FANN C# NeuralNet Float

Overview

The Fann Wrapper for C# provides Six main classes: FANNCSharp.Float::NeuralNet, FANNCSharp.Double::NeuralNet, FANNCSharp.Fixed::NeuralNet, FANNCSharp.Float::TrainingData, FANNCSharp.Double::TrainingData, FANNCSharp.Fixed::TrainingData. To use the wrapper add FANNCSharp.dll, FANNCSharp.Float.dll, FANNCSharp.Double.dll, FANNCSharp.Fixed.dll as a reference to your project.

Note: Notes and differences from C++ API

Method names are the same as the function names in the C++ API except the snake_case style naming has been replaced with CamelCase, getters and setters have been turned into properties, and “num” has been replace with “Count” to be more C#-ish
The arguments to the methods are the same as the C++ API except that the neural_net *net/training_data *data arguments are encapsulated so they are not present in the method signatures or are translated into class references.
The neural network and training data is automatically cleaned up in the Dispose methods

Summary

FANN C# NeuralNet Float	The Fann Wrapper for C# provides Six main classes: FANNCSharp.Float::NeuralNet, FANNCSharp.Double::NeuralNet, FANNCSharp.Fixed::NeuralNet, FANNCSharp.Float::TrainingData, FANNCSharp.Double::TrainingData, FANNCSharp.Fixed::TrainingData.
NeuralNet	NeuralNet is the main neural network class used for both training and execution using floats
Functions and Properties
NeuralNet	Creates a copy the other NeuralNet.
Dispose	Destructs the entire network.
NeuralNet	Creates a neural network of the desired NetworkType net_type.
NeuralNet	Creates a neural network of the desired NetworkType netType, based on a collection of layers.
NeuralNet	Creates a standard backpropagation neural network, which is sparsely connected, this will default the NetworkType to NetworkType::LAYER
NeuralNet	Creates a standard backpropagation neural network, which is sparsely connected, this will default the NetworkType to NetworkType::LAYER
NeuralNet	Constructs a backpropagation neural network from a configuration file, which have been saved by Save.
Run	Will run input through the neural network, returning an array of outputs, the number of which being equal to the number of neurons in the output layer.
Run	Will run input through the neural network, returning an array of outputs, the number of which being equal to the number of neurons in the output layer.
RandomizeWeights	Give each connection a random weight between minWeight and maxWeight
InitWeights	Initialize the weights using Widrow + Nguyen’s algorithm.
PrintConnections	Will print the connections of the network in a compact matrix, for easy viewing of the internals of the network.
Save	Save the entire network to a configuration file.
SaveToFixed	Saves the entire network to a configuration file.
Train	Train one iteration with a set of inputs, and a set of desired outputs.
TrainEpoch	Train one epoch with a set of training data.
TrainOnData	Trains on an entire dataset, for a period of time.
TrainOnFile	Does the same as TrainOnData, but reads the training data directly from a file.
Test	Test with a set of inputs, and a set of desired outputs.
TestData	Test a set of training data and calculates the MSE for the training data.
MSE	Reads the mean square error from the network.
ResetMSE	Resets the mean square error from the network.
SetCallback	Sets the callback function for use during training.
PrintParameters	Prints all of the parameters and options of the neural network
TrainingAlgorithm	Return or set the training algorithm as described by TrainingAlgorithm.
LearningRate	Return or set the learning rate.
GetActivationFunction	Get the activation function for neuron number neuron in layer number layer, counting the input layer as layer 0.
SetActivationFunction	Set the activation function for neuron number neuron in layer number layer, counting the input layer as layer 0.
SetActivationFunctionLayer	Set the activation function for all the neurons in the layer number layer, counting the input layer as layer 0.
ActivationFunctionHidden	Set the activation function for all of the hidden layers.
ActivationFunctionOutput	Set the activation function for the output layer.
GetActivationSteepness	Get the activation steepness for neuron number neuron in layer number layer, counting the input layer as layer 0.
SetActivationSteepness	Set the activation steepness for neuron number neuron in layer number layer, counting the input layer as layer 0.
SetActivationSteepnessLayer	Set the activation steepness all of the neurons in layer number layer, counting the input layer as layer 0.
ActivationSteepnessHidden	Set the steepness of the activation steepness in all of the hidden layers.
ActivationSteepnessOutput	Set the steepness of the activation steepness in the output layer.
TrainErrorFunction	Sets or returns the error function used during training.
QuickpropDecay	Gets or sets the quickprop decay factor.
QuickpropMu	Get or sets the quickprop mu factor.
RpropIncreaseFactor	Gets or sets the increase factor used during RPROP training.
RpropDecreaseFactor	Gets or sets the rprop decrease factor.
RpropDeltaZero	Gets or sets the rprop delta zero.
RpropDeltaMin	Gets or sets the rprop delta min.
RpropDeltaMax	Gets or set the rprop delta max.
SarpropWeightDecayShift	Gets or sets the sarprop weight decay shift.
SarpropStepErrorThresholdFactor	Gets or sets the sarprop step error threshold factor.
SarpropStepErrorShift	Gets or sets the sarprop step error shift.
