/*
* File: PolynomialFunction.java
* Authors: Kevin R. Dixon
* Company: Sandia National Laboratories
* Project: Cognitive Foundry
*
* Copyright September 4, 2007, Sandia Corporation. Under the terms of Contract
* DE-AC04-94AL85000, there is a non-exclusive license for use of this work by
* or on behalf of the U.S. Government. Export of this program may require a
* license from the United States Government. See CopyrightHistory.txt for
* complete details.
*
*/
package gov.sandia.cognition.learning.function.scalar;
import gov.sandia.cognition.annotation.PublicationReference;
import gov.sandia.cognition.annotation.PublicationType;
import gov.sandia.cognition.learning.algorithm.SupervisedBatchLearner;
import gov.sandia.cognition.learning.algorithm.gradient.ParameterGradientEvaluator;
import gov.sandia.cognition.learning.algorithm.minimization.line.InputOutputSlopeTriplet;
import gov.sandia.cognition.learning.algorithm.regression.LinearRegression;
import gov.sandia.cognition.learning.data.DefaultInputOutputPair;
import gov.sandia.cognition.learning.data.InputOutputPair;
import gov.sandia.cognition.learning.function.vector.ScalarBasisSet;
import gov.sandia.cognition.math.AbstractDifferentiableUnivariateScalarFunction;
import gov.sandia.cognition.math.DifferentiableUnivariateScalarFunction;
import gov.sandia.cognition.math.matrix.VectorFactory;
import gov.sandia.cognition.math.matrix.Vector;
import gov.sandia.cognition.util.AbstractCloneableSerializable;
import java.util.ArrayList;
import java.util.Collection;
/**
* A single polynomial term specified by a real-valued exponent. Evaluates for
* "y" such that y=x^a, where "x" is the input and "a" is the real-valued
* exponent.
*
* @author Kevin R. Dixon
* @since 2.0
*
*/
public class PolynomialFunction
extends AbstractDifferentiableUnivariateScalarFunction
implements ParameterGradientEvaluator<Double, Double, Vector>
{
/**
* Real-valued exponent of this polynomial
*/
private double exponent;
/**
* Creates a new instance of PolynomialFunction
* @param exponent
* Real-valued exponent of this polynomial
*/
public PolynomialFunction(
final double exponent )
{
this.setExponent( exponent );
}
/**
* Copy Constructor
* @param other
* PolynomialFunction to copy
*/
public PolynomialFunction(
final PolynomialFunction other )
{
this( other.getExponent() );
}
@Override
public PolynomialFunction clone()
{
return (PolynomialFunction) super.clone();
}
/**
* Returns the value of the exponent
* @return
* Exponent of this polynomial
*/
@Override
public Vector convertToVector()
{
return VectorFactory.getDefault().copyValues( this.getExponent() );
}
/**
* Sets the value of the exponent
* @param parameters
* Exponent of this polynomial
*/
@Override
public void convertFromVector(
final Vector parameters )
{
this.setExponent( parameters.getElement( 0 ) );
}
@Override
public double differentiate(
final double input )
{
// d[x^a]/dx = a*x^{a-1}
double a = this.getExponent();
double dydx = a * Math.pow( input, a - 1.0 );
return dydx;
}
/**
* Evaluates the polynomial for the given input, returning
* Math.pow( input, exponent )
* @param input
* Input about which to compute the polynomial
* @return
* Math.pow(input,exponent)
*/
@Override
public double evaluate(
final double input )
{
// Note that this will fail if "input" is less than zero AND
// "exponent" is negative AND "exponent" isn't an integer:
// for example, Math.pow(-2, -1.1) -> Exception!
