cc.mallet.util.Maths Maven / Gradle / Ivy
Show all versions of mallet Show documentation
/* Copyright (C) 2002 Univ. of Massachusetts Amherst, Computer Science Dept.
This file is part of "MALLET" (MAchine Learning for LanguagE Toolkit).
http://www.cs.umass.edu/~mccallum/mallet
This software is provided under the terms of the Common Public License,
version 1.0, as published by http://www.opensource.org. For further
information, see the file `LICENSE' included with this distribution. */
/**
@author Andrew McCallum [email protected]
*/
package cc.mallet.util;
// Math and statistics functions
public final class Maths {
// From libbow, dirichlet.c
// Written by Tom Minka
public static final double logGamma (double x)
{
double result, y, xnum, xden;
int i;
final double d1 = -5.772156649015328605195174e-1;
final double p1[] = {
4.945235359296727046734888e0, 2.018112620856775083915565e2,
2.290838373831346393026739e3, 1.131967205903380828685045e4,
2.855724635671635335736389e4, 3.848496228443793359990269e4,
2.637748787624195437963534e4, 7.225813979700288197698961e3
};
final double q1[] = {
6.748212550303777196073036e1, 1.113332393857199323513008e3,
7.738757056935398733233834e3, 2.763987074403340708898585e4,
5.499310206226157329794414e4, 6.161122180066002127833352e4,
3.635127591501940507276287e4, 8.785536302431013170870835e3
};
final double d2 = 4.227843350984671393993777e-1;
final double p2[] = {
4.974607845568932035012064e0, 5.424138599891070494101986e2,
1.550693864978364947665077e4, 1.847932904445632425417223e5,
1.088204769468828767498470e6, 3.338152967987029735917223e6,
5.106661678927352456275255e6, 3.074109054850539556250927e6
};
final double q2[] = {
1.830328399370592604055942e2, 7.765049321445005871323047e3,
1.331903827966074194402448e5, 1.136705821321969608938755e6,
5.267964117437946917577538e6, 1.346701454311101692290052e7,
1.782736530353274213975932e7, 9.533095591844353613395747e6
};
final double d4 = 1.791759469228055000094023e0;
final double p4[] = {
1.474502166059939948905062e4, 2.426813369486704502836312e6,
1.214755574045093227939592e8, 2.663432449630976949898078e9,
2.940378956634553899906876e10, 1.702665737765398868392998e11,
4.926125793377430887588120e11, 5.606251856223951465078242e11
};
final double q4[] = {
2.690530175870899333379843e3, 6.393885654300092398984238e5,
4.135599930241388052042842e7, 1.120872109616147941376570e9,
1.488613728678813811542398e10, 1.016803586272438228077304e11,
3.417476345507377132798597e11, 4.463158187419713286462081e11
};
final double c[] = {
-1.910444077728e-03, 8.4171387781295e-04,
-5.952379913043012e-04, 7.93650793500350248e-04,
-2.777777777777681622553e-03, 8.333333333333333331554247e-02,
5.7083835261e-03
};
final double a = 0.6796875;
if((x <= 0.5) || ((x > a) && (x <= 1.5))) {
if(x <= 0.5) {
result = -Math.log(x);
/* Test whether X < machine epsilon. */
if(x+1 == 1) {
return result;
}
}
else {
result = 0;
x = (x - 0.5) - 0.5;
}
xnum = 0;
xden = 1;
for(i=0;i<8;i++) {
xnum = xnum * x + p1[i];
xden = xden * x + q1[i];
}
result += x*(d1 + x*(xnum/xden));
}
else if((x <= a) || ((x > 1.5) && (x <= 4))) {
if(x <= a) {
result = -Math.log(x);
x = (x - 0.5) - 0.5;
}
else {
result = 0;
x -= 2;
}
xnum = 0;
xden = 1;
for(i=0;i<8;i++) {
xnum = xnum * x + p2[i];
xden = xden * x + q2[i];
}
result += x*(d2 + x*(xnum/xden));
}
else if(x <= 12) {
x -= 4;
xnum = 0;
xden = -1;
for(i=0;i<8;i++) {
xnum = xnum * x + p4[i];