SarpropTemperature	Gets or set the sarprop weight decay shift.
InputCount	Get the number of input neurons.
OutputCount	Get the number of output neurons.
TotalNeurons	Get the total number of neurons in the entire network.
TotalConnections	Get the total number of connections in the entire network.
NetworkType	Get the type of neural network it was created as.
ConnectionRate	Get the connection rate used when the network was created
LayerCount	Get the number of layers in the network
LayerArray	Get the number of neurons in each layer in the network.
BiasArray	Get the number of bias in each layer in the network.
ConnectionArray	Get the connections in the network.
WeightArray	Set connections in the network.
SetWeight	Set a connection in the network.
LearningMomentum	Get or set the learning momentum.
TrainStopFunction	Gets or sets the stop function used during training.
BitFailLimit	Gets or sets the bit fail limit used during training.
BitFail	Gets or set the number of fail bits.
CascadetrainOnData	Trains on an entire dataset, for a period of time using the Cascade2 training algorithm.
CascadetrainOnFile	Does the same as CascadetrainOnData, but reads the training data directly from a file.
CascadeOutputChangeFraction	The cascade output change fraction is a number between 0 and 1 determining how large a fraction the MSE value should change within CascadeOutputStagnationEpochs during training of the output connections, in order for the training not to stagnate.
CascadeOutputStagnationEpochs	The number of cascade output stagnation epochs determines the number of epochs training is allowed to continue without changing the MSE by a fraction of CascadeOutputChangeFraction.
CascadeCandidateChangeFraction	The cascade candidate change fraction is a number between 0 and 1 determining how large a fraction the MSE value should change within CascadeCandidateStagnationEpochs during training of the candidate neurons, in order for the training not to stagnate.
CascadeCandidateStagnationEpochs	The number of cascade candidate stagnation epochs determines the number of epochs training is allowed to continue without changing the MSE by a fraction of CascadeCandidateChangeFraction.
CascadeWeightMultiplier	The weight multiplier is a parameter which is used to multiply the weights from the candidate neuron before adding the neuron to the neural network.
CascadeCandidateLimit	The candidate limit is a limit for how much the candidate neuron may be trained.
CascadeMaxOutEpochs	The maximum out epochs determines the maximum number of epochs the output connections may be trained after adding a new candidate neuron.
CascadeMaxCandEpochs	The maximum candidate epochs determines the maximum number of epochs the input connections to the candidates may be trained before adding a new candidate neuron.
CascadeCandidatesCount	The number of candidates used during training (calculated by multiplying CascadeActivationFunctionsCount, CascadeActivationSteepnessesCount and CascadeCandidateGroupsCount).
CascadeActivationFunctionsCount	The number of activation functions in the CascadeActivationFunctions array.
CascadeActivationFunctions	The cascade activation functions array is an array of the different activation functions used by the candidates.
CascadeActivationSteepnessesCount	The number of activation steepnesses in the CascadeActivationFunctions array.
CascadeActivationSteepnesses	The cascade activation steepnesses array is an array of the different activation functions used by the candidates.
CascadeCandidateGroupsCount	The number of candidate groups is the number of groups of identical candidates which will be used during training.
ScaleTrain	Scale input and output data based on previously calculated parameters.
DescaleTrain	Descale input and output data based on previously calculated parameters.
SetInputScalingParams	Calculate scaling parameters for future use based on training data.
SetOutputScalingParams	Calculate scaling parameters for future use based on training data.
SetScalingParams	Calculate scaling parameters for future use based on training data.
ClearScalingParams	Clears scaling parameters.
ScaleInput	Scale data in input vector before feed it to ann based on previously calculated parameters.
ScaleOutput	Scale data in output vector before feed it to ann based on previously calculated parameters.
DescaleInput	Scale data in input vector after get it from ann based on previously calculated parameters.
DescaleOutput	Scale data in output vector after get it from ann based on previously calculated parameters.
ErrorLog	Change where errors are logged to.
ErrNo	Returns the last error number.
ResetErrno	Resets the last error number.
ResetErrstr	Resets the last error string.
ErrStr	Returns the last errstr.
PrintError	Prints the last error to Console.Error.
DisableSeedRand	Disables the automatic random generator seeding that happens in FANN.
EnableSeedRand	Enables the automatic random generator seeding that happens in FANN.
TrainEpochBatchParallel
TrainEpochIrpropmParallel
TrainEpochQuickpropParallel
TrainEpochSarpropParallel
TrainEpochIncrementalMod
TrainEpochBatchParallel
TrainEpochIrpropmParallel
TrainEpochQuickpropParallel
TrainEpochSarpropParallel
TrainEpochIncrementalMod
TestDataParallel
TestDataParallel
Delegates
TrainingCallback	This callback function can be called during training when using TrainOnData, TrainOnFile or CascadetrainOnData

