Get the LossProfile associated with a string (i.e. "MSE" -> mean squared error function).

Parameters

loss	The name of the loss function

Returns

The LossProfile

MSE_gradient()

double CNum::Model::Loss::MSE_gradient	(	double	y,
		double	y_pred )

Calculate the gradient value of the mean squared error.

Parameters

y	A true y sample
y_pred	A predicted y sample

Returns

The gradient value

MSE_hessian()

double CNum::Model::Loss::MSE_hessian	(	double	y,
		double	y_pred )

Calculate the hessian (second order derivative) value of the mean squared error.

Parameters

y	A true y sample
y_pred	A predicted y sample

Returns

The hessian value

MSE_loss()

double CNum::Model::Loss::MSE_loss	(	const CNum::DataStructs::Matrix< double > &	y,
		const CNum::DataStructs::Matrix< double > &	y_pred )

Mean squared error.

Parameters

y	List of true y values (shape=(n,1))
y_pred	List of predicted values (shape=(n,1))

Returns

The loss

RMSE_loss()

double CNum::Model::Loss::RMSE_loss	(	const CNum::DataStructs::Matrix< double > &	y,
		const CNum::DataStructs::Matrix< double > &	y_pred )

Root mean squared error.

Parameters

y	List of true y values (shape=(n,1))
y_pred	List of predicted values (shape=(n,1))

Returns

The loss