You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@commons.apache.org by ce...@apache.org on 2012/05/15 08:21:53 UTC
svn commit: r1338548 -
/commons/proper/math/trunk/src/main/java/org/apache/commons/math3/distribution/GammaDistribution.java
Author: celestin
Date: Tue May 15 06:21:53 2012
New Revision: 1338548
URL: http://svn.apache.org/viewvc?rev=1338548&view=rev
Log:
New implementation of the pdf of Gamma distributions. Solves MATH-753.
Additional unit tests to come.
Modified:
commons/proper/math/trunk/src/main/java/org/apache/commons/math3/distribution/GammaDistribution.java
Modified: commons/proper/math/trunk/src/main/java/org/apache/commons/math3/distribution/GammaDistribution.java
URL: http://svn.apache.org/viewvc/commons/proper/math/trunk/src/main/java/org/apache/commons/math3/distribution/GammaDistribution.java?rev=1338548&r1=1338547&r2=1338548&view=diff
==============================================================================
--- commons/proper/math/trunk/src/main/java/org/apache/commons/math3/distribution/GammaDistribution.java (original)
+++ commons/proper/math/trunk/src/main/java/org/apache/commons/math3/distribution/GammaDistribution.java Tue May 15 06:21:53 2012
@@ -34,12 +34,59 @@ public class GammaDistribution extends A
* @since 2.1
*/
public static final double DEFAULT_INVERSE_ABSOLUTE_ACCURACY = 1e-9;
+
/** Serializable version identifier. */
private static final long serialVersionUID = -3239549463135430361L;
+
/** The shape parameter. */
private final double alpha;
+
/** The scale parameter. */
private final double beta;
+
+ /**
+ * The constant value of {@code alpha + g + 0.5}, where {@code alpha} is
+ * the shape parameter, and {@code g} is the Lanczos constant
+ * {@link Gamma#LANCZOS_G}.
+ */
+ private final double shiftedShape;
+
+ /**
+ * The constant value of
+ * {@code alpha / beta * sqrt(e / (2 * pi * (alpha + g + 0.5))) / L(alpha)},
+ * where {@code alpha} is the shape parameter, {@code beta} is the scale
+ * parameter, and {@code L(alpha)} is the Lanczos approximation returned by
+ * {@link Gamma#lanczos(double)}. This prefactor is used in
+ * {@link #density(double)}, when no overflow occurs with the natural
+ * calculation.
+ */
+ private final double densityPrefactor1;
+
+ /**
+ * The constant value of
+ * {@code alpha * sqrt(e / (2 * pi * (alpha + g + 0.5))) / L(alpha)},
+ * where {@code alpha} is the shape parameter, and {@code L(alpha)} is the
+ * Lanczos approximation returned by {@link Gamma#lanczos(double)}. This
+ * prefactor is used in {@link #density(double)}, when overflow occurs with
+ * the natural calculation.
+ */
+ private final double densityPrefactor2;
+
+ /**
+ * Lower bound on {@code y = x / beta} for the selection of the computation
+ * method in {@link #density(double)}. For {@code y <= minY}, the natural
+ * calculation overflows. {@code beta} is the shape parameter.
+ */
+ private final double minY;
+
+ /**
+ * Upper bound on {@code log(y)} ({@code y = x / beta}) for the selection of
+ * the computation method in {@link #density(double)}. For
+ * {@code log(y) >= maxLogY}, the natural calculation overflows.
+ * {@code beta} is the shape parameter.
+ */
+ private final double maxLogY;
+
/** Inverse cumulative probability accuracy. */
private final double solverAbsoluteAccuracy;
@@ -77,7 +124,15 @@ public class GammaDistribution extends A
this.alpha = alpha;
this.beta = beta;
- solverAbsoluteAccuracy = inverseCumAccuracy;
+ this.solverAbsoluteAccuracy = inverseCumAccuracy;
+ this.shiftedShape = alpha + Gamma.LANCZOS_G + 0.5;
+ final double aux = FastMath.E / (2.0 * FastMath.PI * shiftedShape);
+ this.densityPrefactor2 = alpha * FastMath.sqrt(aux) / Gamma.lanczos(alpha);
+ this.densityPrefactor1 = this.densityPrefactor2 / beta *
+ FastMath.pow(shiftedShape, -alpha) *
+ FastMath.exp(alpha + Gamma.LANCZOS_G);
+ this.minY = alpha + Gamma.LANCZOS_G - FastMath.log(Double.MAX_VALUE);
+ this.maxLogY = FastMath.log(Double.MAX_VALUE) / (alpha - 1.0);
}
/**
@@ -111,11 +166,63 @@ public class GammaDistribution extends A
/** {@inheritDoc} */
public double density(double x) {
+ /* The present method must return the value of
+ *
+ * 1 x a - x
+ * ---------- (-) exp(---)
+ * x Gamma(a) b b
+ *
+ * where a is the shape parameter, and b the scale parameter.
+ * Substituting the Lanczos approximation of Gamma(a) leads to the
+ * following expression of the density
+ *
+ * a e 1 y a
+ * - sqrt(------------------) ---- (-----------) exp(a - y + g),
+ * x 2 pi (a + g + 0.5) L(a) a + g + 0.5
+ *
+ * where y = x / b. The above formula is the "natural" computation, which
+ * is implemented when no overflow is likely to occur. If overflow occurs
+ * with the natural computation, the following identity is used. It is
+ * based on the BOOST library
+ * http://www.boost.org/doc/libs/1_35_0/libs/math/doc/sf_and_dist/html/math_toolkit/special/sf_gamma/igamma.html
+ * Formula (15) needs adaptations, which are detailed below.
+ *
+ * y a
+ * (-----------) exp(a - y + g)
+ * a + g + 0.5
+ * y - a - g - 0.5 y (g + 0.5)
+ * = exp(a log1pm(---------------) - ----------- + g),
+ * a + g + 0.5 a + g + 0.5
+ *
+ * where log1pm(z) = log(1 + z) - z. Therefore, the value to be
+ * returned is
+ *
+ * a e 1
+ * - sqrt(------------------) ----
+ * x 2 pi (a + g + 0.5) L(a)
+ * y - a - g - 0.5 y (g + 0.5)
+ * * exp(a log1pm(---------------) - ----------- + g).
+ * a + g + 0.5 a + g + 0.5
+ */
if (x < 0) {
return 0;
}
- return FastMath.pow(x / beta, alpha - 1) / beta *
- FastMath.exp(-x / beta) / FastMath.exp(Gamma.logGamma(alpha));
+ final double y = x / beta;
+ if ((y <= minY) || (FastMath.log(y) >= maxLogY)) {
+ /*
+ * Overflow.
+ */
+ final double aux1 = (y - shiftedShape) / shiftedShape;
+ final double aux2 = alpha * (FastMath.log1p(aux1) - aux1);
+ final double aux3 = -y * (Gamma.LANCZOS_G + 0.5) / shiftedShape +
+ Gamma.LANCZOS_G + aux2;
+ return densityPrefactor2 / x * FastMath.exp(aux3);
+ }
+ /*
+ * Natural calculation.
+ */
+ return densityPrefactor1 * FastMath.exp(-y) *
+ FastMath.pow(y, alpha - 1);
}
/**