xden = xden * x + q4[i];
}
result = d4 + x*(xnum/xden);
}
/* X > 12 */
else {
y = Math.log(x);
result = x*(y - 1) - y*0.5 + .9189385332046727417803297;
x = 1/x;
y = x*x;
xnum = c[6];
for(i=0;i<6;i++) {
xnum = xnum * y + c[i];
}
xnum *= x;
result += xnum;
}
return result;
}
// This is from "Numeric Recipes in C"
public static double oldLogGamma (double x) {
int j;
double y, tmp, ser;
double [] cof = {76.18009172947146, -86.50532032941677 ,
24.01409824083091, -1.231739572450155 ,
0.1208650973866179e-2, -0.5395239384953e-5};
y = x;
tmp = x + 5.5 - (x + 0.5) * Math.log (x + 5.5);
ser = 1.000000000190015;
for (j = 0; j <= 5; j++)
ser += cof[j] / ++y;
return Math.log (2.5066282746310005 * ser / x) - tmp;
}
public static double logBeta (double a, double b) {
return logGamma(a)+logGamma(b)-logGamma(a+b);
}
public static double beta (double a, double b) {
return Math.exp (logBeta(a,b));
}
public static double gamma (double x) {
return Math.exp (logGamma(x));
}
public static double factorial (int n) {
return Math.exp (logGamma(n+1));
}
public static double logFactorial (int n) {
return logGamma(n+1);
}
/**
* Computes p(x;n,p) where x~B(n,p)
*/
// Copied as the "classic" method from Catherine Loader.
// Fast and Accurate Computation of Binomial Probabilities.
// 2001. (This is not the fast and accurate version.)
public static double logBinom (int x, int n, double p)
{
return logFactorial (n) - logFactorial (x) - logFactorial (n - x)
+ (x*Math.log (p)) + ((n-x)*Math.log (1-p));
}
/**
* Vastly inefficient O(x) method to compute cdf of B(n,p)
*/
public static double pbinom (int x, int n, double p) {
double sum = Double.NEGATIVE_INFINITY;
for (int i = 0; i <= x; i++) {
sum = sumLogProb (sum, logBinom (i, n, p));
}
return Math.exp (sum);
}
public static double sigmod(double beta){
return (double)1.0/(1.0+Math.exp(-beta));
}
public static double sigmod_rev(double sig){
return (double)Math.log(sig/(1-sig));
}
public static double logit (double p)
{
return Math.log (p / (1 - p));
}
// Combination?
public static double numCombinations (int n, int r) {
return Math.exp (logFactorial(n)-logFactorial(r)-logFactorial(n-r));
}
// Permutation?
public static double numPermutations (int n, int r) {
return Math.exp (logFactorial(n)-logFactorial(r));
}
public static double cosh (double a)
{
if (a < 0)
return 0.5 * (Math.exp(-a) + Math.exp(a));
else
return 0.5 * (Math.exp(a) + Math.exp(-a));
}
public static double tanh (double a)
{
return (Math.exp(a) - Math.exp(-a)) / (Math.exp(a) + Math.exp(-a));
}
/**
* Numbers that are closer than this are considered equal
* by almostEquals.
*/
public static double EPSILON = 0.000001;
public static boolean almostEquals (double d1, double d2)
{
return almostEquals (d1, d2, EPSILON);
}
public static boolean almostEquals (double d1, double d2, double epsilon)
{
return Math.abs (d1 - d2) < epsilon;
}
public static boolean almostEquals (double[] d1, double[] d2, double eps)
{
for (int i = 0; i < d1.length; i++) {
double v1 = d1[i];
double v2 = d2[i];
if (!almostEquals (v1, v2, eps)) return false;
}
return true;
}
// gsc
/**
* Checks if min <= value <= max.
*/
public static boolean checkWithinRange(double value, double min, double max) {
return (value > min || almostEquals(value, min, EPSILON)) &&
(value < max || almostEquals(value, max, EPSILON));
}
public static final double log2 = Math.log(2);
// gsc
/**
* Returns the KL divergence, K(p1 || p2).
