Opencl weibull_lpdf, neg_binomial_2_log_lpmf and neg_binomial_2_lpmf

stan/math/opencl/opencl.hpp

@@ -132,6 +132,8 @@
 #include <stan/math/opencl/prim/mdivide_left_tri_low.hpp>
 #include <stan/math/opencl/prim/mdivide_right_tri_low.hpp>
 #include <stan/math/opencl/prim/neg_binomial_2_log_glm_lpmf.hpp>
+#include <stan/math/opencl/prim/neg_binomial_2_log_lpmf.hpp>
+#include <stan/math/opencl/prim/neg_binomial_2_lpmf.hpp>
 #include <stan/math/opencl/prim/normal_id_glm_lpdf.hpp>
 #include <stan/math/opencl/prim/normal_lpdf.hpp>
 #include <stan/math/opencl/prim/ordered_logistic_glm_lpmf.hpp>
@@ -155,6 +157,7 @@
 #include <stan/math/opencl/prim/sum.hpp>
 #include <stan/math/opencl/prim/tcrossprod.hpp>
 #include <stan/math/opencl/prim/uniform_lpdf.hpp>
+#include <stan/math/opencl/prim/weibull_lpdf.hpp>
 
 #include <stan/math/opencl/err.hpp>
 

stan/math/opencl/prim/neg_binomial_2_log_lpmf.hpp

@@ -0,0 +1,122 @@
+#ifndef STAN_MATH_OPENCL_PRIM_NEG_BINOMIAL_2_LOG_LPMF_HPP
+#define STAN_MATH_OPENCL_PRIM_NEG_BINOMIAL_2_LOG_LPMF_HPP
+#ifdef STAN_OPENCL
+
+#include <stan/math/prim/meta.hpp>
+#include <stan/math/prim/err.hpp>
+#include <stan/math/prim/fun/constants.hpp>
+#include <stan/math/prim/fun/elt_divide.hpp>
+#include <stan/math/prim/fun/elt_multiply.hpp>
+#include <stan/math/prim/fun/exp.hpp>
+#include <stan/math/opencl/kernel_generator.hpp>
+#include <stan/math/prim/functor/operands_and_partials.hpp>
+
+namespace stan {
+namespace math {
+
+/** \ingroup opencl
+ * The log of the log transformed negative binomial density for the specified
+ * scalars given the specified mean(s) and deviation(s). n, eta, or phi can each
+ * be either a scalar or a vector matrix_cl. Any vector inputs must be the same
+ * length.
+ *
+ * <p>The result log probability is defined to be the sum of the
+ * log probabilities for each observation/mean/deviation triple.
+ *
+ * @tparam T_n_cl type of scalar
+ * @tparam T_log_location_cl type of location parameter
+ * @tparam T_precision_cl type of precision parameter
+ * @param n (Sequence of) scalar(s).
+ * @param eta (Sequence of) location parameter(s)
+ * @param phi (Sequence of) precision parameters
+ * @return The log of the product of the densities.
+ * @throw std::domain_error if the scale is not positive.
+ */
+template <bool propto, typename T_n_cl, typename T_log_location_cl,
+          typename T_precision_cl,
+          require_all_prim_or_rev_kernel_expression_t<
+              T_n_cl, T_log_location_cl, T_precision_cl>* = nullptr,
+          require_any_not_stan_scalar_t<T_n_cl, T_log_location_cl,
+                                        T_precision_cl>* = nullptr>
+inline return_type_t<T_n_cl, T_log_location_cl, T_precision_cl>
+neg_binomial_2_log_lpmf(const T_n_cl& n, const T_log_location_cl& eta,
+                        const T_precision_cl& phi) {
+  static const char* function = "neg_binomial_2_log_lpmf(OpenCL)";
+  using T_partials_return
+      = partials_return_t<T_n_cl, T_log_location_cl, T_precision_cl>;
