org.deeplearning4j.nn.gradient

Interface Gradient

All Superinterfaces:

Serializable

All Known Implementing Classes:

DefaultGradient

public interface Gradient
extends Serializable

Generic gradient

Author:

Adam Gibson

Method Summary

Modifier and Type	Method and Description
void	clear() Clear residual parameters (useful for returning a gradient and then clearing old objects)
org.nd4j.linalg.api.ndarray.INDArray	getGradientFor(String variable) The gradient for the given variable
org.nd4j.linalg.api.ndarray.INDArray	gradient() The full gradient as one flat vector
org.nd4j.linalg.api.ndarray.INDArray	gradient(List<String> order) The full gradient as one flat vector
Map<String,org.nd4j.linalg.api.ndarray.INDArray>	gradientForVariable() Gradient look up table
org.nd4j.linalg.api.ndarray.INDArray	setGradientFor(String variable, org.nd4j.linalg.api.ndarray.INDArray gradient) Update gradient for the given variable

Method Detail

gradientForVariable

Map<String,org.nd4j.linalg.api.ndarray.INDArray> gradientForVariable()

Gradient look up table

Returns:

the gradient look up table

gradient

org.nd4j.linalg.api.ndarray.INDArray gradient(List<String> order)

The full gradient as one flat vector

Returns:

gradient

org.nd4j.linalg.api.ndarray.INDArray gradient()

The full gradient as one flat vector

Returns:

clear

void clear()

Clear residual parameters (useful for returning a gradient and then clearing old objects)

getGradientFor

org.nd4j.linalg.api.ndarray.INDArray getGradientFor(String variable)

The gradient for the given variable

Parameters:

variable - the variable to get the gradient for

Returns:

the gradient for the given variable or null

setGradientFor

org.nd4j.linalg.api.ndarray.INDArray setGradientFor(String variable,
                                                    org.nd4j.linalg.api.ndarray.INDArray gradient)

Update gradient for the given variable

Parameters:

variable - the variable to get the gradient for

gradient - the gradient values

Returns:

the gradient for the given variable or null