NeuralNet

IDisposable

FANNCSharp.Float.NeuralNet

NeuralNet is the main neural network class used for both training and execution using floats

Encapsulation of a float neural network neural_net and associated C++ API functions.

Summary

Functions and Properties
NeuralNet	Creates a copy the other NeuralNet.
Dispose	Destructs the entire network.
NeuralNet	Creates a neural network of the desired NetworkType net_type.
NeuralNet	Creates a neural network of the desired NetworkType netType, based on a collection of layers.
NeuralNet	Creates a standard backpropagation neural network, which is sparsely connected, this will default the NetworkType to NetworkType::LAYER
NeuralNet	Creates a standard backpropagation neural network, which is sparsely connected, this will default the NetworkType to NetworkType::LAYER
NeuralNet	Constructs a backpropagation neural network from a configuration file, which have been saved by Save.
Run	Will run input through the neural network, returning an array of outputs, the number of which being equal to the number of neurons in the output layer.
Run	Will run input through the neural network, returning an array of outputs, the number of which being equal to the number of neurons in the output layer.
RandomizeWeights	Give each connection a random weight between minWeight and maxWeight
InitWeights	Initialize the weights using Widrow + Nguyen’s algorithm.
PrintConnections	Will print the connections of the network in a compact matrix, for easy viewing of the internals of the network.
Save	Save the entire network to a configuration file.
SaveToFixed	Saves the entire network to a configuration file.
Train	Train one iteration with a set of inputs, and a set of desired outputs.
TrainEpoch	Train one epoch with a set of training data.
TrainOnData	Trains on an entire dataset, for a period of time.
TrainOnFile	Does the same as TrainOnData, but reads the training data directly from a file.
Test	Test with a set of inputs, and a set of desired outputs.
TestData	Test a set of training data and calculates the MSE for the training data.
MSE	Reads the mean square error from the network.
ResetMSE	Resets the mean square error from the network.
SetCallback	Sets the callback function for use during training.
PrintParameters	Prints all of the parameters and options of the neural network
TrainingAlgorithm	Return or set the training algorithm as described by TrainingAlgorithm.
LearningRate	Return or set the learning rate.
GetActivationFunction	Get the activation function for neuron number neuron in layer number layer, counting the input layer as layer 0.
SetActivationFunction	Set the activation function for neuron number neuron in layer number layer, counting the input layer as layer 0.
SetActivationFunctionLayer	Set the activation function for all the neurons in the layer number layer, counting the input layer as layer 0.
ActivationFunctionHidden	Set the activation function for all of the hidden layers.
ActivationFunctionOutput	Set the activation function for the output layer.
GetActivationSteepness	Get the activation steepness for neuron number neuron in layer number layer, counting the input layer as layer 0.
SetActivationSteepness	Set the activation steepness for neuron number neuron in layer number layer, counting the input layer as layer 0.
SetActivationSteepnessLayer	Set the activation steepness all of the neurons in layer number layer, counting the input layer as layer 0.
ActivationSteepnessHidden	Set the steepness of the activation steepness in all of the hidden layers.
ActivationSteepnessOutput	Set the steepness of the activation steepness in the output layer.
TrainErrorFunction	Sets or returns the error function used during training.
QuickpropDecay	Gets or sets the quickprop decay factor.
QuickpropMu	Get or sets the quickprop mu factor.
RpropIncreaseFactor	Gets or sets the increase factor used during RPROP training.
RpropDecreaseFactor	Gets or sets the rprop decrease factor.
RpropDeltaZero	Gets or sets the rprop delta zero.
RpropDeltaMin	Gets or sets the rprop delta min.
RpropDeltaMax	Gets or set the rprop delta max.
SarpropWeightDecayShift	Gets or sets the sarprop weight decay shift.
SarpropStepErrorThresholdFactor	Gets or sets the sarprop step error threshold factor.
SarpropStepErrorShift	Gets or sets the sarprop step error shift.
SarpropTemperature	Gets or set the sarprop weight decay shift.
InputCount	Get the number of input neurons.
OutputCount	Get the number of output neurons.
TotalNeurons	Get the total number of neurons in the entire network.
TotalConnections	Get the total number of connections in the entire network.
NetworkType	Get the type of neural network it was created as.
ConnectionRate	Get the connection rate used when the network was created
LayerCount	Get the number of layers in the network
LayerArray	Get the number of neurons in each layer in the network.
BiasArray	Get the number of bias in each layer in the network.
ConnectionArray	Get the connections in the network.
WeightArray	Set connections in the network.
SetWeight	Set a connection in the network.
LearningMomentum	Get or set the learning momentum.
TrainStopFunction	Gets or sets the stop function used during training.
BitFailLimit	Gets or sets the bit fail limit used during training.
BitFail	Gets or set the number of fail bits.
CascadetrainOnData	Trains on an entire dataset, for a period of time using the Cascade2 training algorithm.
CascadetrainOnFile	Does the same as CascadetrainOnData, but reads the training data directly from a file.
CascadeOutputChangeFraction	The cascade output change fraction is a number between 0 and 1 determining how large a fraction the MSE value should change within CascadeOutputStagnationEpochs during training of the output connections, in order for the training not to stagnate.
CascadeOutputStagnationEpochs	The number of cascade output stagnation epochs determines the number of epochs training is allowed to continue without changing the MSE by a fraction of CascadeOutputChangeFraction.
CascadeCandidateChangeFraction	The cascade candidate change fraction is a number between 0 and 1 determining how large a fraction the MSE value should change within CascadeCandidateStagnationEpochs during training of the candidate neurons, in order for the training not to stagnate.
CascadeCandidateStagnationEpochs	The number of cascade candidate stagnation epochs determines the number of epochs training is allowed to continue without changing the MSE by a fraction of CascadeCandidateChangeFraction.
CascadeWeightMultiplier	The weight multiplier is a parameter which is used to multiply the weights from the candidate neuron before adding the neuron to the neural network.
CascadeCandidateLimit	The candidate limit is a limit for how much the candidate neuron may be trained.
CascadeMaxOutEpochs	The maximum out epochs determines the maximum number of epochs the output connections may be trained after adding a new candidate neuron.
CascadeMaxCandEpochs	The maximum candidate epochs determines the maximum number of epochs the input connections to the candidates may be trained before adding a new candidate neuron.
CascadeCandidatesCount	The number of candidates used during training (calculated by multiplying CascadeActivationFunctionsCount, CascadeActivationSteepnessesCount and CascadeCandidateGroupsCount).
CascadeActivationFunctionsCount	The number of activation functions in the CascadeActivationFunctions array.
CascadeActivationFunctions	The cascade activation functions array is an array of the different activation functions used by the candidates.
CascadeActivationSteepnessesCount	The number of activation steepnesses in the CascadeActivationFunctions array.
CascadeActivationSteepnesses	The cascade activation steepnesses array is an array of the different activation functions used by the candidates.
CascadeCandidateGroupsCount	The number of candidate groups is the number of groups of identical candidates which will be used during training.
ScaleTrain	Scale input and output data based on previously calculated parameters.
DescaleTrain	Descale input and output data based on previously calculated parameters.
SetInputScalingParams	Calculate scaling parameters for future use based on training data.
SetOutputScalingParams	Calculate scaling parameters for future use based on training data.
SetScalingParams	Calculate scaling parameters for future use based on training data.
ClearScalingParams	Clears scaling parameters.
ScaleInput	Scale data in input vector before feed it to ann based on previously calculated parameters.
ScaleOutput	Scale data in output vector before feed it to ann based on previously calculated parameters.
DescaleInput	Scale data in input vector after get it from ann based on previously calculated parameters.
DescaleOutput	Scale data in output vector after get it from ann based on previously calculated parameters.
ErrorLog	Change where errors are logged to.
ErrNo	Returns the last error number.
ResetErrno	Resets the last error number.
ResetErrstr	Resets the last error string.
ErrStr	Returns the last errstr.
PrintError	Prints the last error to Console.Error.
DisableSeedRand	Disables the automatic random generator seeding that happens in FANN.
EnableSeedRand	Enables the automatic random generator seeding that happens in FANN.
TrainEpochBatchParallel
TrainEpochIrpropmParallel
TrainEpochQuickpropParallel
TrainEpochSarpropParallel
TrainEpochIncrementalMod
TrainEpochBatchParallel
TrainEpochIrpropmParallel
TrainEpochQuickpropParallel
TrainEpochSarpropParallel
TrainEpochIncrementalMod
TestDataParallel
TestDataParallel
Delegates
TrainingCallback	This callback function can be called during training when using TrainOnData, TrainOnFile or CascadetrainOnData

Functions and Properties

NeuralNet

public NeuralNet( NeuralNet other )

Creates a copy the other NeuralNet.

Dispose

public void Dispose()

Destructs the entire network. Must be called manually.

NeuralNet

public NeuralNet( NetworkType netType,
uint numLayers,
params uint[] args )

Creates a neural network of the desired NetworkType net_type.

Parameters

numLayers	The total number of layers including the input and the output layer.
...	Integer values determining the number of neurons in each layer starting with the input layer and ending with the output layer.