return Math.pow( input, this.exponent );
}
@Override
public Vector computeParameterGradient(
final Double input )
{
// We're computing the derivative of:
// d/da[x^a] = ln(x)*x^a
// This is degenerate when x==0.0, and we assume that
// 0.0^a == 0.0 for any a, so that the derivative would be zero
// as well
double x = input;
double a = this.exponent;
double dyda;
if (x == 0.0)
{
dyda = 0.0;
}
else
{
dyda = Math.log( x ) * Math.pow( x, a );
}
Vector gradient = VectorFactory.getDefault().copyValues( dyda );
return gradient;
}
@Override
public String toString()
{
return "x^" + this.getExponent();
}
/**
* Getter for exponent
* @return
* Real-valued exponent of this polynomial
*/
public double getExponent()
{
return this.exponent;
}
/**
* Setter for exponent
* @param exponent
* Real-valued exponent of this polynomial
*/
public void setExponent(
final double exponent )
{
this.exponent = exponent;
}
/**
* Creates an array of PolynomialFunctions from the array of their exponents
* @param polynomialExponents
* Exponents for the various polynomials
* @return
* Array of PolynomialFunctions from the given exponents
*/
public static ArrayList<PolynomialFunction> createPolynomials(
final double... polynomialExponents )
{
int num = polynomialExponents.length;
ArrayList<PolynomialFunction> functions =
new ArrayList<PolynomialFunction>( num );
for (int i = 0; i < num; i++)
{
functions.add( new PolynomialFunction( polynomialExponents[i] ) );
}
return functions;
}
/**
* Describes functionality of a closed-form algebraic polynomial function
*/
public interface ClosedForm
extends DifferentiableUnivariateScalarFunction
{
/**
* Finds the real-valued roots (zero crossings) of the polynomial
* @return
* Array of roots, will never be null
*/
public Double[] roots();
/**
* Finds the real-valued stationary points (zero slope) maxima or minima
* of the polynomial
* @return
* Array of stationary points, will never be null
*/
public Double[] stationaryPoints();
}
/**
* Utilities for algebraic treatment of a linear polynomial of the form
* y(x) = q0 + q1*x
*/
public static class Linear
extends AbstractDifferentiableUnivariateScalarFunction
implements PolynomialFunction.ClosedForm
{
/**
* Constant (zeroth-order) coefficient
*/
private double q0;
/**
* Linear (first-order) coefficient
*/
private double q1;
/**
* Tolerance below which to consider something zero, {@value}
*/
public final static double COLLINEAR_TOLERANCE = 0.0;
/**
* Creates a new instance of Linear
* @param q0
* Constant (zeroth-order) coefficient
* @param q1
* Linear (first-order) coefficient
*/
public Linear(
final double q0,
final double q1 )
{
this.setQ0( q0 );
this.setQ1( q1 );
}
@Override
public Linear clone()
{
return (Linear) super.clone();
}
@Override
public double evaluate(
double input )
{
return this.getQ0() + this.getQ1()*input;
}
@Override
public double differentiate(
double input )
{
return this.getQ1();
}
@Override
public Double[] roots()
{
if( Math.abs(this.getQ1()) <= COLLINEAR_TOLERANCE )
{
return new Double[0];
}
else
{
return new Double[] { -this.getQ0() / this.getQ1() };
}
}
@Override
public Double[] stationaryPoints()
{
// No stationary points
// (unless q1==0.0, but then every point is stationary)
return new Double[0];
}
/**
* Fits a linear (straight-line) curve to the given data points
* @param p0
* First point
* @param p1
