Merge pull request #1433 from stan-dev/feature/issue-1426-ibeta_deriv…

…s_large_args

fixes #1426. Feature/issue 1426 ibeta derivs large args

src/stan/math/prim/scal/fun.hpp

@@ -28,6 +28,9 @@
 #include <stan/math/prim/scal/fun/identity_free.hpp>
 #include <stan/math/prim/scal/fun/if_else.hpp>
 #include <stan/math/prim/scal/fun/inc_beta.hpp>
+#include <stan/math/prim/scal/fun/inc_beta_dda.hpp>
+#include <stan/math/prim/scal/fun/inc_beta_ddb.hpp>
+#include <stan/math/prim/scal/fun/inc_beta_ddz.hpp>
 #include <stan/math/prim/scal/fun/int_step.hpp>
 #include <stan/math/prim/scal/fun/inv.hpp>
 #include <stan/math/prim/scal/fun/inv_cloglog.hpp>

src/stan/math/prim/scal/fun/inc_beta_dda.hpp

@@ -0,0 +1,94 @@
+#ifndef STAN_MATH_PRIM_SCAL_FUN_INC_BETA_DDA_HPP
+#define STAN_MATH_PRIM_SCAL_FUN_INC_BETA_DDA_HPP
+
+#include <stan/math/prim/scal/fun/inc_beta.hpp>
+#include <stan/math/prim/scal/fun/inc_beta_ddb.hpp>
+#include <cmath>
+
+namespace stan {
+  namespace math {
+
+    template <typename T>
+    T inc_beta_ddb(T a, T b, T z,
+                   T digamma_b, T digamma_ab);
+
+    /**
+     * Returns the partial derivative of the regularized
+     * incomplete beta function, I_{z}(a, b) with respect to a.
+     * The power series used to compute the deriative tends to
+     * converge slowly when a and b are large, especially if z
+     * approaches 1.  The implementation will throw an exception
+     * if the series have not converged within 100,000 iterations.
+     * The current implementation has been tested for values
+     * of a and b up to 12500 and z = 0.999.
+     *
+     * @tparam T scalar types of arguments
+     * @param a a
+     * @param b b
+     * @param z upper bound of the integral
+     * @param digamma_a value of digamma(a)
+     * @param digamma_ab value of digamma(b)
+     * @return partial derivative of the incomplete beta with respect to a
+     *
+     * @pre a >= 0
+     * @pre b >= 0
+     * @pre 0 <= z <= 1
+     */
+    template <typename T>
+    T inc_beta_dda(T a, T b, T z,
+                   T digamma_a, T digamma_ab) {
+      using std::log;
+
+      if (b > a)
+        if ((0.1 < z && z <= 0.75 && b > 500)
+            || (0.01 < z && z <= 0.1 && b > 2500)
+            || (0.001 < z && z <= 0.01 && b > 1e5))
+          return -inc_beta_ddb(b, a, 1 - z, digamma_a, digamma_ab);
+
+      if (z > 0.75 && a < 500)
+        return -inc_beta_ddb(b, a, 1 - z, digamma_a, digamma_ab);
+      if (z > 0.9 && a < 2500)
+        return -inc_beta_ddb(b, a, 1 - z, digamma_a, digamma_ab);
+      if (z > 0.99 && a < 1e5)
+        return -inc_beta_ddb(b, a, 1 - z, digamma_a, digamma_ab);
+      if (z > 0.999)
+        return -inc_beta_ddb(b, a, 1 - z, digamma_a, digamma_ab);
+
+      double threshold = 1e-10;
+
+      digamma_a += 1.0 / a;  // Need digamma(a + 1), not digamma(a);
+
+      // Common prefactor to regularize numerator and denomentator
+      T prefactor = (a + 1) / (a + b);
+      prefactor = prefactor * prefactor * prefactor;
+
+      T sum_numer = (digamma_ab - digamma_a) * prefactor;
+      T sum_denom = prefactor;
+
+      T summand = prefactor * z * (a + b) / (a + 1);
+
+      T k = 1;
+      digamma_ab += 1.0 / (a + b);
+      digamma_a += 1.0 / (a + 1);
+
+      while (fabs(summand) > threshold) {
+        sum_numer += (digamma_ab - digamma_a) * summand;
+        sum_denom += summand;
+
+        summand *= (1 + (a + b) / k) * (1 + k) / (1 + (a + 1) / k);
+        digamma_ab += 1.0 / (a + b + k);
+        digamma_a += 1.0 / (a + 1 + k);
+        ++k;
+        summand *= z / k;
+
+        if (k > 1e5)
+          throw std::domain_error("stan::math::inc_beta_dda did "
+                                  "not converge within 100000 iterations");
+      }
+      return inc_beta(a, b, z) * (log(z) + sum_numer / sum_denom);
+    }
+
+  }  // math
+}   // stan
+
+#endif

