org.knime.base.data.neural
Class MultiLayerPerceptron

java.lang.Object
  org.knime.base.data.neural.MultiLayerPerceptron

public class MultiLayerPerceptron
extends Object

Representation of a MultiLayer Perceptron, a neural net with one or more hidden layers.

Author:

Nicolas Cebron, University of Konstanz

Field Summary

static String	ALLLAYERS_KEY Key to store all layers of the MLP in the ModelContent.
static int	CLASSIFICATION_MODE Indicates whether the MLP does classification with multiple output neurons, one neuron for each class.
static String	CLASSVALUE_KEY Key to store the class value of a neuron in the ModelContent.
static String	INPUT_KEY Key to store the input value of a neuron in the ModelContent.
static String	MODE_KEY Key to store the mode of the MLP in the ModelContent.
static int	REGRESSION_MODE Indicates whether the MLP does regression with one output neuron.
static String	THRESHOLD_KEY Key to store the threshold value of a neuron in the ModelContent.
static String	WEIGHT_KEY Key to store the weights of a neuron in the ModelContent.

Constructor Summary

MultiLayerPerceptron()
Constructor for an empty Neural Net.

MultiLayerPerceptron(Architecture a)
Constructs a net with a given architecture.

MultiLayerPerceptron(Layer[] layers)
Constructs a net with the given layers.

Method Summary

Architecture	getArchitecture()
HashMap<DataCell,Integer>	getClassMapping() Allows to get the class mapping from output neurons to class values.
String	getClassOutput(double[] in) Finds the winner and returns its class output value.
HashMap<String,Integer>	getInputMapping() Allows to get the input mapping from input neurons to columns.
Layer	getLayer(int i) Returns layer at a given position.
Layer[]	getLayers()
int	getMode()
int	getNrLayers()
static MultiLayerPerceptron	loadPredictorParams(ModelContentRO predParams)
double[]	output(double[] in) Evaluates input and returns output of output neurons.
double[]	output(Double[] in) Computes the output for given input.
void	savePredictorParams(ModelContentWO predParams) Stores this MLP model to config.
void	setArchitecture(Architecture architecture) Sets the architecture.
void	setClassMapping(HashMap<DataCell,Integer> map) Allows for setting the class mapping from output neurons to class values.
void	setInputMapping(HashMap<String,Integer> map) Allows for setting the input mapping from input neurons to class values.
void	setLayer(int i, Layer layer) Sets the layer at a given position.
void	setLayers(Layer[] layers) Sets all layers.
void	setMode(int mode) Sets the mode of the MLP.

Methods inherited from class java.lang.Object

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

Field Detail

INPUT_KEY

public static final String INPUT_KEY

Key to store the input value of a neuron in the ModelContent.

See Also:

Constant Field Values

WEIGHT_KEY

public static final String WEIGHT_KEY

Key to store the weights of a neuron in the ModelContent.

See Also:

Constant Field Values

THRESHOLD_KEY

public static final String THRESHOLD_KEY

Key to store the threshold value of a neuron in the ModelContent.

See Also:

Constant Field Values

CLASSVALUE_KEY

public static final String CLASSVALUE_KEY

Key to store the class value of a neuron in the ModelContent.

See Also:

Constant Field Values

MODE_KEY

public static final String MODE_KEY

Key to store the mode of the MLP in the ModelContent.

See Also:

Constant Field Values

ALLLAYERS_KEY

public static final String ALLLAYERS_KEY

Key to store all layers of the MLP in the ModelContent.

See Also:

Constant Field Values

REGRESSION_MODE

public static final int REGRESSION_MODE

Indicates whether the MLP does regression with one output neuron.

See Also:

Constant Field Values

CLASSIFICATION_MODE

public static final int CLASSIFICATION_MODE

Indicates whether the MLP does classification with multiple output neurons, one neuron for each class.

See Also:

Constant Field Values

Constructor Detail

MultiLayerPerceptron

public MultiLayerPerceptron(Layer[] layers)

Constructs a net with the given layers.

Parameters:

layers - layers for the new net

MultiLayerPerceptron

public MultiLayerPerceptron(Architecture a)

Constructs a net with a given architecture.

Parameters:

a - architecture for the new net

MultiLayerPerceptron

public MultiLayerPerceptron()

Constructor for an empty Neural Net.

Method Detail

setClassMapping

public void setClassMapping(HashMap<DataCell,Integer> map)

Allows for setting the class mapping from output neurons to class values.

Parameters:

map - a HashMap containing the mapping.

setInputMapping

public void setInputMapping(HashMap<String,Integer> map)

Allows for setting the input mapping from input neurons to class values.

Parameters:

map - a HashMap containing the mapping

getClassMapping

public HashMap<DataCell,Integer> getClassMapping()

Allows to get the class mapping from output neurons to class values.

Returns:

a HashMap containing the mapping.

getInputMapping

public HashMap<String,Integer> getInputMapping()

Allows to get the input mapping from input neurons to columns.

Returns:

a HashMap containing the mapping

getLayers

public Layer[] getLayers()

Returns:

the layers of the net.

getLayer

public Layer getLayer(int i)

Returns layer at a given position.

Parameters:

i - position

Returns:

layer at position i

getNrLayers

public int getNrLayers()

Returns:

number of Layers in MLP.

setLayers

public void setLayers(Layer[] layers)

Sets all layers.

Parameters:

layers - new layers for the net

setLayer

public void setLayer(int i,
                     Layer layer)

Sets the layer at a given position.

Parameters:

i - position

layer - new layer for position i

output

public double[] output(double[] in)

Evaluates input and returns output of output neurons.

Parameters:

in - input for the mlp

Returns:

output of the output neurons after having processed a forward wave through the net

output

public double[] output(Double[] in)

Computes the output for given input.

Parameters:

in - the input values

Returns:

output values of MLP.

getClassOutput

public String getClassOutput(double[] in)

Finds the winner and returns its class output value.

Parameters:

in - input for the mlp

Returns:

class value

getArchitecture

public Architecture getArchitecture()

Returns:

Architecture of the net

setArchitecture

public void setArchitecture(Architecture architecture)

Sets the architecture.

Parameters:

architecture - Architecture for the net

savePredictorParams

public void savePredictorParams(ModelContentWO predParams)

Stores this MLP model to config.

Parameters:

predParams - ModelContent to write into.

getMode

public int getMode()

Returns:

the mode of the MLP

See Also:

CLASSIFICATION_MODE, REGRESSION_MODE

setMode

public void setMode(int mode)

Sets the mode of the MLP. This can either be CLASSIFICATION_MODE or REGRESSION_MODE, other values are ignored.

Parameters:

mode - the mode of the MLP

See Also:

CLASSIFICATION_MODE, REGRESSION_MODE

loadPredictorParams

public static MultiLayerPerceptron loadPredictorParams(ModelContentRO predParams)
                                                throws InvalidSettingsException

Parameters:

predParams - the ConfigObject containing the model of the mlp

Returns:

a new MultiLayerPerceptron based on the config

Throws:

InvalidSettingsException - if settings are incorrect