Package deepnetts.net

Class FeedForwardNetwork.Builder

java.lang.Object

deepnetts.net.FeedForwardNetwork.Builder

Enclosing class:

FeedForwardNetwork

public static class FeedForwardNetwork.Builder
extends Object

Builder of a FeedForwardNetwork instance. Provides methods for setting all the components of a feed forward neural network and performs basic validation of settings in order to prevent illegal configuration.

See Also:

• FeedForwardNetwork

Constructor Summary

Constructors

Constructor	Description
Builder()

Method Summary

Modifier and Type	Method	Description
FeedForwardNetwork.Builder	addFullyConnectedLayer(int layerWidth)	Adds to the network a fully connected layer with specified width and Relu activation function by default.
FeedForwardNetwork.Builder	addFullyConnectedLayer(int layerWidth, ActivationType activationType)	Adds fully connected addLayer with specified width and activation function to the network.
FeedForwardNetwork.Builder	addHiddenFullyConnectedLayers(int... layerWidths)	Adds to the network several hidden fully connected layers with specified widths and default hidden activation function by default.
FeedForwardNetwork.Builder	addHiddenFullyConnectedLayers(ActivationType activationType, int... layerWidths)	Adds fully connected hidden layers with widths given in layerWidths param and given activation function type.
FeedForwardNetwork.Builder	addInputLayer(int layerWidth)	Adds input layer with the specified layerWidth (number of inputs) to the network.
FeedForwardNetwork.Builder	addInputLayer(int layerWidth, int batchSize)
FeedForwardNetwork.Builder	addLayer(AbstractLayer layer)	Adds custom layer to this network (which inherits from AbstractLayer)
FeedForwardNetwork.Builder	addOutputLayer(int width, ActivationType activationType)	Adds output layer to the neural network with specified width (number of outputs) and activation function type.
FeedForwardNetwork	build()	Builds an instance of FeedForwardNetwork with settings specified in this builder.
FeedForwardNetwork.Builder	hiddenActivationFunction(ActivationType activationType)	Sets default type of the activation function to use for all hidden layers in the network.
FeedForwardNetwork.Builder	lossFunction(LossType lossType)	Sets loss function to be used by created neural network.
FeedForwardNetwork.Builder	randomSeed(long seed)	Initializes random number generator with the specified seed in order to get same random number sequences used for weights initialization.

Methods inherited from class java.lang.Object

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

Constructor Details

Builder

public Builder()

Method Details

addInputLayer

public FeedForwardNetwork.Builder addInputLayer(int layerWidth)

Adds input layer with the specified layerWidth (number of inputs) to the network. Input layer is always the first layer in the network which accepts the external input.

Parameters:

layerWidth - width of the input layer that corresponds to the number of network's inputs

Returns:

builder instance

addInputLayer

public FeedForwardNetwork.Builder addInputLayer(int layerWidth,
 int batchSize)

addFullyConnectedLayer

public FeedForwardNetwork.Builder addFullyConnectedLayer(int layerWidth)

Adds to the network a fully connected layer with specified width and Relu activation function by default.

Parameters:

layerWidth - width of the layer / number of neurons

Returns:

builder instance

See Also:

• FullyConnectedLayer

addHiddenFullyConnectedLayers

public FeedForwardNetwork.Builder addHiddenFullyConnectedLayers(int... layerWidths)

Adds to the network several hidden fully connected layers with specified widths and default hidden activation function by default.

Parameters:

layerWidths - an array with widths for hidden fully connected layers.

Returns:

builder instance

See Also:

• FullyConnectedLayer

addFullyConnectedLayer

public FeedForwardNetwork.Builder addFullyConnectedLayer(int layerWidth,
 ActivationType activationType)

Adds fully connected addLayer with specified width and activation function to the network.

Parameters:

layerWidth - width of the layer to add

activationType - type of the activation function for layer to add

Returns:

builder instance

See Also:

• ActivationFunction

addHiddenFullyConnectedLayers

public FeedForwardNetwork.Builder addHiddenFullyConnectedLayers(ActivationType activationType,
 int... layerWidths)

Adds fully connected hidden layers with widths given in layerWidths param and given activation function type.

Parameters:

activationType - type of activation function in hidden layers

layerWidths - widths of the hidden layers

Returns:

addLayer

public FeedForwardNetwork.Builder addLayer(AbstractLayer layer)

Adds custom layer to this network (which inherits from AbstractLayer)

Parameters:

layer -

Returns:

builder instance

addOutputLayer

public FeedForwardNetwork.Builder addOutputLayer(int width,
 ActivationType activationType)

Adds output layer to the neural network with specified width (number of outputs) and activation function type.

Parameters:

width - layer with which corresponds to number of network's outputs

activationType - type of the activation function to use in output layer

Returns:

builder instance

hiddenActivationFunction

public FeedForwardNetwork.Builder hiddenActivationFunction(ActivationType activationType)

Sets default type of the activation function to use for all hidden layers in the network.

Parameters:

activationType - type of activation function

Returns:

instance of the current builder

See Also:

• ActivationType

lossFunction

public FeedForwardNetwork.Builder lossFunction(LossType lossType)

Sets loss function to be used by created neural network. Loss function calculates the network's error during the training as a difference between actual and target output provided in training set.

Parameters:

lossType - type of a loss function

Returns:

instance of the current builder

randomSeed

public FeedForwardNetwork.Builder randomSeed(long seed)

Initializes random number generator with the specified seed in order to get same random number sequences used for weights initialization. Specifying this value enables getting same/repeatable initialization.

Parameters:

seed -

Returns:

instance of the current builder

build

public FeedForwardNetwork build()

Builds an instance of FeedForwardNetwork with settings specified in this builder.

Returns:

an instance of the FeedForwardNetwork created by this builder