jsat.math.optimization

Class NelderMead

java.lang.Object

jsat.math.optimization.NelderMead

All Implemented Interfaces:

Serializable, Optimizer

public class NelderMead
extends Object
implements Optimizer

The Nelder-Mean algorithm is a simple directed search method. As such, it does not need any information about the target functions derivative, or any data points. To perform best, the Nelder-Mean method needs N+1 reasonable initial guesses for an N dimensional problem.
The Nelder-Mean method has the advantage that the only information it needs about the function it is going to minimize, is the function itself.

Author:

Edward Raff

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
NelderMead()

Method Summary

Modifier and Type	Method and Description
Vec	optimize(double eps, int iterationLimit, Function f, Function fd, Vec vars, List<Vec> inputs, Vec outputs) Performs optimization on the given inputs to find the minima of the function.
Vec	optimize(double eps, int iterationLimit, Function f, Function fd, Vec vars, List<Vec> inputs, Vec outputs, ExecutorService threadpool) Performs optimization on the given inputs to find the minima of the function.
Vec	optimize(double eps, int iterationLimit, Function f, List<Vec> initalPoints) Attempts to find the minimal value of the given function.
void	setContraction(double contraction) Sets the contraction constant, which must be in the range (0, 1)
void	setExpansion(double expansion) Sets the expansion constant, which must be greater than 1 and the reflection constant
void	setReflection(double reflection) Sets the reflection constant, which must be greater than 0
void	setShrink(double shrink) Sets the shrinkage constant, which must be in the range (0, 1)

Methods inherited from class java.lang.Object

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

Constructor Detail

NelderMead

public NelderMead()

Method Detail

setReflection

public void setReflection(double reflection)

Sets the reflection constant, which must be greater than 0

Parameters:

reflection - the reflection constant

setExpansion

public void setExpansion(double expansion)

Sets the expansion constant, which must be greater than 1 and the reflection constant

Parameters:

expansion -

setContraction

public void setContraction(double contraction)

Sets the contraction constant, which must be in the range (0, 1)

Parameters:

contraction - the contraction constant

setShrink

public void setShrink(double shrink)

Sets the shrinkage constant, which must be in the range (0, 1)

Parameters:

shrink -

optimize

public Vec optimize(double eps,
                    int iterationLimit,
                    Function f,
                    Function fd,
                    Vec vars,
                    List<Vec> inputs,
                    Vec outputs,
                    ExecutorService threadpool)

Description copied from interface: Optimizer

Performs optimization on the given inputs to find the minima of the function. Not all optimization functions require all the parameters to work, and may except some null values.

Specified by:

optimize in interface Optimizer

Parameters:

eps - the desired accuracy of the result.

iterationLimit - the maximum number of iteration steps to allow. This value must be positive

f - the function to optimize. This value can not be null

fd - the derivative of the function to optimize

vars - contains the initial estimate of the minima. The length should be equal to the number of variables being solved for. This value may be altered. This value can not be null.

inputs - a list of input data point values to learn from

outputs - a vector containing the true values for each data point in inputs

threadpool - a source of threads to perform computation with

Returns:

the computed value for the optimization.

optimize

public Vec optimize(double eps,
                    int iterationLimit,
                    Function f,
                    Function fd,
                    Vec vars,
                    List<Vec> inputs,
                    Vec outputs)

Description copied from interface: Optimizer

Performs optimization on the given inputs to find the minima of the function. Not all optimization functions require all the parameters to work, and may except some null values.

Specified by:

optimize in interface Optimizer

Parameters:

eps - the desired accuracy of the result.

iterationLimit - the maximum number of iteration steps to allow. This value must be positive

f - the function to optimize. This value can not be null

fd - the derivative of the function to optimize

vars - contains the initial estimate of the minima. The length should be equal to the number of variables being solved for. This value may be altered. This value can not be null.

inputs - a list of input data point values to learn from

outputs - a vector containing the true values for each data point in inputs

Returns:

the computed value for the optimization.

optimize

public Vec optimize(double eps,
                    int iterationLimit,
                    Function f,
                    List<Vec> initalPoints)

Attempts to find the minimal value of the given function.

Parameters:

eps - the desired accuracy of the result.

iterationLimit - the maximum number of iteration steps to allow. This value must be positive

f - the function to optimize. This value can not be null

initalPoints - the list of initial guess points. If too small, new ones will be generated. if too large, the extra ones will be ignored. This list may not be empty

Returns:

the computed value for the optimization.