+  using std::isfinite;
+  using std::isnan;
+
+  check_consistent_sizes(function, "Failures variable", n,
+                         "Log location parameter", eta, "Precision parameter",
+                         phi);
+  const size_t N = max_size(n, eta, phi);
+  if (N == 0) {
+    return 0.0;
+  }
+  if (!include_summand<propto, T_n_cl, T_log_location_cl,
+                       T_precision_cl>::value) {
+    return 0.0;
+  }
+
+  const auto& eta_val = value_of(eta);
+  const auto& phi_val = value_of(phi);
+
+  auto check_n_nonnegative
+      = check_cl(function, "Failures variable", n, "nonnegative");
+  auto n_nonnegative = n >= 0;
+  auto check_eta_finite
+      = check_cl(function, "Log location parameter", eta_val, "finite");
+  auto eta_finite = isfinite(eta_val);
+  auto check_phi_positive_finite
+      = check_cl(function, "Precision parameter", phi_val, "positive finite");
+  auto phi_positive_finite = 0 < phi_val && isfinite(phi_val);
+
+  auto log_phi = log(phi_val);
+  auto exp_eta = exp(eta_val);
+  auto exp_eta_over_exp_eta_phi
+      = elt_divide(1.0, elt_divide(phi_val, exp_eta) + 1.0);
+  auto log1p_exp_eta_m_logphi = log1p_exp(eta_val - log_phi);
+  auto n_plus_phi = n + phi_val;
+
+  auto logp1 = -elt_multiply(phi_val, log1p_exp_eta_m_logphi)
+               - elt_multiply(n, log_phi + log1p_exp_eta_m_logphi);
+  auto logp2 = static_select<include_summand<propto, T_precision_cl>::value>(
+      logp1 + binomial_coefficient_log(n_plus_phi - 1, n), logp1);
+  auto logp_expr = colwise_sum(
+      static_select<include_summand<propto, T_log_location_cl>::value>(
+          logp2 + elt_multiply(n, eta_val), logp2));
+
+  auto eta_deriv = n - elt_multiply(n_plus_phi, exp_eta_over_exp_eta_phi);
+  auto phi_deriv = exp_eta_over_exp_eta_phi - elt_divide(n, exp_eta + phi_val)
+                   - log1p_exp_eta_m_logphi - digamma(phi_val)
+                   + digamma(n_plus_phi);
+
+  matrix_cl<double> logp_cl;
+  matrix_cl<double> eta_deriv_cl;
+  matrix_cl<double> phi_deriv_cl;
+
+  results(check_n_nonnegative, check_eta_finite, check_phi_positive_finite,
+          logp_cl, eta_deriv_cl, phi_deriv_cl)
+      = expressions(n_nonnegative, eta_finite, phi_positive_finite, logp_expr,
+                    calc_if<!is_constant<T_log_location_cl>::value>(eta_deriv),
+                    calc_if<!is_constant<T_precision_cl>::value>(phi_deriv));
+
+  T_partials_return logp = sum(from_matrix_cl(logp_cl));
+
+  operands_and_partials<T_log_location_cl, T_precision_cl> ops_partials(eta,
+                                                                        phi);
+
+  if (!is_constant<T_log_location_cl>::value) {
+    ops_partials.edge1_.partials_ = std::move(eta_deriv_cl);
+  }
+  if (!is_constant<T_precision_cl>::value) {
+    ops_partials.edge2_.partials_ = std::move(phi_deriv_cl);
+  }
+  return ops_partials.build(logp);
+}
+
+}  // namespace math
+}  // namespace stan
+#endif
+#endif

