org.encog.neural.networks.training.pnn
Class GlobalMinimumSearch
java.lang.Object
org.encog.neural.networks.training.pnn.GlobalMinimumSearch
public class GlobalMinimumSearch
- extends Object
Search sigma's for a global minimum. First do a rough search, and then use
the "Brent Method" to refine the search for an optimal sigma. This class uses
the same sigma for each kernel. Multiple sigmas will be introduced in a later
step.
Some of the algorithms in this class are based on C++ code from:
Advanced Algorithms for Neural Networks: A C++ Sourcebook by Timothy Masters
John Wiley & Sons Inc (Computers); April 3, 1995 ISBN: 0471105880
Field Summary |
static double |
CGOLD
The golden section. |
Method Summary |
double |
brentmin(int maxIterations,
double maxError,
double eps,
double tol,
CalculationCriteria network,
double y)
Use the "Brent Method" to find a better minimum. |
void |
findBestRange(double low,
double high,
int numberOfPoints,
boolean useLog,
double minError,
CalculationCriteria network)
Find the best common gamma. |
double |
getX1()
|
double |
getX2()
|
double |
getX3()
|
double |
getY1()
|
double |
getY2()
|
double |
getY3()
|
void |
setX1(double x1)
|
void |
setX2(double x2)
|
void |
setX3(double x3)
|
void |
setY1(double y1)
|
void |
setY2(double y2)
|
void |
setY3(double y3)
|
Methods inherited from class java.lang.Object |
clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait |
CGOLD
public static final double CGOLD
- The golden section.
- See Also:
- Constant Field Values
GlobalMinimumSearch
public GlobalMinimumSearch()
brentmin
public double brentmin(int maxIterations,
double maxError,
double eps,
double tol,
CalculationCriteria network,
double y)
- Use the "Brent Method" to find a better minimum.
- Parameters:
maxIterations
- THe maximum number of iterations.maxError
- We can stop if we reach this error.eps
- The approximate machine precision.tol
- Brent's tolerance, must be >= sqrt( eps )network
- The network to obtain the error from.y
- The error at x2.
- Returns:
- The best error.
findBestRange
public void findBestRange(double low,
double high,
int numberOfPoints,
boolean useLog,
double minError,
CalculationCriteria network)
- Find the best common gamma. Use the same gamma for all kernels. This is a
crude brute-force search. The range found should be refined using the
"Brent Method", also provided in this class.
- Parameters:
low
- The low gamma to begin the search with.high
- The high gamma to end the search with.numberOfPoints
- The number of points between the low and high. Set this value
to negative to prevent the first point from being calculated.
If you do set this to negative, set x2 and y2 to the correct
values.useLog
- Should we progress "logarithmically" from low to high.minError
- We are done if the error is below this.network
- The network to evaluate.
getX1
public double getX1()
- Returns:
- X1, which is a gamma to the left(lower) of the best(middle)
gamma.
getX2
public double getX2()
- Returns:
- X2, which is the middle(best) gamma.
getX3
public double getX3()
- Returns:
- X3, which is a gamma to the right(higher) of the middle(best)
gamma.
getY1
public double getY1()
- Returns:
- Y1, which is the value y1 is the error for x1.
getY2
public double getY2()
- Returns:
- Y2, which is the value y2 is the error for x2. This is the
best(middle) error.
getY3
public double getY3()
- Returns:
- Y3, which is the value y1 is the error for x1.
setX1
public void setX1(double x1)
- Parameters:
x1
- Set X1, which is a gamma to the left(lower) of the
best(middle) gamma.
setX2
public void setX2(double x2)
- Parameters:
x2
- Set X2, which is the middle(best) gamma.
setX3
public void setX3(double x3)
- Parameters:
x3
- Set X3, which is a gamma to the right(higher) of the
middle(best) gamma.
setY1
public void setY1(double y1)
- Parameters:
y1
- Set Y1, which is the value y1 is the error for x1.
setY2
public void setY2(double y2)
- Parameters:
y2
- Set Y2, which is the value y2 is the error for x2. This is the
best(middle) error.
setY3
public void setY3(double y3)
- Parameters:
y3
- Set Y3, which is the value y3 is the error for x3.
Copyright © 2014. All Rights Reserved.