This is Neuron.rtf in view mode; [Download] [Up]
Neuron V1.2
15-JAN-1992
Ralph Zazula
zazula@pri.com
Neuron
INHERITS FROM Object
DECLARED IN Neuron.h
CLASS DESCRIPTION
The Neuron class implements the fundamental unit of neural-network computation. The Neuron can use various activation functions. The following Neuron types are defined in Neuron.h: Binary, Sigmoid, Sign, Tanh. The Neuron class also provides for stochastic networks by storing the current temperature and modifying the output of the Neuron accordingly.
INSTANCE VARIABLES
Inherited from Object Class isa;
Declared in Neuron id inputs;
double lastOutput;
id random;
struct {
id source;
double weight;
connection *next;
} connection;
int nodeType;
double T;
inputs a Storage object containing a list of connections
lastOutput the output of this Neuron based on the inputs
random a Random instance (random number generator)
connection a struct that contains the source and weight of an input connection
nodeType the type of activation function being used (Binary, Sigmoid, Sign, Tanh).
the default is Sigmoid.
T the current temperature
METHOD TYPES
Initializing a new instance - init
Creating connections ± connect:
± connect:withWeight:
Setting parameters ± setWeightFor:to:
± setOutput:
± setType:
± setTemp:
± setRandom:
± setSymmetric:
Getting parameters ± getWeightFor:
± getType
± getTemp
± getSymmetric
± lastOutput
± inputs
Generating new output ± step
± activation:
INSTANCE METHODS
activation:
- (double)activation:(double)net
Returns the value for the activation function of the receiver with input net. Does not change the output of the neuron.
See also: - step, ± lastOutput
connect:
- connect:anObject
Creates a connection from the senders output to the receivers input with a random weight value between 0.0 and 0.01.
See also: - connect:withWeight:
connect:withWeight:
- connect:anObject withWeight:(double)weight
Creates a connection from the senders output to the receivers input with the weight value specified by weight.
See also: - connect:
changeWeightFor:by:
- changeWeightFor:anObject by:(double)delta
Changes the weight of the connection from the senders output to the receivers input by delta.
See also: - connect:withWeight:, ± setWeightFor:to:, ± getWeightFor:
getTemp
- (double)getTemp
Returns the current temperature of the neuron. The default temperature is 0.0.
See also: - setTemp:
getType
- (int)getType
Returns the type of activation function applied to this neurons inputs. Valid types are Binary, Sigmoid, Sign and Tanh. The default type is Sigmoid.
See also: - setType:
getSymmetric
- (BOOL)getWeightFor
Returns the symmetric status of the receiver.
See also: - setSymmetric:
getWeightFor:
- (double)getWeightFor:anObject
Returns the weight value for the receivers input connection from anObject. Returns NAN if anObject isn't in the input list of the receiver.
See also: - setWeightFor:to:, ± connect:withWeight:
init
- init
Initializes the receiver, a new Neuron object.
inputs
- inputs
Returns a pointer to the Storage object that contains the receivers input connections and associated weights.
See also: - connect:, - connect:withWeight:
lastOutput
- (double)lastOutput
Returns the value of the lastOutput instance variable.
See also: - setOutput:, ± step
setOutput:
- setOutput:(double)weight
Sets the instance variable lastOutput of the receiver to weight. This is used to set the output values of the input Neurons, for instance.
See also: - lastOutput
setRandom:
- setRandom:theRandom
Sets the the instance variable random to point to theRandom. This should be used for large nets in order to have only one instance of the Random class. This also assures the same seed for all Neuron instances. If random == nil when connect: or connect:withWeight: are called, a new instance is created.
See also:
setSymmetric:
- setRandom:(BOOL)isSymmetric
Controls whether a Neuron trys to update the reverse connection when it's changeWeightFor:by: method is called. Default is NO.
See also: ± changeWeightFor:by:
setTemp:
- setType:(double)newTemp
Sets the current temperature to newTemp. The default temperature is 0.0.
See also: - (double)getTemp
setType:
- setType:(int)type
Sets the activation function to type type. Valid types are: Binary, Sigmoid, Sign and Tanh. The default type is Sigmoid.
See also: - getType
setWeightFor:to:
- setWeightFor:anObject to:(double)weight
Sets the weight for the connection from the output of anObject to the receivers input to weight.
See also: - getWeightFor:, - connect:withWeight:
step
- step
This method causes the receiver to generate a new value for lastOutput by applying it's activation function to the weighted sum of it's inputs.
These are the contents of the former NiCE NeXT User Group NeXTSTEP/OpenStep software archive, currently hosted by Netfuture.ch.