Example

unt numLayes = 3;
uint numInput = 2;
uint numHidden = 3;
uint numOutput = 1;

NeuralNet net(numLayers, numInput, numHidden, numOutput);

This function appears in FANN >= 2.3.0.

NeuralNet

public NeuralNet( NetworkType netType,
ICollection<uint> layers )

Creates a neural network of the desired NetworkType netType, based on a collection of layers.

Parameters

netType	The desired network type of the neural network
layers	the collection of layer sizes

Example

NeuralNet net(NetworkType.LAYER, new uint[] {2, 3, 1});

This function appears in FANN >= 2.3.0.

NeuralNet

public NeuralNet( float connectionRate,
uint numLayers,
params uint[] args )

Creates a standard backpropagation neural network, which is sparsely connected, this will default the NetworkType to NetworkType::LAYER

Parameters

connectionRate	The connection rate controls how many connections there will be in the network. If the connection rate is set to 1, the network will be fully connected, but if it is set to 0.5 only half of the connections will be set. A connection rate of 1 will yield the same result as fann_create_standard
numLayers	The total number of layers including the input and the output layer.
...	Integer values determining the number of neurons in each layer starting with the input layer and ending with the output layer.

This function appears in FANN >= 2.3.0.

NeuralNet

public NeuralNet( float connectionRate,
ICollection<uint> layers )

Creates a standard backpropagation neural network, which is sparsely connected, this will default the NetworkType to NetworkType::LAYER

Parameters

connectionRate	The connection rate controls how many connections there will be in the network. If the connection rate is set to 1, the network will be fully connected, but if it is set to 0.5 only half of the connections will be set. A connection rate of 1 will yield the same result as fann_create_standard
numLayers	The total number of layers including the input and the output layer.
layers	Integer values determining the number of neurons in each layer starting with the input layer and ending with the output layer.

This function appears in FANN >= 2.3.0.

NeuralNet

public NeuralNet( string filename )

Constructs a backpropagation neural network from a configuration file, which have been saved by Save.

Run

public float[] Run( float[] input )

Will run input through the neural network, returning an array of outputs, the number of which being equal to the number of neurons in the output layer.

Parameters

input

An array of inputs to run through the neural network

Run

public float[] Run( DataAccessor input )

Will run input through the neural network, returning an array of outputs, the number of which being equal to the number of neurons in the output layer.

Parameters

input

A DataAccessor that points to the inputs to run through the neural network

RandomizeWeights

public void RandomizeWeights( float minWeight,
float maxWeight )

Give each connection a random weight between minWeight and maxWeight

From the beginning the weights are random between -0.1 and 0.1.

InitWeights

public void InitWeights( TrainingData data )

Initialize the weights using Widrow + Nguyen’s algorithm.

This function behaves similarly to fann_randomize_weights. It will use the algorithm developed by Derrick Nguyen and Bernard Widrow to set the weights in such a way as to speed up training. This technique is not always successful, and in some cases can be less efficient than a purely random initialization.

The algorithm requires access to the range of the input data (ie, largest and smallest input), and therefore accepts a second argument, data, which is the training data that will be used to train the network.

PrintConnections

public void PrintConnections()

Will print the connections of the network in a compact matrix, for easy viewing of the internals of the network.

The output from fann_print_connections on a small (2 2 1) network trained on the xor problem

Layer / Neuron 012345
L   1 / N    3 BBa...
L   1 / N    4 BBA...
L   1 / N    5 ......
L   2 / N    6 ...BBA
L   2 / N    7 ......

This network have five real neurons and two bias neurons. This gives a total of seven neurons named from 0 to 6. The connections between these neurons can be seen in the matrix. “.” is a place where there is no connection, while a character tells how strong the connection is on a scale from a-z. The two real neurons in the hidden layer (neuron 3 and 4 in layer 1) has connection from the three neurons in the previous layer as is visible in the first two lines. The output neuron (6) has connections form the three neurons in the hidden layer 3 - 5 as is visible in the fourth line.

To simplify the matrix output neurons is not visible as neurons that connections can come from, and input and bias neurons are not visible as neurons that connections can go to.

This function appears in FANN >= 1.2.0.

Save

public bool Save( string file )

Save the entire network to a configuration file.

The configuration file contains all information about the neural network and enables <NeuralNet(string filename)> to create an exact copy of the neural network and all of the parameters associated with the neural network.

These two parameters (SetCallback, ErrorLog) are NOT saved to the file because they cannot safely be ported to a different location. Also temporary parameters generated during training like MSE is not saved.

Return

The function returns true on success and false on failure.

SaveToFixed

public int SaveToFixed( string file )

Saves the entire network to a configuration file. But it is saved in fixed point format no matter which format it is currently in.

This is useful for training a network in floating points, and then later executing it in fixed point.

The function returns the bit position of the fix point, which can be used to find out how accurate the fixed point network will be. A high value indicates high precision, and a low value indicates low precision.

A negative value indicates very low precision, and a very strong possibility for overflow. (the actual fix point will be set to 0, since a negative fix point does not make sence).

Generally, a fix point lower than 6 is bad, and should be avoided. The best way to avoid this, is to have less connections to each neuron, or just less neurons in each layer.

The fixed point use of this network is only intended for use on machines that have no floating point processor, like an iPAQ. On normal computers the floating point version is actually faster.

Train

public void Train( float[] input,
float[] desiredOutput )

Train one iteration with a set of inputs, and a set of desired outputs. This training is always incremental training (see TrainingAlgorithm), since only one pattern is presented.

Parameters

input	an array of inputs. This array must be exactly InputCount long.
desiredOutput	an array of desired outputs. This array must be exactly OutputCount long.

TrainEpoch

public float TrainEpoch( TrainingData data )

Train one epoch with a set of training data.

Train one epoch with the training data stored in data. One epoch is where all of the training data is considered exactly once.

This function returns the MSE error as it is calculated either before or during the actual training. This is not the actual MSE after the training epoch, but since calculating this will require to go through the entire training set once more, it is more than adequate to use this value during training.

The training algorithm used by this function is chosen by the TrainingAlgorithm function.

TrainOnData

public void TrainOnData( TrainingData data,
uint maxEpochs,
uint epochsBetweenReports,
float desiredError )

Trains on an entire dataset, for a period of time.

This training uses the training algorithm chosen by TrainingAlgorithm, and the parameters set for these training algorithms.

