dl::loss::HuberLoss< T > Class Template Reference

Huber Loss with autograd support. More...

#include <losses.hpp>

Inheritance diagram for dl::loss::HuberLoss< T >:

Collaboration diagram for dl::loss::HuberLoss< T >:

Public Member Functions
	HuberLoss (T delta=1.0, const std::string &reduction="mean")
	Constructor.
Variable< T >	forward (const Variable< T > &predictions, const Variable< T > &targets) override
	Forward pass: compute Huber loss.
Public Member Functions inherited from dl::loss::AutogradLoss< T >
virtual	~AutogradLoss ()=default
Variable< T >	operator() (const Variable< T > &predictions, const Variable< T > &targets)
	Convenience operator for computing loss.

Detailed Description

template<typename T>
class dl::loss::HuberLoss< T >

Huber Loss with autograd support.

Combines MSE and MAE for robustness to outliers.

Huber(y_pred, y_true) = { 0.5 * (y_pred - y_true)² if |y_pred - y_true| <= delta delta * |y_pred - y_true| - 0.5 * delta² otherwise }

Definition at line 207 of file losses.hpp.

Constructor & Destructor Documentation

HuberLoss()

template<typename T >

dl::loss::HuberLoss< T >::HuberLoss ( T  delta = 1.0, const std::string &  reduction = "mean"  )

inlineexplicit

Constructor.

Parameters

delta	Threshold for switching between MSE and MAE
reduction	Type of reduction ('mean', 'sum', 'none')

Definition at line 214 of file losses.hpp.

Member Function Documentation

forward()

template<typename T >

Variable< T > dl::loss::HuberLoss< T >::forward ( const Variable< T > &  predictions, const Variable< T > &  targets  )

overridevirtual

Forward pass: compute Huber loss.

Parameters

predictions	Predicted values
targets	Target values

Returns

Huber loss

Implements dl::loss::AutogradLoss< T >.

Definition at line 166 of file losses.cpp.

---

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

• include/loss/losses.hpp
• src/loss/losses.cpp