* Second point
* @return
* closed-form Linear function representing the data points
*/
public static PolynomialFunction.Linear fit(
final InputOutputPair<Double,Double> p0,
final InputOutputPair<Double,Double> p1 )
{
double x0 = p0.getInput();
double x1 = p1.getInput();
double y0 = p0.getOutput();
double y1 = p1.getOutput();
double denom = x1-x0;
if( Double.isInfinite(y0) )
{
y0 = Math.signum(y0) * Double.MAX_VALUE/10.0;
}
if( Double.isInfinite(y1) )
{
y1 = Math.signum(y1) * Double.MAX_VALUE/10.0;
}
if( Math.abs(denom) <= COLLINEAR_TOLERANCE )
{
throw new IllegalArgumentException(
"Linear interpolation points are effectively collinear: " + denom );
}
// y = mx+b
// Solving for m = (y1-y0) / (x1-x0) ("rise over run")
// Solving for b = y-mx ({x0,y0} or {x1,y1} will yield same result)
double q1 = (y1-y0) / denom;
double q0 = y0 - q1*x0;
return new PolynomialFunction.Linear( q0, q1 );
}
/**
* Fits a linear (stright-line) curve to the given data point
* @param p0
* First point
* @return
* closed-form Linear function representing the data points
*/
public static PolynomialFunction.Linear fit(
final InputOutputSlopeTriplet p0 )
{
double x0 = p0.getInput();
double y0 = p0.getOutput();
double q1 = p0.getSlope();
double q0 = y0 - q1*x0;
return new PolynomialFunction.Linear( q0, q1 );
}
/**
* Getter for q0
* @return
* Zeroth order coefficient
*/
public double getQ0()
{
return this.q0;
}
/**
* Setter for q0
* @param q0
* Zeroth order coefficient
*/
public void setQ0(
final double q0 )
{
this.q0 = q0;
}
/**
* Getter for q1
* @return
* First-order coefficient
*/
public double getQ1()
{
return this.q1;
}
/**
* Setter for q1
* @param q1
* First-order coefficient
*/
public void setQ1(
final double q1 )
{
this.q1 = q1;
}
@Override
public String toString()
{
return "f(x) = " + this.getQ0() + " + " + this.getQ1() + "x";
}
}
/**
* Utilities for algebraic treatment of a quadratic polynomial of the form
* y(x) = q0 + q1*x + q2*x^2.
*/
public static class Quadratic
extends Linear
{
/**
* Quadratic (second-order) coefficient
*/
private double q2;
/**
* Creates a new instance of Quadratic
* @param q0
* Constant (zeroth-order) coefficient
* @param q1
* Linear (first-order) coefficient
* @param q2
* Quadratic (second-order) coefficient
*/
public Quadratic(
final double q0,
final double q1,
final double q2 )
{
super( q0, q1 );
this.setQ2( q2 );
}
/**
* Copy constructor
* @param other
* Quadratic to copy
*/
public Quadratic(
Quadratic other )
{
this( other.getQ0(), other.getQ1(), other.getQ2() );
}
@Override
public Quadratic clone()
{
return (Quadratic) super.clone();
}
@Override
public String toString()
{
return super.toString() + " + " + this.getQ2() + "x^2";
}
@Override
public double evaluate(
double input )
{
return Quadratic.evaluate( input, this.getQ0(), this.getQ1(), this.getQ2() );
}
@Override
public double differentiate(
final double input )
{
// dy/dx = q1 + 2*q2*x
return this.getQ1() + 2.0*this.getQ2()*input;
}
/**
* Finds the roots (zero-crossings) of the quadratic, which has at most
* two, but possibly one or zero
* @return
* Array of roots
*/
@Override
public Double[] roots()
{
return Quadratic.roots( this.getQ0(), this.getQ1(), this.getQ2() );
}
/**
* Finds the real-valued stationary points (zero-derivatives) of the
* quadratic. A quadratic has at most one stationary point, it may be
* a minimum or maximum.
* @return
* Zero- or One-length array of stationary points
*/
@Override
public Double[] stationaryPoints()
{
return Quadratic.stationaryPoints( this.getQ0(), this.getQ1(), this.getQ2() );
}
/**
* Evaluates a quadratic polynomial of the form
* y(x) = q0 + q1*x + q2*x^2 for a given value of "x"
* @param x
* Value at which to evaluate the polynomial
* @param q0
* Constant-term coefficient
* @param q1
* Linear-term coefficient
* @param q2
* Quadratic-term coefficient
* @return
* Value of the polynomial at "x"
*/
public static double evaluate(
final double x,
final double q0,
final double q1,
final double q2 )
{
return q0 + x*(q1 + x*q2);
}
/**
* Fits a quadratic to three points
* @param p0
* First point
* @param p1
* Second point
* @param p2
* Third point
* @return
* Quadratic fitting the three points
*/
public static Quadratic fit(
final InputOutputPair<Double,Double> p0,
final InputOutputPair<Double,Double> p1,
final InputOutputPair<Double,Double> p2 )
{
double x0 = p0.getInput();
double x1 = p1.getInput();
double x2 = p2.getInput();
double x02 = x0*x0;
double x12 = x1*x1;
double x22 = x2*x2;
double y0 = p0.getOutput();
double y1 = p1.getOutput();
double y2 = p2.getOutput();
// This was computed with MATLAB's symbolic toolbox:
// >> syms x0 x1 x2 y0 y1 y2
// >> A = [ x0^2 x0 1; x1^2 x1 1; x2^2 x2 1 ];
// >> y = [ y0; y1; y2 ];
// >> factor(A\y)
// ans=
// -(y0*x1-y0*x2-y2*x1+x0*y2-x0*y1+x2*y1)/(x0-x2)/(x1-x2)/(-x0+x1)
// (-x1^2*y2+x1^2*y0-x2^2*y0-y1*x0^2+y1*x2^2+y2*x0^2)/(x0-x2)/(x1-x2)/(-x0+x1)
// -(-x2^2*y0*x1+x2^2*x0*y1+x1^2*y0*x2-y1*x0^2*x2+y2*x0^2*x1-x1^2*x0*y2)/(x0-x2)/(x1-x2)/(-x0+x1)
double denom = (x0-x2) * (x1-x2) * (x1-x0);
if( Math.abs( denom ) <= COLLINEAR_TOLERANCE )
{
throw new IllegalArgumentException(
"Parabolic interpolation points are effectively collinear: " + denom );
}
double v2 = x0*(y1-y2) + x1*(y2-y0) + x2*(y0-y1);
double v1 = x02*(y2-y1) + x12*(y0-y2) + x22*(y1-y0);
double v0 = x02*(x2*y1-x1*y2) + x12*(x0*y2-x2*y0) + x22*(x1*y0-x0*y1);
double q2 = v2 / denom;
double q1 = v1 / denom;
double q0 = v0 / denom;
return new Quadratic( q0, q1, q2 );
}
/**
* Fits a quadratic to two points, one of which has slope information.
* @param p0
* @param p1
* @return The quadratic fit.
*/
public static Quadratic fit(
final InputOutputSlopeTriplet p0,
final InputOutputPair<Double, Double> p1 )
{
// These are from the MATLAB command:
// >> syms x0 x1 x2 y0 y1 m0
// >> A = [ x0^2 x0 1; 2*x0 1 0; x1^2 x1 1 ]
// >> y = [ y0; m0; y1 ]
// >> factor(A\y)
// ans =
// -(-x0*m0+y0+x1*m0-y1)/(-x0+x1)^2
// (-x0^2*m0+2*y0*x0-2*x0*y1+m0*x1^2)/(-x0+x1)^2
// (x0^2*x1*m0+y1*x0^2-x0*m0*x1^2+x1^2*y0-2*y0*x0*x1)/(-x0+x1)^2
double x0 = p0.getInput();
double x1 = p1.getInput();
double x02 = x0*x0;
double x12 = x1*x1;
double y0 = p0.getOutput();
double m0 = p0.getSlope();
double y1 = p1.getOutput();
double dy = y0-y1;
double dx = x0-x1;
double denom = dx*dx;
if( Math.abs( denom ) <= COLLINEAR_TOLERANCE )
{
throw new IllegalArgumentException(
"Parabolic interpolation points are effectively collinear: " + denom );
}
double v2 = m0*dx-dy;
double v1 = m0*(x12-x02) + 2.0*x0*dy;
double v0 = x02*(y1+x1*m0) + x12*(y0-x0*m0) - 2.0*y0*x0*x1;
double q2 = v2 / denom;
double q1 = v1 / denom;
double q0 = v0 / denom;
return new Quadratic( q0, q1, q2 );
}
/**
* Finds the roots of the quadratic equation using the quadratic
* formula. That is, finding the values of "x" such that
* y(x) = q0 + q1*x + q2*x^2 = 0.0.
* There will be at most two roots, but there can also be a single root,
* or no roots. In the case of two roots, the return Pair will have
* the "x" value for value. In the case of a single root, the Pair
* will have an "x" value for the First, but null for the second. In
* the case when there are no REAL roots, the return value will be null.
* @param q0
* Constant-term coefficient
* @param q1
* Linear-term coefficient
* @param q2
* Quadratic-term coefficient
* @return
* In the case of two roots, the return Pair will have
* the "x" value for value. In the case of a single root, the Pair
* will have an "x" value for the First, but null for the second. In
* the case when there are no REAL roots, the return value will be a
* zero-length array.
*
*/
@PublicationReference(
author="Wikipedia",
title="Quadratic formula",
type=PublicationType.WebPage,
year=2008,
url="http://en.wikipedia.org/wiki/Quadratic_formula#Quadratic_formula"
)
public static Double[] roots(
final double q0,
final double q1,
final double q2 )
{
// If there's no quadratic term, then just solve the linear quation
// such that x = -q0/q1
if( Math.abs(q2) <= COLLINEAR_TOLERANCE )
{
// There are no roots because this equation is a constant q0
if( Math.abs(q1) <= COLLINEAR_TOLERANCE )
{
return new Double[0];
}
else
{
double xstar = -q0/q1;
return new Double[]{ xstar };
}
}
double discriminant = q1*q1 - 4.0*q2*q0;
// Here there are only complex roots, so just return nothing
if( discriminant < 0.0 )
{
return new Double[0];
}
// One repeated root, so just return that one
else if( Math.abs(discriminant) <= COLLINEAR_TOLERANCE )
{
// There's no way that q2 can be effectively zero, as that
// was caught in the first if-statement
double xstar = -0.5 * q1 / q2;
return new Double[]{ xstar };
}
// Here we've got two real roots!
else
{
// There's no way that q2 can be effectively zero, as that
// was caught in the first if-statement
double denom = -2.0*q2;
double sqrtdisc = Math.sqrt( discriminant );
double xpos = (q1+sqrtdisc) / denom;
double xneg = (q1-sqrtdisc) / denom;
return new Double[]{ xpos, xneg };
}
}
/**
* Finds the stationary point of the quadratic equation. That is,
* the point when the derivative f'(x)=0.0
* @param q0
* Constant-term coefficient
* @param q1
* Linear-term coefficient
* @param q2
* Quadratic-term coefficient
* @return
* Value of "x" when the derivative is zero, null when none is found
*/
public static Double[] stationaryPoints(
final double q0,
final double q1,
final double q2 )
{
// The derivative of the quadratic is
// f'(x) = q1 + 2.0*q2*x
// The stationary point is when f'(x)=0.0
// No quadratic term means there is no stationary point
if( Math.abs(q2) <= COLLINEAR_TOLERANCE )
{
return new Double[0];
}
return new Double[]{ q1 / (-2.0*q2) };
}
/**
* Getter for q2
* @return
* Second-order coefficient
*/
public double getQ2()
{
return this.q2;
}
/**
* Setter for q2
* @param q2
* Second-order coefficient
*/
public void setQ2(
double q2 )
{
this.q2 = q2;
}
}
/**
* Algebraic treatment for a polynomial of the form
* y(x) = q0 + q1*x + q2*x^2 + q3*x^3
*/
public static class Cubic
extends Quadratic
implements PolynomialFunction.ClosedForm
{
/**
* Cubic (third-order) coefficient
*/
private double q3;
/**
* Creates a new instance of Quadratic
* @param q0
* Constant (zeroth-order) coefficient
* @param q1
* Linear (first-order) coefficient
* @param q2
* Quadratic (second-order) coefficient
* @param q3
* Cubic (third-order) coefficient
*/
public Cubic(
final double q0,
final double q1,
final double q2,
final double q3 )
{
super( q0, q1, q2 );
this.setQ3( q3 );
}
/**
* Copy constructor
* @param other
* Cubic to copy
*/
public Cubic(
final Cubic other )
{
super( other );
this.setQ3( other.getQ3() );
}
@Override
public Cubic clone()
{
return (Cubic) super.clone();
}
@Override
public String toString()
{
return super.toString() + " + " + this.getQ3() + "x^3";
}
@Override
public double evaluate(
final double input )
{
return Cubic.evaluate( input, this.getQ0(), this.getQ1(), this.getQ2(), this.getQ3() );
}
@Override
public double differentiate(
final double input )
{
return Quadratic.evaluate( input, this.getQ1(), 2.0*this.getQ2(), 3.0*this.getQ3());
}
@Override
public Double[] roots()
{
throw new IllegalArgumentException(
"Not yet implemented" );
}
@Override
public Double[] stationaryPoints()
{
return Cubic.stationaryPoints( this.getQ0(), this.getQ1(), this.getQ2(), this.getQ3() );
}
/**
* Evaluates a quadratic polynomial of the form
* y(x) = q0 + q1*x + q2*x^2 + q3*x^3 for a given value of "x"
* @param x
* Value at which to evaluate the polynomial
* @param q0
* Constant-term coefficient
* @param q1
* Linear-term coefficient
* @param q2
* Quadratic-term coefficient
* @param q3
* Cubic-term coefficient
* @return
* Value of the polynomial at "x"
*/
public static double evaluate(
final double x,
final double q0,
final double q1,
final double q2,
final double q3 )
{
return q0 + x*(q1 + x*(q2 + x*q3));
}
/**
*
* Finds the stationary point of the quadratic equation. That is,
* the point when the derivative f'(x)=0.0
* @param q0
* Constant-term coefficient
* @param q1
* Linear-term coefficient
* @param q2
* Quadratic-term coefficient
* @param q3
* Cubic-term coefficient
* @return
* Value of "x" when the derivative is zero, null when none is found
*/
public static Double[] stationaryPoints(
final double q0,
final double q1,
final double q2,
final double q3 )
{
// The derivative is given as:
// f'(x) = q1 + 2.0*q2*x + 3.0*q3*x^2
// Then we can just use the quadratic root finder
double p0 = q1;
double p1 = 2.0 * q2;
double p2 = 3.0 * q3;
return Quadratic.roots( p0, p1, p2);
}
/**
* Getter for q3
* @return
* Cubic (third-order) coefficient
*/
public double getQ3()
{
return this.q3;
}
/**
* Setter for q3
* @param q3
* Cubic (third-order) coefficient
*/
public void setQ3(
final double q3 )
{
this.q3 = q3;
}
/**
* Fits a cubic to two InputOutputSlopeTriplets using a closed-form
* solution
* @param p0
* First point
* @param p1
* Second point
* @return
* Cubic fitting the points
*/
public static Cubic fit(
final InputOutputSlopeTriplet p0,
final InputOutputSlopeTriplet p1 )
{
// From the MATLAB symbolic toolbox command sequence:
// >> syms x0 x1 y0 y1 m0 m1
// >> A = [ x0^3 x0^2 x0 1; 3*x0^2 2*x0 1 0; x1^3 x1^2 x1 1; 3*x1^2 2*x1 1 0 ]
// >> y = [ y0; m0; y1; m1 ]
// >> factor(A\y)
// ans =
// (x1*m1-2*y1-x0*m1-x0*m0+x1*m0+2*y0)/(-x0+x1)^3
// -(2*m0*x1^2+x1^2*m1+3*y0*x1-x0*x1*m0-3*x1*y1+x1*x0*m1-2*x0^2*m1+3*y0*x0-x0^2*m0-3*x0*y1)/(-x0+x1)^3
// (2*x1^2*x0*m1+x0*m0*x1^2+x1^3*m0-x0^2*x1*m1+6*y0*x0*x1-2*x0^2*x1*m0-6*x1*x0*y1-x0^3*m1)/(-x0+x1)^3
// (-x1^3*x0*m0+x1^3*y0-x1^2*x0^2*m1-3*y0*x0*x1^2+x1^2*x0^2*m0+x1*m1*x0^3-x0^3*y1+3*y1*x0^2*x1)/(-x0+x1)^3
double x0 = p0.getInput();
double x1 = p1.getInput();
double x02 = x0*x0;
double x12 = x1*x1;
double y0 = p0.getOutput();
double m0 = p0.getSlope();
double y1 = p1.getOutput();
double m1 = p1.getSlope();
double dy = y1-y0;
double dx = x1-x0;
double m1pm0 = m1+m0;
double denom = dx*dx*dx;
if( Math.abs( denom ) <= COLLINEAR_TOLERANCE )
{
throw new IllegalArgumentException(
"Cubic interpolation points are effectively collinear: " + denom );
}
double v3 = dx*m1pm0 - 2.0*dy;
double v2 = x02*(m1+m1pm0) - x12*(m0+m1pm0) + 3.0*(x1+x0)*dy - x0*x1*(m1-m0);
double v1 = x12*(x0*(m1+m1pm0)+x1*m0) - x02*(x1*(m0+m1pm0)+x0*m1) - 6.0*x0*x1*dy;
double v0 = x12*(x1*(y0-x0*m0)+x0*(x0*(m0-m1)-3.0*y0)) + x02*(x0*(x1*m1-y1)+3.0*y1*x1);
double q3 = v3 / denom;
double q2 = v2 / denom;
double q1 = v1 / denom;
double q0 = v0 / denom;
return new Cubic( q0, q1, q2, q3 );
}
}
/**
* Performs Linear Regression using an arbitrary set of
* PolynomialFunction basis functions
*/
public static class Regression
extends AbstractCloneableSerializable
implements SupervisedBatchLearner<Double, Double, VectorFunctionLinearDiscriminant<Double>>
{
/**
* Polynomials to use in the regression
*/
private ScalarBasisSet<Double> polynomials;
/**
* Creates a new instance of Regression
* @param polynomialExponents
* Set of polynomial exponents to use during the regression
*/
public Regression(
final double... polynomialExponents )
{
this.setPolynomials( new ScalarBasisSet<Double>(
PolynomialFunction.createPolynomials( polynomialExponents ) ) );
}
/**
* Performs LinearRegression using all integer-exponent polynomials
* less than or equal to the maxOrder
* @param maxOrder
* Uses all polynomials below the maxOrder: a0*x^0 + a1*x^1 + ... am*a^m
* @param data
* Data set to use for the LinearRegression
* @return
* LinearCombinationFunction that combines the desired
* PolynomialFunctions with weighting coefficients determined by
* the LinearRegression algorithm
*/
public static VectorFunctionLinearDiscriminant<Double> learn(
final int maxOrder,
final Collection<? extends InputOutputPair<Double, Double>> data )
{
// We don't need to include a polynomial with exponent 0 because
// LinearRegression already estimates a bias
double[] polynomialExponents = new double[maxOrder];
for (int i = 0; i < polynomialExponents.length; i++)
{
polynomialExponents[i] = i+1;
}
PolynomialFunction.Regression r =
new PolynomialFunction.Regression( polynomialExponents );
return r.learn( data );
}
@Override
public VectorFunctionLinearDiscriminant<Double> learn(
final Collection<? extends InputOutputPair<? extends Double, Double>> data)
{
// The first task is to create the Vector-space representation
ArrayList<InputOutputPair<Vector,Double>> vectorData =
new ArrayList<InputOutputPair<Vector, Double>>( data.size() );
for( InputOutputPair<? extends Double,Double> pair : data )
{
Vector phi = this.polynomials.evaluate( pair.getInput() );
vectorData.add( DefaultInputOutputPair.create( phi, pair.getOutput() ) );
}
LinearRegression regression = new LinearRegression();
LinearDiscriminantWithBias linearResult = regression.learn(vectorData);
return new VectorFunctionLinearDiscriminant<Double>(
this.getPolynomials(), linearResult );
}
/**
* Getter for polynomials
* @return
* Polynomials to use in the regression
*/
public ScalarBasisSet<Double> getPolynomials()
{
return this.polynomials;
}
/**
* Setter for polynomials
* @param polynomials
* Polynomials to use in the regression
*/
public void setPolynomials(
ScalarBasisSet<Double> polynomials)
{
this.polynomials = polynomials;
}
}
}