src/stan/math/prim/scal/fun/inc_beta_ddb.hpp

@@ -0,0 +1,89 @@
+#ifndef STAN_MATH_PRIM_SCAL_FUN_INC_BETA_DDB_HPP
+#define STAN_MATH_PRIM_SCAL_FUN_INC_BETA_DDB_HPP
+
+#include <stan/math/prim/scal/fun/inc_beta.hpp>
+#include <stan/math/prim/scal/fun/inc_beta_dda.hpp>
+#include <cmath>
+
+namespace stan {
+  namespace math {
+
+    template <typename T>
+    T inc_beta_dda(T a, T b, T z,
+                   T digamma_a, T digamma_ab);
+
+    /**
+     * Returns the partial derivative of the regularized
+     * incomplete beta function, I_{z}(a, b) with respect to b.
+     * The power series used to compute the deriative tends to
+     * converge slowly when a and b are large, especailly if z
+     * approaches 1.  The implementation will throw an exception
+     * if the series have not converged within 100,000 iterations.
+     * The current implementation has been tested for values
+     * of a and b up to 12500 and z = 0.999.
+     *
+     * @tparam T scalar types of arguments
+     * @param a a
+     * @param b b
+     * @param z upper bound of the integral
+     * @param digamma_a value of digamma(a)
+     * @param digamma_ab value of digamma(b)
+     * @return partial derivative of the incomplete beta with respect to b
+     *
+     * @pre a >= 0
+     * @pre b >= 0
+     * @pre 0 <= z <= 1
+     */
+    template <typename T>
+    T inc_beta_ddb(T a, T b, T z,
+                   T digamma_b, T digamma_ab) {
+      using std::log;
+
+      if (b > a)
+        if ((0.1 < z && z <= 0.75 && b > 500)
+            || (0.01 < z && z <= 0.1 && b > 2500)
+            || (0.001 < z && z <= 0.01 && b > 1e5))
+          return -inc_beta_dda(b, a, 1 - z, digamma_b, digamma_ab);
+
+      if ((z > 0.75 && a < 500)
+          || (z > 0.9 && a < 2500)
+          || (z > 0.99 && a < 1e5)
+          || (z > 0.999))
+        return -inc_beta_dda(b, a, 1 - z, digamma_b, digamma_ab);
+
+      double threshold = 1e-10;
+
+      // Common prefactor to regularize numerator and denomentator
+      T prefactor = (a + 1) / (a + b);
+      prefactor = prefactor * prefactor * prefactor;
+
+      T sum_numer = digamma_ab * prefactor;
+      T sum_denom = prefactor;
+
+      T summand = prefactor * z * (a + b) / (a + 1);
+
+      T k = 1;
+      digamma_ab += 1.0 / (a + b);
+
+      while (fabs(summand) > threshold) {
+        sum_numer += digamma_ab * summand;
+        sum_denom += summand;
+
+        summand *= (1 + (a + b) / k) * (1 + k) / (1 + (a + 1) / k);
+        digamma_ab += 1.0 / (a + b + k);
+        ++k;
+        summand *= z / k;
+
+        if (k > 1e5)
+          throw std::domain_error("stan::math::inc_beta_ddb did "
+                                  "not converge within 100000 iterations");
+      }
+
+      return inc_beta(a, b, z)
+        * (log(1 - z) - digamma_b + sum_numer / sum_denom);
+    }
+
+  }  // math
+}   // stan
+
+#endif

src/stan/math/prim/scal/fun/inc_beta_ddz.hpp

@@ -0,0 +1,44 @@
+#ifndef STAN_MATH_PRIM_SCAL_FUN_INC_BETA_DERIVATIVES_HPP
+#define STAN_MATH_PRIM_SCAL_FUN_INC_BETA_DERIVATIVES_HPP
+
+#include <stan/math/prim/scal/fun/lgamma.hpp>
+#include <stan/math/prim/scal/fun/inc_beta.hpp>
+#include <boost/math/special_functions/beta.hpp>
+#include <cmath>
+
+namespace stan {
+  namespace math {
+
+
+    /**
+     * Returns the partial derivative of the regularized
+     * incomplete beta function, I_{z}(a, b) with respect to z.
+     *
+     * @tparam T scalar types of arguments
+     * @param a a
+     * @param b b
+     * @param z upper bound of the integral
+     * @return partial derivative of the incomplete beta with respect to z
+     *
+     * @pre a > 0
+     * @pre b > 0
+     * @pre 0 < z <= 1
+     */
+    template <typename T>
+    T inc_beta_ddz(T a, T b, T z) {
+      using std::exp;
+      using std::log;
+      return exp((b - 1) * log(1 - z) + (a - 1) * log(z)
+                 + lgamma(a + b) - lgamma(a) - lgamma(b));
+    }
+
+    template <>
+    double inc_beta_ddz(double a, double b, double z) {
+      using boost::math::ibeta_derivative;
+      return ibeta_derivative(a, b, z);
+    }
+
+  }  // math
+}   // stan
+
+#endif