TrainOptions Struct Reference

Contains a set of training options. More...

#include <SimdNeural.hpp>

Public Types
enum	InitType { Xavier }
enum	LossType {   Mse ,   CrossEntropy ,   CrossEntropyMulticlass }
enum	UpdateType { AdaptiveGradient }

Public Member Functions
	TrainOptions ()
	Default constructor.

Data Fields
InitType	initType
	Method to initialize weights.
LossType	lossType
	Loss function type.
UpdateType	updateType
	Weights' update type.
size_t	threadNumber
	Number of threads used to train. Use -1 to auto detect thread number.
size_t	epochStart
	Start epoch. It is used to continue training process.
size_t	epochFinish
	Finish epoch. Describes total epoch number.
size_t	batchSize
	A batch size.
float	alpha
	Describes training speed.
float	epsilon
	Used to prevent division by zero.
bool	shuffle
	A flag to shuffle training set.

Detailed Description

Contains a set of training options.

Member Enumeration Documentation

InitType

enum InitType

Describes method to initialize weights of neural network.

Enumerator
Xavier	Use fan-in and fan-out for scaling Xavier Glorot, Yoshua Bengio. "Understanding the difficulty of training deep feedforward neural networks" Proc. AISTATS 10, May 2010, vol.9, pp249-256

LossType

enum LossType

Describes loss function.

Enumerator
Mse	Mean-Squared-Error loss function for regression.
CrossEntropy	Cross-entropy loss function for (multiple independent) binary classifications.
CrossEntropyMulticlass	Cross-entropy loss function for multi-class classification.

UpdateType

enum UpdateType

Method of weights' updating.

Enumerator
AdaptiveGradient	Adaptive gradients method. J Duchi, E Hazan and Y Singer, "Adaptive subgradient methods for online learning and stochastic optimization" The Journal of Machine Learning Research, pages 2121-2159, 2011. Note See SimdNeuralAdaptiveGradientUpdate.