Package engine.neural.networks

Class Feedforward

java.lang.Object

engine.neural.networks.Feedforward

All Implemented Interfaces:

Network, Reconstructible

public class Feedforward
extends Object
implements Network

"Simple" multi-layer neural network

Field Summary

Fields

Modifier and Type	Field	Description
private final Activation	activation	Variable that tracks the activation function being used
private Vector[]	activations
private Vector[]	biases
private final Cost	cost	Variable that tracks the cost function being used
private float	currCost	Variable that keeps track of the last cost of the network
private final Initializer	initializer	Variable that tracks the initializer being used
private final int[]	networkSizes	Keeps track of the network geometry
private final Optimizer	optimizer	Variable that tracks the optimizer being used
private Vector[]	preActivations
private Matrix[]	weights

Constructor Summary

Constructors

Constructor	Description
Feedforward(int[] networkSizes, Initializer initializer, Activation activation, Cost cost, Optimizer optimizer)	Constructs a new multilayer neural network
Feedforward(String name)

Method Summary

Modifier and Type	Method	Description
private void	backward(Vector output, Vector expected)
private Vector	forward(Vector input)
float	getCost()
Vector	predict(Vector input)
private String	printMatrices(Matrix[] m)
void	saveNetwork(String name)
private Object	search(String str, Scanner scan)	Function for finding and initializing network params
void	step(Vector input, Vector expected)
private void	textToMatrix(Scanner scan, Matrix[] m)
void	train(Vector[] inputs, Vector[] expected, int epochs)

Methods inherited from class java.lang.Object

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

Methods inherited from interface engine.neural.Reconstructible

getClassInfo

Field Details

networkSizes

private final int[] networkSizes

Keeps track of the network geometry

biases

private Vector[] biases

activations

private Vector[] activations

preActivations

private Vector[] preActivations

weights

private Matrix[] weights

initializer

private final Initializer initializer

Variable that tracks the initializer being used

activation

private final Activation activation

Variable that tracks the activation function being used

cost

private final Cost cost

Variable that tracks the cost function being used

optimizer

private final Optimizer optimizer

Variable that tracks the optimizer being used

currCost

private float currCost

Variable that keeps track of the last cost of the network

Constructor Details

Feedforward

public Feedforward(int[] networkSizes,
 Initializer initializer,
 Activation activation,
 Cost cost,
 Optimizer optimizer)

Constructs a new multilayer neural network

Parameters:

networkSizes - The size of each layer of the network, for instance, [2, 4, 4, 1] has an input of size 2, 2 hidden layers of size 4, and an output of size 1

initializer - The specific way you want to populate the weights at creation

activation - The specific way you want neurons to fire

cost - The specific way you want to show how wrong the network is

optimizer - The specific way you want to push the network towards minima

Feedforward

public Feedforward(String name)
            throws Exception

Throws:

Exception

Method Details

textToMatrix

private void textToMatrix(Scanner scan,
 Matrix[] m)

search

private Object search(String str,
 Scanner scan)
               throws Exception

Function for finding and initializing network params

Parameters:

str - Header information

scan - Scanner object

Returns:

Class within text file

Throws:

Exception

forward

private Vector forward(Vector input)

backward

private void backward(Vector output,
 Vector expected)

train

public void train(Vector[] inputs,
 Vector[] expected,
 int epochs)

Specified by:

train in interface Network

step

public void step(Vector input,
 Vector expected)

Specified by:

step in interface Network

predict

public Vector predict(Vector input)

Specified by:

predict in interface Network

getCost

public float getCost()

Specified by:

getCost in interface Network

saveNetwork

public void saveNetwork(String name)

Specified by:

saveNetwork in interface Network

printMatrices

private String printMatrices(Matrix[] m)