Package deepnetts.net

Class FeedForwardNetwork

java.lang.Object

deepnetts.net.NeuralNetwork<BackpropagationTrainer>

deepnetts.net.FeedForwardNetwork

All Implemented Interfaces:

TrainerProvider<BackpropagationTrainer>, Serializable

public final class FeedForwardNetwork
extends NeuralNetwork<BackpropagationTrainer>

Feed forward neural network architecture, also known as Multi Layer Perceptron. It consists of a sequence of neural network layers deepnetts.net.layers trained by Back-propagation BackpropagationTrainer algorithm. As a minimum network must have input InputLayer and output layer OutputLayer. For non-trivial problems it will also need several hidden fully connected layers FullyConnectedLayer. The easiest and recommended way to create an instance of a neural network is by using FeedForwardNetwork.Builder This type of network can be used for both classification and regression tasks depending how it is configured.

For a quick explanation about essential principles behind feed forward neural networks see the tutorial From Basic Machine Learning to Deep Learning in 5 Minutes

For a quick overview of machine learning basics required to understand Feed Forward Network see Machine Learning Tutorial for Java Developers

See Also:

• FeedForwardNetwork.Builder
• NeuralNetwork
• BackpropagationTrainer
• Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static class	FeedForwardNetwork.Builder	Builder of a FeedForwardNetwork instance.

Method Summary

Modifier and Type	Method	Description
static FeedForwardNetwork.Builder	builder()	Returns a builder for the FeedForwardNetwork
float[]	getOutput(float[] inputs)	Deprecated.
float[]	predict(float... inputs)	Returns the network's prediction (outputs) for the given input.
void	setInput(float... inputs)	Sets network's input using given inputs and invokes the calculation of the network for the given input (forward pass).

Methods inherited from class deepnetts.net.NeuralNetwork

applyWeightChanges, backward, getInputLayer, getL1RegSum, getL2RegSum, getLabel, getLayerAt, getLayers, getLossFunction, getMode, getNormalizer, getOutput, getOutputLabel, getOutputLabels, getOutputLayer, getPreprocessing, getThreadPool, getTrainer, load, predict, save, setInput, setLabel, setLossFunction, setMode, setNormalizer, setOutputError, setOutputLabels, setPreprocessing, setTrainer, test, toString, train

Methods inherited from class java.lang.Object

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

Method Details

setInput

public void setInput(float... inputs)

Sets network's input using given inputs and invokes the calculation of the network for the given input (forward pass). This method is usually used only during the training. You don't have to use it before the call to predict(float...) since predict() method sets given inputs and returns calculated outputs.

Parameters:

inputs - array of inputs to the network given as array of float values

Throws:

IllegalArgumentException - if size of the input vector does not match the number of the inputs of a network

predict

public float[] predict(float... inputs)

Returns the network's prediction (outputs) for the given input.

Parameters:

inputs - array of inputs to the network given as array of float values

Returns:

model's prediction as network's output

Throws:

IllegalArgumentException - if size of the input vector does not match the number of the inputs of a network

getOutput

public float[] getOutput(float[] inputs)

Deprecated.

Returns network output for the given input. This method is deprecated and predict method should be used instead.

Parameters:

inputs -

Returns:

builder

public static FeedForwardNetwork.Builder builder()

Returns a builder for the FeedForwardNetwork

Returns:

builder instance