stan/math/opencl/prim/neg_binomial_2_lpmf.hpp

@@ -0,0 +1,119 @@
+#ifndef STAN_MATH_OPENCL_PRIM_NEG_BINOMIAL_2_LPMF_HPP
+#define STAN_MATH_OPENCL_PRIM_NEG_BINOMIAL_2_LPMF_HPP
+#ifdef STAN_OPENCL
+
+#include <stan/math/prim/meta.hpp>
+#include <stan/math/prim/err.hpp>
+#include <stan/math/prim/fun/constants.hpp>
+#include <stan/math/prim/fun/elt_divide.hpp>
+#include <stan/math/prim/fun/elt_multiply.hpp>
+#include <stan/math/prim/fun/exp.hpp>
+#include <stan/math/opencl/kernel_generator.hpp>
+#include <stan/math/prim/functor/operands_and_partials.hpp>
+
+namespace stan {
+namespace math {
+
+/** \ingroup opencl
+ * The log of the negative binomial density for the specified scalars given
+ * the specified mean(s) and deviation(s). n, mu, or phi can
+ * each be either a scalar or a vector matrix_cl. Any vector inputs
+ * must be the same length.
+ *
+ * <p>The result log probability is defined to be the sum of the
+ * log probabilities for each observation/mean/deviation triple.
+ *
+ * @tparam T_n_cl type of scalar
+ * @tparam T_location_cl type of location parameter
+ * @tparam T_precision_cl type of precision parameter
+ * @param n (Sequence of) scalar(s).
+ * @param mu (Sequence of) location parameter(s)
+ * @param phi (Sequence of) precision parameters
+ * @return The log of the product of the densities.
+ * @throw std::domain_error if the scale is not positive.
+ */
+template <bool propto, typename T_n_cl, typename T_location_cl,
+          typename T_precision_cl,
+          require_all_prim_or_rev_kernel_expression_t<
+              T_n_cl, T_location_cl, T_precision_cl>* = nullptr,
+          require_any_not_stan_scalar_t<T_n_cl, T_location_cl,
+                                        T_precision_cl>* = nullptr>
+inline return_type_t<T_n_cl, T_location_cl, T_precision_cl> neg_binomial_2_lpmf(
+    const T_n_cl& n, const T_location_cl& mu, const T_precision_cl& phi) {
+  static const char* function = "neg_binomial_2_lpmf(OpenCL)";
+  using T_partials_return
+      = partials_return_t<T_n_cl, T_location_cl, T_precision_cl>;
+  using std::isfinite;
+  using std::isnan;
+
+  check_consistent_sizes(function, "Failures variable", n, "Location parameter",
+                         mu, "Precision parameter", phi);
+  const size_t N = max_size(n, mu, phi);
+  if (N == 0) {
+    return 0.0;
+  }
+  if (!include_summand<propto, T_n_cl, T_location_cl, T_precision_cl>::value) {
+    return 0.0;
+  }
+
+  const auto& mu_val = value_of(mu);
+  const auto& phi_val = value_of(phi);
+
+  auto check_n_nonnegative
+      = check_cl(function, "Failures variable", n, "nonnegative");
+  auto n_nonnegative = n >= 0;
+  auto check_mu_positive_finite
+      = check_cl(function, "Log location parameter", mu_val, "positive finite");
+  auto mu_positive_finite = 0 < mu_val && isfinite(mu_val);
+  auto check_phi_positive_finite
+      = check_cl(function, "Precision parameter", phi_val, "positive finite");
+  auto phi_positive_finite = 0 < phi_val && isfinite(phi_val);
+
+  auto log_phi = log(phi_val);
+  auto mu_plus_phi = mu_val + phi_val;
+  auto log_mu_plus_phi = log(mu_plus_phi);
+  auto n_plus_phi = n + phi_val;
+
+  auto logp1 = -elt_multiply(phi_val, log1p(elt_divide(mu_val, phi_val)))
+               - elt_multiply(n, log_mu_plus_phi);
+  auto logp2 = static_select<include_summand<propto, T_precision_cl>::value>(
+      logp1 + binomial_coefficient_log(n_plus_phi - 1, n), logp1);
+  auto logp_expr = colwise_sum(
+      static_select<include_summand<propto, T_location_cl>::value>(
+          logp2 + multiply_log(n, mu_val), logp2));
+
+  auto mu_deriv = elt_divide(n, mu_val) - elt_divide(n + phi_val, mu_plus_phi);
+  auto log_term
+      = select(mu_val < phi_val, log1p(-elt_divide(mu_val, mu_plus_phi)),
+               log_phi - log_mu_plus_phi);
+  auto phi_deriv = elt_divide(mu_val - n, mu_plus_phi) + log_term
+                   - digamma(phi_val) + digamma(n_plus_phi);
+
+  matrix_cl<double> logp_cl;
+  matrix_cl<double> mu_deriv_cl;
+  matrix_cl<double> phi_deriv_cl;
+
+  results(check_n_nonnegative, check_mu_positive_finite,
+          check_phi_positive_finite, logp_cl, mu_deriv_cl, phi_deriv_cl)
+      = expressions(n_nonnegative, mu_positive_finite, phi_positive_finite,
+                    logp_expr,
+                    calc_if<!is_constant<T_location_cl>::value>(mu_deriv),
+                    calc_if<!is_constant<T_precision_cl>::value>(phi_deriv));
+
+  T_partials_return logp = sum(from_matrix_cl(logp_cl));
+
+  operands_and_partials<T_location_cl, T_precision_cl> ops_partials(mu, phi);
+
+  if (!is_constant<T_location_cl>::value) {
+    ops_partials.edge1_.partials_ = std::move(mu_deriv_cl);
+  }
+  if (!is_constant<T_precision_cl>::value) {
+    ops_partials.edge2_.partials_ = std::move(phi_deriv_cl);
+  }
+  return ops_partials.build(logp);
+}
+
+}  // namespace math
+}  // namespace stan
+#endif
+#endif