Parameters

data	The data, which should be used during training
maxEpochs	The maximum number of epochs the training should continue
epochsBetweenReports	The number of epochs between printing a status report to the console. A value of zero means no reports should be printed.
desiredError	The desired MSE or BitFail, depending on which stop function is chosen by TrainStopFunction.

Instead of printing out reports every epochs_between_reports, a callback function can be called (see SetCallback).

TrainOnFile

public void TrainOnFile( string filename,
uint maxEpochs,
uint epochsBetweenReports,
float desiredError )

Does the same as TrainOnData, but reads the training data directly from a file.

Test

public float[] Test( float[] input,
float[] desiredOutput )

Test with a set of inputs, and a set of desired outputs. This operation updates the mean square error, but does not change the network in any way.

TestData

public float TestData( TrainingData data )

Test a set of training data and calculates the MSE for the training data.

This function updates the MSE and the bit fail values.

MSE

public float MSE { get }

Reads the mean square error from the network.

Reads the mean square error from the network. This value is calculated during training or testing, and can therefore sometimes be a bit off if the weights have been changed since the last calculation of the value.

ResetMSE

public void ResetMSE()

Resets the mean square error from the network.

This function also resets the number of bits that fail.

SetCallback

public void SetCallback( TrainingCallback callback,
Object userData )

Sets the callback function for use during training. The user_data is passed to the callback. It can point to arbitrary data that the callback might require and can be NULL if it is not used.

See FANN::callback_type for more information about the callback function.

The default callback function simply prints out some status information.

This function appears in FANN >= 2.0.0.

PrintParameters

public void PrintParameters()

Prints all of the parameters and options of the neural network

TrainingAlgorithm

public TrainingAlgorithm TrainingAlgorithm { get, set }

Return or set the training algorithm as described by TrainingAlgorithm. This training algorithm is used by TrainOnData and associated functions.

Note that this algorithm is also used during CascadetrainOnData, although only TrainingAlgorithm.TRAIN_RPROP and TrainingAlgorithm.:TRAIN_QUICKPROP is allowed during cascade training.

The default training algorithm is TrainingAlgorithm.TRAIN_RPROP.

LearningRate

public float LearningRate { get, set }

Return or set the learning rate.

The learning rate is used to determine how aggressive training should be for some of the training algorithms (TrainingAlgorithm.TRAIN_INCREMENTAL, TrainingAlgorithm.TRAIN_BATCH, TrainingAlgorithm.TRAIN_QUICKPROP). Do however note that it is not used in TrainingAlgorithm.TRAIN_RPROP.

The default learning rate is 0.7.

GetActivationFunction

public ActivationFunction GetActivationFunction( int layer,
int neuron )

Get the activation function for neuron number neuron in layer number layer, counting the input layer as layer 0.

It is not possible to get activation functions for the neurons in the input layer.

Information about the individual activation functions is available at ActivationFunction.

Returns

The activation function for the neuron or -1 if the neuron is not defined in the neural network.

SetActivationFunction

public void SetActivationFunction( ActivationFunction function,
int layer,
int neuron )

Set the activation function for neuron number neuron in layer number layer, counting the input layer as layer 0.

It is not possible to set activation functions for the neurons in the input layer.

When choosing an activation function it is important to note that the activation functions have different range. FANN::SIGMOID is e.g. in the 0 - 1 range while FANN::SIGMOID_SYMMETRIC is in the -1 - 1 range and FANN::LINEAR is unbounded.

Information about the individual activation functions is available at ActivationFunction.

The default activation function is FANN::SIGMOID_STEPWISE.

SetActivationFunctionLayer

public void SetActivationFunctionLayer( ActivationFunction function,
int layer )

Set the activation function for all the neurons in the layer number layer, counting the input layer as layer 0.

It is not possible to set activation functions for the neurons in the input layer.

ActivationFunctionHidden

public ActivationFunction ActivationFunctionHidden { set }

Set the activation function for all of the hidden layers.

ActivationFunctionOutput

public ActivationFunction ActivationFunctionOutput { set }

Set the activation function for the output layer.

GetActivationSteepness

public float GetActivationSteepness( int layer,
int neuron )

Get the activation steepness for neuron number neuron in layer number layer, counting the input layer as layer 0.

It is not possible to get activation steepness for the neurons in the input layer.

The steepness of an activation function says something about how fast the activation function goes from the minimum to the maximum. A high value for the activation function will also give a more aggressive training.

When training neural networks where the output values should be at the extremes (usually 0 and 1, depending on the activation function), a steep activation function can be used (e.g. 1.0).

The default activation steepness is 0.5.

Returns

The activation steepness for the neuron or -1 if the neuron is not defined in the neural network.

SetActivationSteepness

public void SetActivationSteepness( float steepness,
int layer,
int neuron )

Set the activation steepness for neuron number neuron in layer number layer, counting the input layer as layer 0.

It is not possible to set activation steepness for the neurons in the input layer.

When training neural networks where the output values should be at the extremes (usually 0 and 1, depending on the activation function), a steep activation function can be used (e.g. 1.0).

The default activation steepness is 0.5.

SetActivationSteepnessLayer

public void SetActivationSteepnessLayer( float steepness,
int layer )

Set the activation steepness all of the neurons in layer number layer, counting the input layer as layer 0.

It is not possible to set activation steepness for the neurons in the input layer.

ActivationSteepnessHidden

public float ActivationSteepnessHidden { set }

Set the steepness of the activation steepness in all of the hidden layers.

ActivationSteepnessOutput

public float ActivationSteepnessOutput { set }

Set the steepness of the activation steepness in the output layer.

TrainErrorFunction

public ErrorFunction TrainErrorFunction { get, set }

Sets or returns the error function used during training.

The error functions is described further in ErrorFunction

The default error function is ErrorFunction.ERRORFUNC_TANH

QuickpropDecay

public float QuickpropDecay { get, set }

Gets or sets the quickprop decay factor.

The decay is a small negative valued number which is the factor that the weights should become smaller in each iteration during quickprop training. This is used to make sure that the weights do not become too high during training.

The default decay is -0.0001.

QuickpropMu

public float QuickpropMu { get, set }

Get or sets the quickprop mu factor.

The mu factor is used to increase and decrease the step-size during quickprop training. The mu factor should always be above 1, since it would otherwise decrease the step-size when it was suppose to increase it.

The default mu factor is 1.75.

RpropIncreaseFactor

public float RpropIncreaseFactor { get, set }

Gets or sets the increase factor used during RPROP training.

The increase factor is a value larger than 1, which is used to increase the step-size during RPROP training.

The default increase factor is 1.2.

RpropDecreaseFactor

public float RpropDecreaseFactor { get, set }

Gets or sets the rprop decrease factor.

The decrease factor is a value smaller than 1, which is used to decrease the step-size during RPROP training.

The default decrease factor is 0.5.

RpropDeltaZero

public float RpropDeltaZero { get, set }

Gets or sets the rprop delta zero.

The initial step-size is a small positive number determining how small the initial step-size may be.

The default value delta zero is 0.1.

RpropDeltaMin

public float RpropDeltaMin { get, set }

Gets or sets the rprop delta min.

The minimum step-size is a small positive number determining how small the minimum step-size may be.

The default value delta min is 0.0.

RpropDeltaMax

public float RpropDeltaMax { get, set }

Gets or set the rprop delta max.

The maximum step-size is a positive number determining how large the maximum step-size may be.

The default delta max is 50.0.

SarpropWeightDecayShift

public float SarpropWeightDecayShift { get, set }

Gets or sets the sarprop weight decay shift.

The default delta max is -6.644.

SarpropStepErrorThresholdFactor

public float SarpropStepErrorThresholdFactor { get, set }

Gets or sets the sarprop step error threshold factor.

The default delta max is 0.1.

SarpropStepErrorShift

public float SarpropStepErrorShift { get, set }

Gets or sets the sarprop step error shift.

The default delta max is 1.385.

SarpropTemperature

public float SarpropTemperature { get, set }

Gets or set the sarprop weight decay shift.

The default delta max is 0.015.

InputCount

public uint InputCount { get }

Get the number of input neurons.

This function appears in FANN >= 1.0.0.

OutputCount

public uint OutputCount { get }

Get the number of output neurons.

This function appears in FANN >= 1.0.0.

TotalNeurons

public uint TotalNeurons { get }

Get the total number of neurons in the entire network. This number does also include the bias neurons, so a 2-4-2 network has 2+4+2 +2(bias) = 10 neurons.

This function appears in FANN >= 1.0.0.

TotalConnections

public uint TotalConnections { get }

Get the total number of connections in the entire network.

This function appears in FANN >= 1.0.0.

NetworkType

public NetworkType NetworkType { get }

Get the type of neural network it was created as.

Returns

The neural network type from enum NetworkType

ConnectionRate

public float ConnectionRate { get }

Get the connection rate used when the network was created

Returns

The connection rate

LayerCount

public uint LayerCount { get }

Get the number of layers in the network

Returns

The number of layers in the neural network

LayerArray

public uint[] LayerArray { get }

Get the number of neurons in each layer in the network.

Bias is not included so the layers match the create methods.

BiasArray

public uint[] BiasArray { get }

Get the number of bias in each layer in the network.

ConnectionArray

public Connection[] ConnectionArray { get }

Get the connections in the network.

WeightArray

public Connection[] WeightArray { set }

Set connections in the network.

Only the weights can be changed, connections and weights are ignored if they do not already exist in the network.

SetWeight

public void SetWeight( uint fromNeuron,
uint toNeuron,
float weight )

Set a connection in the network.

Only the weights can be changed. The connection/weight is ignored if it does not already exist in the network.

LearningMomentum

public float LearningMomentum { get, set }

Get or set the learning momentum.

The learning momentum can be used to speed up TrainingAlgorithm.TRAIN_INCREMENTAL training. A too high momentum will however not benefit training. Setting momentum to 0 will be the same as not using the momentum parameter. The recommended value of this parameter is between 0.0 and 1.0.

The default momentum is 0.

TrainStopFunction

public StopFunction TrainStopFunction { get, set }

Gets or sets the stop function used during training.

The stop function is described further in StopFunction

The default stop function is StopFunction.STOPFUNC_MSE

BitFailLimit

public float BitFailLimit { get, set }

Gets or sets the bit fail limit used during training.

The bit fail limit is used during training when the StopFunction is set to StopFunction.FANN_STOPFUNC_BIT.

The limit is the maximum accepted difference between the desired output and the actual output during training. Each output that diverges more than this limit is counted as an error bit. This difference is divided by two when dealing with symmetric activation functions, so that symmetric and not symmetric activation functions can use the same limit.

The default bit fail limit is 0.35.

This function appears in FANN >= 2.0.0.

BitFail

public uint BitFail { get }

Gets or set the number of fail bits. Means the number of output neurons which differ more than the bit fail limit (see BitFailLimit). The bits are counted in all of the training data, so this number can be higher than the number of training data.

This value is reset by ResetMSE and updated by all the same functions which also updates the MSE value (e.g. TestData, TrainEpoch)

CascadetrainOnData

public void CascadetrainOnData( TrainingData data,
uint maxNeurons,
uint neuronsBetweenReports,
float desiredError )

Trains on an entire dataset, for a period of time using the Cascade2 training algorithm. This algorithm adds neurons to the neural network while training, which means that it needs to start with an ANN without any hidden layers. The neural network should also use shortcut connections, so NeuralNet(NetworkType.SHORTCUT, ...) should be used to create the NeuralNetwork like this:

NeuralNet net(NetworkType.SHORTCUT, ...);

This training uses the parameters set using the Cascade..., but it also uses another training algorithm as it’s internal training algorithm. This algorithm can be set to either TrainingAlgorithm.TRAIN_RPROP or TrainingAlgorithm.TRAIN_QUICKPROP by TrainingAlgorithm, and the parameters set for these training algorithms will also affect the cascade training.

Parameters

data	The data, which should be used during training
maxNeurons	The maximum number of neurons to be added to neural network
neuronsBetweenReports	The number of neurons between printing a status report to the console. A value of zero means no reports should be printed.
desiredError	The desired MSE or BitFail, depending on which stop function is chosen by TrainStopFunction.

Instead of printing out reports every neuronsBetweenReports, a callback function can be called (see SetCallback).

CascadetrainOnFile

public void CascadetrainOnFile( string filename,
uint maxNeurons,
uint neuronsBetweenReports,
float desiredError )

Does the same as CascadetrainOnData, but reads the training data directly from a file.

