class FreeNetwork A generic and extensible neural network. More... Definition #include <termination.h> Inherits ANNetwork, Attributed (common) Inherited by RegulationNet List of all Methods

Annotated List Files Globals Hierarchy Index

Public Members

• FreeNetwork (const char* description=NULL)
• FreeNetwork (const FreeNetwork& orig)
• virtual ~FreeNetwork ()
• void makeUnits (const char* topology)
• virtual void make (const char* top)
• virtual void copy (const FreeNetwork& fnet, bool onlyWeights=false)
• void empty ()
• void removeUnit (int i)
• void connectFfw (const PackTable<int>& conns)
• void connectFullFfw ()
• void connectFull ()
• void setUnitTemplate (FreeNeuron* t)
• virtual void copyFreeNet (const FreeNetwork& orig, bool onlyWeights=false)
• virtual void init (double r=0.0)
• void reset ()
• virtual void update ()
• const FreeNeuron& operator[] (int i) const
• FreeNeuron& operator[] (int i)
• int size () const
• const ANNLayering& getLayering () const
• void setInitializer (NeuronInitializer* initer)
• FreeConnection* connect (int i, int j)
• virtual void operator>> (OStream& out) const
• String drawEPS (double xsize=-1, double ysize=-1) const
• void cleanup (bool removeDisableds=false, bool removePassthroughs=false)
• void drawFeedForward ()
• void check () const

Protected Members

• ANNLayering mLayering
• Array<FreeNeuron> mUnits
• FreeNeuron* mUnitTemplate
• NeuronInitializer* mInitializer

Detailed Description

A generic and extensible neural network.

Constructs the neural network with the given topology description string. See FreeNetwork::makeUnits() for more information about the format of the description string.

Builds a non-connected network according to the given layer topology (for example: "7-4-2" or "8").

See ANNLayering for some more information about the description strings.

Implementation for ANNetwork.

See also: makeUnits

Implementation for ANNetwork.

Destroys all units.

Removes the i:th unit and all connections from it and to it.

Builds the network from connection matrix, restricting the topology to only feedforward connections.

Connects all units in subsequent layers to form feedforward topology.

Connects all units to all units, including self.

Sets the unit template. The object t must be an instance of an inheritor of FreeNeuron.

NOTE: The template is destructed along with the FreeNetwork, so do not destroy it yourself.

Implementation for ANNetwork. Copies the other FreeNet unto self.

Implementation for ANNetwork. Initializes the weights and biases randomly according to the range, or if an initializer has been given, using that.

Resets activations to 0.0.

Transfers signals once for all units.

Returns i:th unit. Const version.

Returns i:th unit. Non-const version.

Returns the number of neurons in the network.

Returns current layering.

Sets the given initializer.

NOTE: Takes the ownership of the initializer and destroys it along with the FreeNetwork.

Creates a connection from neuron i to neuron j.

Implementation for Object. Prints out a formatted printout of the network.

Draws a picture of the network in EPS (Encapsulated PostScript). If no size is given, the image is drawn using default scaling.

Returns: Picture as EPS code.

Removes unnecessary connections and disables lonely units.

Parameters:

removeDisableds	Should disabled units be purged? Default: false (not purged).
removePassthroughs	Remove units that have only one input and one output connection. Default: false (not purged).

Order the coordinates of the FreeNeuron neurons (which do have coordinates) in the network in a pretty way in two-dimensional space.

Implementation for Object.

Layering information.

The neurons.

Neuron template.

Neuron initializer method.

Generated by: magi@magi.yok.utu.fi on Wed Sep 29 17:27:22 199.