NearOCR

Class NetStack

java.lang.Object

NearOCR.NetStack

public class NetStack
extends java.lang.Object

The net stack holds all nets that are to be run durin the OCR process. All of the nets are run from the same input, but produce outputs which are serialised into a large array before analysis. This means that the symbols must hold enough weights not just to parse with one net but for all that are loaded.

Version:

0.1

Author:

Stephen Wattam

Constructor Summary

Constructors

Constructor and Description
NetStack() Creates an empty net stack.

Method Summary

Methods

Modifier and Type	Method and Description
void	appendNet(NeuralNet tempNet) Appends a net to the end of the stack.
int	countNets() Returns the number of nets that are currently loaded.
int	countOutput() Returns the number of outputs this OCR system will generate.
int	getInputCount() Returns the number of input neurons that this OCR manager is currently set to using.
NeuralNet	getNet(int p_index) Returns a net by index from this OCR manager.
NeuralNet[]	getNets() Returns an array of all nets held in this OCR manager, ordered as they are to have their outputs serialised.
void	moveNet(int p_indexFrom, int p_indexTo) Moves a net in the list of nets.
void	removeNet(NeuralNet p_net) Removes a net from the internal list.
void	setInput(int p_number) Attempts to adjust the number of input neurons.

Methods inherited from class java.lang.Object

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

Constructor Detail

NetStack

public NetStack()

Creates an empty net stack.

Method Detail

appendNet

public void appendNet(NeuralNet tempNet)
               throws ValueOutOfBoundsException

Appends a net to the end of the stack.

Parameters:

tempNet - The net to append.

Throws:

ValueOutOfBoundsException - in the event that the net given has the wrong amount of input neurons.

setInput

public void setInput(int p_number)
              throws ValueOutOfBoundsException

Attempts to adjust the number of input neurons. This number is fixed across all currently loaded nets and regulated by the OCR manager.

Parameters:

p_number - The number of inputs to attempt to set

Throws:

ValueOutOfBoundsException - in the event that the desired number of inputs is incompatible with currently loaded nets.

getInputCount

public int getInputCount()

Returns the number of input neurons that this OCR manager is currently set to using. This is fixed and based on all nets that are currently loaded. It can be set but must equal the same number of input neurons on every single loaded net.

Returns:

The number of input neurons. All nets loaded must match this.

moveNet

public void moveNet(int p_indexFrom,
           int p_indexTo)
             throws java.lang.IndexOutOfBoundsException

Moves a net in the list of nets.

The order in which nets are loaded is vital, as all nets' outputs are serialised before poat-parsing and symbol comparison. Note that the removal of the net from the vector shuffles everything up - indexTo will have to be one less than one would normally assume.

Parameters:

p_indexFrom - The index to request a net from

p_indexTo - The index to move the requested net to

Throws:

java.lang.IndexOutOfBoundsException - in the event that either index does not exist

getNet

public NeuralNet getNet(int p_index)
                 throws java.lang.IndexOutOfBoundsException

Returns a net by index from this OCR manager.

Parameters:

p_index - The index of the net to return

Throws:

java.lang.IndexOutOfBoundsException - Thrown in the event that the index requested contains no net or does not exist

removeNet

public void removeNet(NeuralNet p_net)
               throws ItemNotFoundException

Removes a net from the internal list. This allows for removal of nets without knowing index numbers.

Parameters:

p_net - the net to remove

Throws:

ItemNotFoundException - In the event that the net is not in the list

getNets

public NeuralNet[] getNets()

Returns an array of all nets held in this OCR manager, ordered as they are to have their outputs serialised. From this one can get enough data to prune, re-order, manipulate or replace any nets from outside the confines of the OCR manager. Severe net restructuring may break the OCR process, and it is generally safer to use this for visualisation and edge operations only.

Returns:

An array of all neural nets held in this OCR manager.

countNets

public int countNets()

Returns the number of nets that are currently loaded.

Returns:

The number of nets currently in the stack.

countOutput

public int countOutput()

Returns the number of outputs this OCR system will generate. This is equal to the sum of all outputs from all loaded nets and must be equal to the number of weighted nodes in each symbol of the symbol table.

Returns:

The number of outputs