CascadeOutputChangeFraction

public float CascadeOutputChangeFraction { get, set }

The cascade output change fraction is a number between 0 and 1 determining how large a fraction the MSE value should change within CascadeOutputStagnationEpochs during training of the output connections, in order for the training not to stagnate. If the training stagnates, the training of the output connections will be ended and new candidates will be prepared.

This means

If the MSE does not change by a fraction of CascadeOutputStagnationEpochs during a period of CascadeOutputStagnationEpochs, the training of the output connections is stopped because the training has stagnated.

If the cascade output change fraction is low, the output connections will be trained more and if the fraction is high they will be trained less.

The default cascade output change fraction is 0.01, which is equalent to a 1% change in MSE.

CascadeOutputStagnationEpochs

public uint CascadeOutputStagnationEpochs { get, set }

The number of cascade output stagnation epochs determines the number of epochs training is allowed to continue without changing the MSE by a fraction of CascadeOutputChangeFraction.

See more info about this parameter in CascadeOutputChangeFraction.

The default number of cascade output stagnation epochs is 12.

CascadeCandidateChangeFraction

public float CascadeCandidateChangeFraction { get, set }

The cascade candidate change fraction is a number between 0 and 1 determining how large a fraction the MSE value should change within CascadeCandidateStagnationEpochs during training of the candidate neurons, in order for the training not to stagnate. If the training stagnates, the training of the candidate neurons will be ended and the best candidate will be selected.

This means

If the MSE does not change by a fraction of CascadeCandidateChangeFraction during a period of CascadeCandidateStagnationEpochs, the training of the candidate neurons is stopped because the training has stagnated.

If the cascade candidate change fraction is low, the candidate neurons will be trained more and if the fraction is high they will be trained less.

The default cascade candidate change fraction is 0.01, which is equalent to a 1% change in MSE.

CascadeCandidateStagnationEpochs

public uint CascadeCandidateStagnationEpochs { get, set }

The number of cascade candidate stagnation epochs determines the number of epochs training is allowed to continue without changing the MSE by a fraction of CascadeCandidateChangeFraction.

See more info about this parameter in CascadeCandidateChangeFraction.

The default number of cascade candidate stagnation epochs is 12.

CascadeWeightMultiplier

public float CascadeWeightMultiplier { get, set }

The weight multiplier is a parameter which is used to multiply the weights from the candidate neuron before adding the neuron to the neural network. This parameter is usually between 0 and 1, and is used to make the training a bit less aggressive.

The default weight multiplier is 0.4

CascadeCandidateLimit

public float CascadeCandidateLimit { get, set }

The candidate limit is a limit for how much the candidate neuron may be trained. The limit is a limit on the proportion between the MSE and candidate score.

Set this to a lower value to avoid overfitting and to a higher if overfitting is not a problem.

The default candidate limit is 1000.0

CascadeMaxOutEpochs

public uint CascadeMaxOutEpochs { get, set }

The maximum out epochs determines the maximum number of epochs the output connections may be trained after adding a new candidate neuron.

The default max out epochs is 150

CascadeMaxCandEpochs

public uint CascadeMaxCandEpochs { get, set }

The maximum candidate epochs determines the maximum number of epochs the input connections to the candidates may be trained before adding a new candidate neuron.

The default max candidate epochs is 150

CascadeCandidatesCount

public uint CascadeCandidatesCount { get }

The number of candidates used during training (calculated by multiplying CascadeActivationFunctionsCount, CascadeActivationSteepnessesCount and CascadeCandidateGroupsCount).

The actual candidates is defined by the CascadeActivationFunctions and CascadeActivationSteepnesses arrays. These arrays define the activation functions and activation steepnesses used for the candidate neurons. If there are 2 activation functions in the activation function array and 3 steepnesses in the steepness array, then there will be 2x3=6 different candidates which will be trained. These 6 different candidates can be copied into several candidate groups, where the only difference between these groups is the initial weights. If the number of groups is set to 2, then the number of candidate neurons will be 2x3x2=12. The number of candidate groups is defined by CascadeCandidateGroupsCount.

The default number of candidates is 6x4x2 = 48

CascadeActivationFunctionsCount

public uint CascadeActivationFunctionsCount { get }

The number of activation functions in the CascadeActivationFunctions array.

The default number of activation functions is 10.

CascadeActivationFunctions

public ActivationFunction[] CascadeActivationFunctions { get, set }

The cascade activation functions array is an array of the different activation functions used by the candidates.

See CascadeCandidatesCount for a description of which candidate neurons will be generated by this array.

CascadeActivationSteepnessesCount

public uint CascadeActivationSteepnessesCount { get }

The number of activation steepnesses in the CascadeActivationFunctions array.

The default number of activation steepnesses is 4.

CascadeActivationSteepnesses

public float[] CascadeActivationSteepnesses { get, set }

The cascade activation steepnesses array is an array of the different activation functions used by the candidates.

See CascadeCandidatesCount for a description of which candidate neurons will be generated by this array.

The default activation steepnesses is {0.25, 0.50, 0.75, 1.00}

CascadeCandidateGroupsCount

public uint CascadeCandidateGroupsCount { get, set }

The number of candidate groups is the number of groups of identical candidates which will be used during training.

This number can be used to have more candidates without having to define new parameters for the candidates.

See CascadeCandidatesCount for a description of which candidate neurons will be generated by this parameter.

The default number of candidate groups is 2

ScaleTrain

public void ScaleTrain( TrainingData data )

Scale input and output data based on previously calculated parameters.

DescaleTrain

public void DescaleTrain( TrainingData data )

Descale input and output data based on previously calculated parameters.

SetInputScalingParams

public bool SetInputScalingParams( TrainingData data,
float newInputMin,
float newInputMax )

Calculate scaling parameters for future use based on training data.

SetOutputScalingParams

public bool SetOutputScalingParams( TrainingData data,
float newOutputMin,
float newOutputMax )

Calculate scaling parameters for future use based on training data.

SetScalingParams

public bool SetScalingParams( TrainingData data,
float newInputMin,
float newInputMax,
float newOutputMin,
float newOutputMax )

Calculate scaling parameters for future use based on training data.

ClearScalingParams

public bool ClearScalingParams()

Clears scaling parameters.

ScaleInput

public void ScaleInput( DataAccessor input )

Scale data in input vector before feed it to ann based on previously calculated parameters.

ScaleOutput

public void ScaleOutput( DataAccessor output )

Scale data in output vector before feed it to ann based on previously calculated parameters.

