ca.nengo.math.impl

Class WeightedCostApproximator.Factory

java.lang.Object

ca.nengo.math.impl.WeightedCostApproximator.Factory

All Implemented Interfaces:

ApproximatorFactory, java.io.Serializable, java.lang.Cloneable

Enclosing class:

WeightedCostApproximator

public static class WeightedCostApproximator.Factory
extends java.lang.Object
implements ApproximatorFactory

An ApproximatorFactory that produces WeightedCostApproximators.

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
WeightedCostApproximator.Factory(float noise)
WeightedCostApproximator.Factory(float noise, boolean quiet)
WeightedCostApproximator.Factory(float noise, int NSV)
WeightedCostApproximator.Factory(float noise, int NSV, boolean quiet)

Method Summary

Methods

Modifier and Type	Method and Description
ApproximatorFactory	clone()
LinearApproximator	getApproximator(float[][][] evaluationSignals, float[][][] values) Similar to getApproximator(float[][], float[][]) but uses evaluation signals and outputs computed over time.
LinearApproximator	getApproximator(float[][] evalPoints, float[][] values)
Function	getCostFunction(int dimension) Note: override to use non-uniform error weighting.
float	getNoise()
int	getNSV()
boolean	getQuiet()
void	setNoise(float noise)
void	setNSV(int nSV)
void	setQuiet(boolean quiet)

Methods inherited from class java.lang.Object

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

Constructor Detail

WeightedCostApproximator.Factory

public WeightedCostApproximator.Factory(float noise)

Parameters:

noise - Random noise to add to component functions (proportion of largest value over all functions)

WeightedCostApproximator.Factory

public WeightedCostApproximator.Factory(float noise,
                                boolean quiet)

Parameters:

noise - Random noise to add to component functions (proportion of largest value over all functions)

quiet - Turn off logging?

WeightedCostApproximator.Factory

public WeightedCostApproximator.Factory(float noise,
                                int NSV)

Parameters:

noise - Random noise to add to component functions (proportion of largest value over all functions)

NSV - Number of singular values to keep

WeightedCostApproximator.Factory

public WeightedCostApproximator.Factory(float noise,
                                int NSV,
                                boolean quiet)

Parameters:

noise - Random noise to add to component functions (proportion of largest value over all functions)

NSV - Number of singular values to keep

quiet - Turn off logging?

Method Detail

getNoise

public float getNoise()

Returns:

Random noise to add to component functions (proportion of largest value over all functions)

setNoise

public void setNoise(float noise)

Parameters:

noise - Random noise to add to component functions (proportion of largest value over all functions)

getNSV

public int getNSV()

Returns:

Maximum number of singular values to use in pseudoinverse of correlation matrix (zero or less means use as many as possible to a threshold magnitude determined by noise).

setNSV

public void setNSV(int nSV)

Parameters:

nSV - Maximum number of singular values to use in pseudoinverse of correlation matrix (zero or less means use as many as possible to a threshold magnitude determined by noise).

getQuiet

public boolean getQuiet()

Returns:

Whether or not information will be printed out to console during make process.

setQuiet

public void setQuiet(boolean quiet)

Parameters:

quiet - Controls whether or not information will be printed out to console during make process.

getApproximator

public LinearApproximator getApproximator(float[][] evalPoints,
                                 float[][] values)

Specified by:

getApproximator in interface ApproximatorFactory

Parameters:

evalPoints - Points at which component functions are evaluated. These should usually be uniformly distributed, because the sum of error at these points is treated as an integral over the domain of interest.

values - The values of component functions at the evalPoints. The first dimension makes up the list of functions, and the second the values of these functions at each evaluation point.

Returns:

A LinearApproximator that can be used to approximate new Functions as a wieghted sum of the given components.

See Also:

ApproximatorFactory.getApproximator(float[][], float[][])

getApproximator

public LinearApproximator getApproximator(float[][][] evaluationSignals,
                                 float[][][] values)

Similar to getApproximator(float[][], float[][]) but uses evaluation signals and outputs computed over time.

Parameters:

evaluationSignals - Signals over which component functions are evaluated. First dimension is the signal, second is the dimension, and third is time.

values - values of component functions over the evaluation signals. First dimension is the component, second is the signal, and third is time.

Returns:

A LinearApproximator that can be used to approximate new Functions as a weighted sum of the given components.

getCostFunction

public Function getCostFunction(int dimension)

Note: override to use non-uniform error weighting.

Parameters:

dimension - Dimension of the function to be approximated

Returns:

A function over the input space that defines relative importance of error at each point (defaults to a ConstantFunction)

clone

public ApproximatorFactory clone()
                          throws java.lang.CloneNotSupportedException

Specified by:

clone in interface ApproximatorFactory

Overrides:

clone in class java.lang.Object

Returns:

Valid clone

Throws:

java.lang.CloneNotSupportedException - if clone can't be made