deepnetts.net.train.opt

Interface Optimizer

All Known Implementing Classes:

AdaDeltaOptimizer, AdaGradOptimizer, AdamOptimizer, MomentumOptimizer, RmsPropOptimizer, SgdOptimizer

public interface Optimizer

Optimization technique to tune network's weights parameters used by training algorithm.

Field Summary

Fields

Modifier and Type	Field and Description
static int	COL_IDX
static int	DEPTH_IDX
static float	EPS Smoothing term to prevent division by zero if sqr grad sum becomes zero 1e-8 should be also tried https://d2l.ai/chapter_optimization/adagrad.html 1e-6 The value to use is 1e-6, 1e-8, Keras uses 1e-7 for adam
static int	FOURTH_DIM_IDX
static int	ROW_IDX

Method Summary

Modifier and Type	Method and Description
float	calculateDeltaBias(float gradient, int idx)
float	calculateDeltaWeight(float gradient, int... index)
static Optimizer	create(OptimizerType type, AbstractLayer layer) Factory method to create different types of optimizers
void	setLearningRate(float learningRate)

Field Detail

EPS

static final float EPS

Smoothing term to prevent division by zero if sqr grad sum becomes zero 1e-8 should be also tried https://d2l.ai/chapter_optimization/adagrad.html 1e-6 The value to use is 1e-6, 1e-8, Keras uses 1e-7 for adam

See Also:

Constant Field Values

ROW_IDX

static final int ROW_IDX

See Also:

Constant Field Values

COL_IDX

static final int COL_IDX

See Also:

Constant Field Values

DEPTH_IDX

static final int DEPTH_IDX

See Also:

Constant Field Values

FOURTH_DIM_IDX

static final int FOURTH_DIM_IDX

See Also:

Constant Field Values

Method Detail

calculateDeltaWeight

float calculateDeltaWeight(float gradient,
                           int... index)

calculateDeltaBias

float calculateDeltaBias(float gradient,
                         int idx)

setLearningRate

void setLearningRate(float learningRate)

create

static Optimizer create(OptimizerType type,
                        AbstractLayer layer)

Factory method to create different types of optimizers

Parameters:

type -

layer -

Returns: