* In this implementation object of type
* CumulativeDiscreteDistribution are immutable.
*
* This implementation is part of the master's thesis: "Studie * en implementatie van instantie-gebaseerde algoritmen voor gesuperviseerd * rangschikken", Stijn Lievens, Ghent University, 2004. *
* * @author Stijn Lievens (stijn.lievens@ugent.be) * @version $Revision: 5922 $ */ public class CumulativeDiscreteDistribution implements Serializable, RevisionHandler { /** for serialization */ private static final long serialVersionUID = -2959806903004453176L; /** * small tolerance to account for rounding errors when working * with doubles */ private static final double TOLERANCE = Utils.SMALL; /** The cumulative probabilities */ private double[] m_cdf; /** * Create a discrete cumulative probability distribution based on a * DiscreteEstimator.
*
* @param e the DiscreteEstimator on which the
* cumulative probability distribution will be based
*/
public CumulativeDiscreteDistribution(DiscreteEstimator e) {
m_cdf = new double[e.getNumSymbols()];
if (m_cdf.length != 0) {
m_cdf[0] = e.getProbability(0);
}
for (int i = 1; i < m_cdf.length; i++) {
m_cdf[i] = m_cdf[i - 1] + e.getProbability(i);
}
}
/**
* Create a CumulativeDiscreteDistribution based on a
* DiscreteDistribution.
*
* @param d the DiscreteDistribution on which the
* cumulative probability distribution will be based
*/
public CumulativeDiscreteDistribution(DiscreteDistribution d) {
m_cdf = new double[d.getNumSymbols()];
if (m_cdf.length != 0) {
m_cdf[0] = d.getProbability(0);
}
for (int i = 1; i < m_cdf.length; i++) {
m_cdf[i] = m_cdf[i - 1] + d.getProbability(i);
}
}
/**
* Create a CumulativeDiscreteDistribution based on an
* array of doubles. The array cdf is copied, so
* the caller can reuse the same array.
*
* @param cdf an array that represents a valid discrete cumulative
* probability distribution
* @throws IllegalArgumentException if the array doesn't
* represent a valid cumulative discrete distribution function
*/
public CumulativeDiscreteDistribution(double[] cdf) throws IllegalArgumentException {
if (!validCumulativeDistribution(cdf)) {
throw new IllegalArgumentException
("Not a cumulative probability distribution");
}
m_cdf = new double[cdf.length];
System.arraycopy(cdf, 0, m_cdf, 0, cdf.length);
}
/**
* Get the number of elements over which the cumulative
* probability distribution is defined.
*
* @return the number of elements over which the cumulative
* probability distribution is defined.
*/
public int getNumSymbols() {
return (m_cdf != null) ? m_cdf.length : 0;
}
/**
* Get the probability of finding an element
* smaller or equal than index.
*
* @param index the required index
* @return the probability of finding an element <= index
*/
public double getCumulativeProbability(int index) {
return m_cdf[index];
}
/**
* Get an array representation of the cumulative probability
* distribution.
*
* @return an array of doubles representing the cumulative
* probability distribution
*/
public double[] toArray() {
double cdf[] = new double[m_cdf.length];
System.arraycopy(m_cdf, 0, cdf, 0, cdf.length);
return cdf;
}
/**
* Returns if this is dominated by cdf.
* This means that we check if, for all indices i , it
* holds that this.getProbability(i) >= cdf.getProbability(i).
*
*
* @param cdf the CumulativeDiscreteDistribution
* this is compared to
* @return true if this is dominated by
* cdf , false otherwise
* @throws IllegalArgumentException if the two distributions don't
* have the same length
*/
public boolean stochasticDominatedBy(CumulativeDiscreteDistribution cdf) throws IllegalArgumentException {
if (getNumSymbols() != cdf.getNumSymbols()) {
throw new IllegalArgumentException
("Cumulative distributions are not defined over"
+ " the same number of symbols");
}
for (int i = 0; i < m_cdf.length; i++) {
if (m_cdf[i] < cdf.m_cdf[i]) {
return false;
}
}
return true;
}
/**
* Indicates if the object o equals this.
* Note: for practical reasons I was forced to use a small tolerance
* whilst comparing the distributions, meaning that the transitivity
* property of equals is not guaranteed.
*
* @param o the reference object with which to compare
* @return true if o equals
* this, false otherwise
*/
public boolean equals(Object o) {
if (!(o instanceof CumulativeDiscreteDistribution)) {
return false;
}
CumulativeDiscreteDistribution cdf =
(CumulativeDiscreteDistribution) o;
if (m_cdf.length != cdf.getNumSymbols()) {
return false;
}
for (int i = 0; i < m_cdf.length; i++) {
if (Math.abs(m_cdf[i] - cdf.m_cdf[i]) > TOLERANCE) {
return false;
}
}
return true;
}
/**
* Get a string representation of the cumulative probability
* distribution.
*
* @return a string representation of the distribution.
*/
public String toString() {
// XXX MAYBE WE SHOULD USE STRINGBUFFER AND USE A FIXED
// NUMBER OF DECIMALS BEHIND THE COMMA
String s = "[" + getNumSymbols() + "]:";
for (int i = 0; i < getNumSymbols(); i++)
s += " " + getCumulativeProbability(i);
return s;
}
/**
* Checks if the given array represents a valid cumulative
* distribution.
*
* @param cdf an array holding the presumed cumulative distribution
* @return true if cdf represents
* a valid cumulative discrete distribution function, false
* otherwise
*/
private static boolean validCumulativeDistribution(double[] cdf) {
if (cdf == null || cdf.length == 0 ||
Math.abs(cdf[cdf.length - 1] - 1.) > TOLERANCE || cdf[0] < 0) {
return false;
}
for (int i = 1; i < cdf.length; i++) {
// allow small tolerance for increasing check
if (cdf[i] < cdf[i - 1] - TOLERANCE) {
return false;
}
}
return true;
}
/**
* Returns the revision string.
*
* @return the revision
*/
public String getRevision() {
return RevisionUtils.extract("$Revision: 5922 $");
}
}