org.deeplearning4j.nn

Class BaseMultiLayerNetwork.Builder<E extends BaseMultiLayerNetwork>

java.lang.Object

org.deeplearning4j.nn.BaseMultiLayerNetwork.Builder<E>

Direct Known Subclasses:

CDBN.Builder, DBN.Builder, StackedDenoisingAutoEncoder.Builder

Enclosing class:

BaseMultiLayerNetwork

public static class BaseMultiLayerNetwork.Builder<E extends BaseMultiLayerNetwork>
extends Object

Field Summary

Fields

Modifier and Type	Field and Description
protected boolean	backProp
protected Class<? extends BaseMultiLayerNetwork>	clazz
protected boolean	shouldForceEpochs
protected Map<Integer,MatrixTransform>	weightTransforms

Constructor Summary

Constructors

Constructor and Description
BaseMultiLayerNetwork.Builder()

Method Summary

Methods

Modifier and Type	Method and Description
E	build()
E	buildEmpty()
BaseMultiLayerNetwork.Builder<E>	decodeNetwork(boolean decode) Whether the network is a decoder for an auto encoder
BaseMultiLayerNetwork.Builder<E>	disableBackProp() Disables back propagation
BaseMultiLayerNetwork.Builder<E>	forceEpochs() Forces use of number of epochs for training SGD style rather than conjugate gradient
BaseMultiLayerNetwork.Builder<E>	hiddenLayerSizes(int[] hiddenLayerSizes)
BaseMultiLayerNetwork.Builder<E>	numberOfInputs(int nIns)
BaseMultiLayerNetwork.Builder<E>	numberOfOutPuts(int nOuts)
BaseMultiLayerNetwork.Builder<E>	renderWeights(int everyN) Whether to plot weights or not
BaseMultiLayerNetwork.Builder<E>	transformWeightsAt(int layer, MatrixTransform transform) Transform the weights at the given layer
BaseMultiLayerNetwork.Builder<E>	transformWeightsAt(Map<Integer,MatrixTransform> transforms) A map of transformations for transforming the given layers
BaseMultiLayerNetwork.Builder<E>	useRegularization(boolean useRegularization) Use l2 reg
BaseMultiLayerNetwork.Builder<E>	withActivation(ActivationFunction activation) Pick an activation function, default is sigmoid
BaseMultiLayerNetwork.Builder<E>	withClazz(Class<? extends BaseMultiLayerNetwork> clazz)
BaseMultiLayerNetwork.Builder<E>	withDist(org.apache.commons.math3.distribution.RealDistribution dist) Probability distribution for generating weights
BaseMultiLayerNetwork.Builder<E>	withFanIn(Double fanIn)
BaseMultiLayerNetwork.Builder<E>	withInput(org.jblas.DoubleMatrix input)
BaseMultiLayerNetwork.Builder<E>	withL2(double l2) L2 coefficient
BaseMultiLayerNetwork.Builder<E>	withLabels(org.jblas.DoubleMatrix labels)
BaseMultiLayerNetwork.Builder<E>	withMomentum(double momentum) Specify momentum
BaseMultiLayerNetwork.Builder<E>	withRng(org.apache.commons.math3.random.RandomGenerator gen)
BaseMultiLayerNetwork.Builder<E>	withSparsity(double sparsity)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

clazz

protected Class<? extends BaseMultiLayerNetwork> clazz

weightTransforms

protected Map<Integer,MatrixTransform> weightTransforms

backProp

protected boolean backProp

shouldForceEpochs

protected boolean shouldForceEpochs

Constructor Detail

BaseMultiLayerNetwork.Builder

public BaseMultiLayerNetwork.Builder()

Method Detail

withSparsity

public BaseMultiLayerNetwork.Builder<E> withSparsity(double sparsity)

forceEpochs

public BaseMultiLayerNetwork.Builder<E> forceEpochs()

Forces use of number of epochs for training SGD style rather than conjugate gradient

Returns:

disableBackProp

public BaseMultiLayerNetwork.Builder<E> disableBackProp()

Disables back propagation

Returns:

transformWeightsAt

public BaseMultiLayerNetwork.Builder<E> transformWeightsAt(int layer,
                                                  MatrixTransform transform)

Transform the weights at the given layer

Parameters:

layer - the layer to transform

transform - the function used for transformation

Returns:

transformWeightsAt

public BaseMultiLayerNetwork.Builder<E> transformWeightsAt(Map<Integer,MatrixTransform> transforms)

A map of transformations for transforming the given layers

Parameters:

transforms -

Returns:

withDist

public BaseMultiLayerNetwork.Builder<E> withDist(org.apache.commons.math3.distribution.RealDistribution dist)

Probability distribution for generating weights

Parameters:

dist -

Returns:

withMomentum

public BaseMultiLayerNetwork.Builder<E> withMomentum(double momentum)

Specify momentum

Parameters:

momentum -

Returns:

useRegularization

public BaseMultiLayerNetwork.Builder<E> useRegularization(boolean useRegularization)

Use l2 reg

Parameters:

useRegularization -

Returns:

withL2

public BaseMultiLayerNetwork.Builder<E> withL2(double l2)

L2 coefficient

Parameters:

l2 -

Returns:

renderWeights

public BaseMultiLayerNetwork.Builder<E> renderWeights(int everyN)

Whether to plot weights or not

Parameters:

everyN -

Returns:

withFanIn

public BaseMultiLayerNetwork.Builder<E> withFanIn(Double fanIn)

withActivation

public BaseMultiLayerNetwork.Builder<E> withActivation(ActivationFunction activation)

Pick an activation function, default is sigmoid

Parameters:

activation -

Returns:

numberOfInputs

public BaseMultiLayerNetwork.Builder<E> numberOfInputs(int nIns)

decodeNetwork

public BaseMultiLayerNetwork.Builder<E> decodeNetwork(boolean decode)

Whether the network is a decoder for an auto encoder

Parameters:

decode -

Returns:

hiddenLayerSizes

public BaseMultiLayerNetwork.Builder<E> hiddenLayerSizes(int[] hiddenLayerSizes)

numberOfOutPuts

public BaseMultiLayerNetwork.Builder<E> numberOfOutPuts(int nOuts)

withRng

public BaseMultiLayerNetwork.Builder<E> withRng(org.apache.commons.math3.random.RandomGenerator gen)

withInput

public BaseMultiLayerNetwork.Builder<E> withInput(org.jblas.DoubleMatrix input)

withLabels

public BaseMultiLayerNetwork.Builder<E> withLabels(org.jblas.DoubleMatrix labels)

withClazz

public BaseMultiLayerNetwork.Builder<E> withClazz(Class<? extends BaseMultiLayerNetwork> clazz)

buildEmpty

public E buildEmpty()

build

public E build()