*
* The log is w.r.t. base 2.
*
* *Note*: If any value in p2 is 0.0 then the KL-divergence
* is infinite.
*
*/
public static double klDivergence(double[] p1, double[] p2) {
assert (p1.length == p2.length);
double klDiv = 0.0;
for (int i = 0; i < p1.length; ++i) {
if (p1[i] == 0) {
continue;
}
if (p2[i] == 0) {
return Double.POSITIVE_INFINITY;
}
klDiv += p1[i] * Math.log(p1[i] / p2[i]);
}
return klDiv / log2; // moved this division out of the loop -DM
}
// gsc
/**
* Returns the Jensen-Shannon divergence.
*/
public static double jensenShannonDivergence(double[] p1, double[] p2) {
assert(p1.length == p2.length);
double[] average = new double[p1.length];
for (int i = 0; i < p1.length; ++i) {
average[i] += (p1[i] + p2[i])/2;
}
return (klDivergence(p1, average) + klDivergence(p2, average))/2;
}
/**
* Returns the sum of two doubles expressed in log space,
* that is,
*
* sumLogProb = log (e^a + e^b)
* = log e^a(1 + e^(b-a))
* = a + log (1 + e^(b-a))
*
*
* By exponentiating b-a, we obtain better numerical precision than
* we would if we calculated e^a or e^b directly.
*
* Note: This function is just like
* {@link cc.mallet.fst.Transducer#sumNegLogProb sumNegLogProb}
* in Transducer,
* except that the logs aren't negated.
*/
public static double sumLogProb (double a, double b)
{
if (a == Double.NEGATIVE_INFINITY)
return b;
else if (b == Double.NEGATIVE_INFINITY)
return a;
else if (b < a)
return a + Math.log (1 + Math.exp(b-a));
else
return b + Math.log (1 + Math.exp(a-b));
}
// Below from Stanford NLP package, SloppyMath.java
private static final double LOGTOLERANCE = 30.0;
/**
* Sums an array of numbers log(x1)...log(xn). This saves some of
* the unnecessary calls to Math.log in the two-argument version.
*
* Note that this implementation IGNORES elements of the input
* array that are more than LOGTOLERANCE (currently 30.0) less
* than the maximum element.
*
* Cursory testing makes me wonder if this is actually much faster than
* repeated use of the 2-argument version, however -cas.
* @param vals An array log(x1), log(x2), ..., log(xn)
* @return log(x1+x2+...+xn)
*/
public static double sumLogProb (double[] vals)
{
double max = Double.NEGATIVE_INFINITY;
int len = vals.length;
int maxidx = 0;
for (int i = 0; i < len; i++) {
if (vals[i] > max) {
max = vals[i];
maxidx = i;
}
}
boolean anyAdded = false;
double intermediate = 0.0;
double cutoff = max - LOGTOLERANCE;
for (int i = 0; i < maxidx; i++) {
if (vals[i] >= cutoff) {
anyAdded = true;
intermediate += Math.exp(vals[i] - max);
}
}
for (int i = maxidx + 1; i < len; i++) {
if (vals[i] >= cutoff) {
anyAdded = true;
intermediate += Math.exp(vals[i] - max);
}
}
if (anyAdded) {
return max + Math.log(1.0 + intermediate);
} else {
return max;
}
}
/**
* Returns the difference of two doubles expressed in log space,
* that is,
*
* sumLogProb = log (e^a - e^b)
* = log e^a(1 - e^(b-a))
* = a + log (1 - e^(b-a))
*
*
* By exponentiating b-a, we obtain better numerical precision than
* we would if we calculated e^a or e^b directly.
*
* Returns NaN if b > a (so that log(e^a - e^b) is undefined).
*/
public static double subtractLogProb (double a, double b)
{
if (b == Double.NEGATIVE_INFINITY)
return a;
else
return a + Math.log (1 - Math.exp(b-a));
}
public static double getEntropy(double[] dist) {
double entropy = 0;
for (int i = 0; i < dist.length; i++) {
if (dist[i] != 0) {
entropy -= dist[i] * Math.log(dist[i]);
}
}
return entropy;
}
}