org.joone.engine
Class ExtendedKalmanFilterFFN

java.lang.Object
  org.joone.engine.ExtendedKalmanFilterFFN

public class ExtendedKalmanFilterFFN
extends java.lang.Object

Implements the extended Kalman filter (EKF) as described in "Using an extended Kalman filter learning algorithm for feed-forward neural networks to describe tracer correlations" by Lary and Mussa (2004) in order to train a feed-forward neural network. This requires some matrix algebra for which colt is utilised.

Author:

mg

Field Summary

protected double	kalmanBoost A boost factor used to dampen the Kalman gain
protected java.util.List<NodesAndWeights.Node>	nodes The list of nodes we use in the filter
protected NodesAndWeights	nodesAndWeights The network structure
protected DoubleMatrix2D	P The P matrix
protected double	p0 The initial (diagonal) values to enter into the P matrix
protected int	patternCount The current pattern count
protected int	processorCount Number of processors to use, 2 or more indicates SMP
protected DoubleMatrix2D	Q The Q matrix
protected double	qDecay The Q decay factor
protected DoubleMatrix2D	R The R matrix
protected double	rDecay The R decay factor
protected int	verbose Zero for no verbosity, otherwise the number of patterns after which we croak
protected DoubleMatrix1D	W The weight estimate
protected java.util.List<NodesAndWeights.Weight>	weights The list of weights that we use in the filter

Constructor Summary

ExtendedKalmanFilterFFN(NodesAndWeights nodesAndWeights, double kalmanBoost, int verbose)
Create a new instance of the filter

ExtendedKalmanFilterFFN(NodesAndWeights nodesAndWeights, double kalmanBoost, int verbose, int processorCount)
Create a new instance of the filter - use for SMP

Method Summary

protected void	init() Initialise the filter
void	update(double[] error) Update the filter.

Methods inherited from class java.lang.Object

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

Field Detail

nodesAndWeights

protected NodesAndWeights nodesAndWeights

The network structure

W

protected DoubleMatrix1D W

The weight estimate

P

protected DoubleMatrix2D P

The P matrix

p0

protected double p0

The initial (diagonal) values to enter into the P matrix

Q

protected DoubleMatrix2D Q

The Q matrix

qDecay

protected double qDecay

The Q decay factor

R

protected DoubleMatrix2D R

The R matrix

rDecay

protected double rDecay

The R decay factor

patternCount

protected int patternCount

The current pattern count

kalmanBoost

protected double kalmanBoost

A boost factor used to dampen the Kalman gain

nodes

protected java.util.List<NodesAndWeights.Node> nodes

The list of nodes we use in the filter

weights

protected java.util.List<NodesAndWeights.Weight> weights

The list of weights that we use in the filter

verbose

protected int verbose

Zero for no verbosity, otherwise the number of patterns after which we croak

processorCount

protected int processorCount

Number of processors to use, 2 or more indicates SMP

Constructor Detail

ExtendedKalmanFilterFFN

public ExtendedKalmanFilterFFN(NodesAndWeights nodesAndWeights,
                               double kalmanBoost,
                               int verbose)

Create a new instance of the filter

Parameters:

nodesAndWeights - the network to be optimised's structure

kalmanBoost - the boost factor (multiplier) to use

verbose - zero for no verbosity or the number of patterns after which we croak

ExtendedKalmanFilterFFN

public ExtendedKalmanFilterFFN(NodesAndWeights nodesAndWeights,
                               double kalmanBoost,
                               int verbose,
                               int processorCount)

Create a new instance of the filter - use for SMP

Parameters:

nodesAndWeights - the network to be optimised's structure

kalmanBoost - the boost factor (multiplier) to use

verbose - zero for no verbosity or the number of patterns after which we croak

processorCount - number of processors to use

Method Detail

init

protected void init()

Initialise the filter

update

public void update(double[] error)

Update the filter. Call this with the most recent error pattern as soon as one becomes available.

Parameters:

error - the most recently seen error pattern