DescaleInput

public void DescaleInput( float[] input )

Scale data in input vector after get it from ann based on previously calculated parameters.

DescaleOutput

public void DescaleOutput( float[] output )

Scale data in output vector after get it from ann based on previously calculated parameters.

ErrorLog

public FannFile ErrorLog { set }

Change where errors are logged to.

If the value is NULL, no errors will be printed.

If NeuralNet is empty the default log will be set. The default log is the log used when creating a NeuralNet. This default log will also be the default for all new NeuralNet that are created.

The default behavior is to log them to Console.Error.

ErrNo

public uint ErrNo { get }

Returns the last error number.

ResetErrno

public void ResetErrno()

Resets the last error number.

This function appears in FANN >= 1.1.0.

ResetErrstr

public void ResetErrstr()

Resets the last error string.

This function appears in FANN >= 1.1.0.

ErrStr

public string ErrStr { get }

Returns the last errstr.

This function calls fann_reset_errno and fann_reset_errstr

This function appears in FANN >= 1.1.0.

PrintError

public void PrintError()

Prints the last error to Console.Error.

This function appears in FANN >= 1.1.0.

DisableSeedRand

public void DisableSeedRand()

Disables the automatic random generator seeding that happens in FANN.

Per default FANN will always seed the random generator when creating a new network, unless FANN_NO_SEED is defined during compilation of the library. This method can disable this at runtime.

This function appears in FANN >= 2.3.0

EnableSeedRand

public void EnableSeedRand()

Enables the automatic random generator seeding that happens in FANN.

Per default FANN will always seed the random generator when creating a new network, unless FANN_NO_SEED is defined during compilation of the library. This method can disable this at runtime.

This function appears in FANN >= 2.3.0

TrainEpochBatchParallel

public float TrainEpochBatchParallel( TrainingData data,
uint threadNumb )

Parameters

data	the data to train on
threadNumb	the thread to do training on

TrainEpochIrpropmParallel

public float TrainEpochIrpropmParallel( TrainingData data,
uint threadNumb )

Parameters

data	the data to train on
threadNumb	the thread to do training on

TrainEpochQuickpropParallel

public float TrainEpochQuickpropParallel( TrainingData data,
uint threadNumb )

Parameters

data	the data to train on
threadNumb	the thread to do training on

TrainEpochSarpropParallel

public float TrainEpochSarpropParallel( TrainingData data,
uint threadNumb )

Parameters

data	the data to train on
threadNumb	the thread to do training on

TrainEpochIncrementalMod

public float TrainEpochIncrementalMod( TrainingData data )

Parameters

data	the data to train on

TrainEpochBatchParallel

public float TrainEpochBatchParallel( TrainingData data,
uint threadNumb,
List<List<float>> predictedOutputs )

Parameters

data	the data to train on
threadNumb	the thread to do training on
predictedOutputs	the predicted outputs

TrainEpochIrpropmParallel

public float TrainEpochIrpropmParallel( TrainingData data,
uint threadNumb,
List<List<float>> predictedOutputs )

Parameters

data	the data to train on
threadNumb	the thread to do training on
predictedOutputs	the predicted outputs

TrainEpochQuickpropParallel

public float TrainEpochQuickpropParallel( TrainingData data,
uint threadNumb,
List<List<float>> predictedOutputs )

Parameters

data	the data to train on
threadNumb	the thread to do training on
predictedOutputs	the predicted outputs

TrainEpochSarpropParallel

public float TrainEpochSarpropParallel( TrainingData data,
uint threadNumb,
List<List<float>> predictedOutputs )

Parameters

data	the data to train on
threadNumb	the thread to do training on
predictedOutputs	the predicted outputs

TrainEpochIncrementalMod

public float TrainEpochIncrementalMod( TrainingData data,
List<List<float>> predictedOutputs )

Parameters

data	the data to train on
predictedOutputs	the predicted outputs

TestDataParallel

public float TestDataParallel( TrainingData data,
uint threadNumb )

Parameters

data	the data to train on
threadNumb	the thread to do training on

TestDataParallel

public float TestDataParallel( TrainingData data,
uint threadNumb,
List<List<float>> predictedOutputs )

Parameters

data	the data to train on
threadNumb	the thread to do training on
predictedOutputs	the predicted outputs

Delegates

TrainingCallback

public delegate int TrainingCallback( NeuralNet net,
TrainingData data,
uint maxEpochs,
uint epochsBetweenReports,
float desiredError,
uint epochs,
Object userData )

This callback function can be called during training when using TrainOnData, TrainOnFile or CascadetrainOnData

The callback can be set by using SetCallback and is very useful for doing custom things during training. It is recommended to use this function when implementing custom training procedures, or when visualizing the training in a GUI etc. The parameters which the callback function takes is the parameters given to the TrainOnData, plus an epochs parameter which tells how many epochs the training have taken so far.

The callback function should return an integer, if the callback function returns -1, the training will terminate.

Example of a callback function that prints information to the Console

int PrintCallback(NeuralNet net, TrainingData data,
    uint maxEpochs, uint epochsBetweenReports,
    float desiredError, uint epochs, Object userData)
{
    Console.WriteLine("Epochs     {0}. Current Error: {1}", epochs.ToString("00000000"), net.MSE.ToString().PadRight(8));
}

FANN C# NeuralNet Float

Overview

NeuralNet

Functions and Properties

NeuralNet

Dispose

NeuralNet

Parameters

Example

NeuralNet

Parameters

Example

NeuralNet

Parameters

NeuralNet

Parameters

NeuralNet

See also

Run

Parameters

See also

Run

Parameters

See also

RandomizeWeights

See also

InitWeights

See also

PrintConnections

Save

Return

See also

SaveToFixed

See also

Train

Parameters

See also

TrainEpoch

See also

TrainOnData

Parameters

See also

TrainOnFile

See also

Test

See also

TestData

See also

MSE

See also

ResetMSE

See also

SetCallback

PrintParameters

See also

TrainingAlgorithm

See also

LearningRate

See also

GetActivationFunction

Returns

See also

SetActivationFunction

See also

SetActivationFunctionLayer

See also

ActivationFunctionHidden

See also

ActivationFunctionOutput

See also

GetActivationSteepness

Returns

See also

SetActivationSteepness

See also

SetActivationSteepnessLayer

See also

ActivationSteepnessHidden

See also

ActivationSteepnessOutput