stan/math/opencl/prim/weibull_lpdf.hpp

@@ -0,0 +1,129 @@
+#ifndef STAN_MATH_OPENCL_PRIM_WEIBULL_LPDF_HPP
+#define STAN_MATH_OPENCL_PRIM_WEIBULL_LPDF_HPP
+#ifdef STAN_OPENCL
+
+#include <stan/math/prim/meta.hpp>
+#include <stan/math/prim/err.hpp>
+#include <stan/math/prim/fun/constants.hpp>
+#include <stan/math/prim/fun/elt_divide.hpp>
+#include <stan/math/prim/fun/elt_multiply.hpp>
+#include <stan/math/opencl/kernel_generator.hpp>
+#include <stan/math/prim/functor/operands_and_partials.hpp>
+
+namespace stan {
+namespace math {
+
+/** \ingroup opencl
+ * Returns the Weibull log probability density for the given
+ * location and scale. Given containers of matching sizes, returns the
+ * log sum of probability densities.
+ *
+ * @tparam T_y_cl type of real parameter
+ * @tparam T_shape_cl type of shape parameter
+ * @tparam T_scale_cl type of scale parameter
+ * @param y real parameter
+ * @param alpha shape parameter
+ * @param sigma scale parameter
+ * @return log probability density or log sum of probability densities
+ * @throw std::domain_error if y is negative, alpha or sigma are nonpositive
+ */
+template <
+    bool propto, typename T_y_cl, typename T_shape_cl, typename T_scale_cl,
+    require_all_prim_or_rev_kernel_expression_t<T_y_cl, T_shape_cl,
+                                                T_scale_cl>* = nullptr,
+    require_any_not_stan_scalar_t<T_y_cl, T_shape_cl, T_scale_cl>* = nullptr>
+inline return_type_t<T_y_cl, T_shape_cl, T_scale_cl> weibull_lpdf(
+    const T_y_cl& y, const T_shape_cl& alpha, const T_scale_cl& sigma) {
+  static const char* function = "weibull_lpdf(OpenCL)";
+  using T_partials_return = partials_return_t<T_y_cl, T_shape_cl, T_scale_cl>;
+  using std::isfinite;
+  using std::isnan;
+
+  check_consistent_sizes(function, "Random variable", y, "Shape parameter",
+                         alpha, "Scale parameter", sigma);
+  const size_t N = max_size(y, alpha, sigma);
+  if (N == 0) {
+    return 0.0;
+  }
+  if (!include_summand<propto, T_y_cl, T_shape_cl, T_scale_cl>::value) {
+    return 0.0;
+  }
+
+  const auto& y_val = value_of(y);
+  const auto& alpha_val = value_of(alpha);
+  const auto& sigma_val = value_of(sigma);
+
+  auto check_y_finite = check_cl(function, "Random variable", y_val, "finite");
+  auto y_finite = isfinite(y_val);
+  auto check_alpha_positive_finite
+      = check_cl(function, "Shape parameter", alpha_val, "positive finite");
+  auto alpha_positive_finite = isfinite(alpha_val) && 0 < alpha_val;
+  auto check_sigma_positive_finite
+      = check_cl(function, "Scale parameter", sigma_val, "positive finite");
+  auto sigma_positive_finite = isfinite(sigma_val) && 0 < sigma_val;
+
+  auto any_y_negative = colwise_max(constant(0, N, 1) + (y_val < 0));
+  auto log_y = log(y_val);
+  auto log_sigma = log(sigma_val);
+  auto inv_sigma = elt_divide(1., sigma_val);
+  auto y_div_sigma_pow_alpha = pow(elt_multiply(y_val, inv_sigma), alpha_val);
+
+  auto logp1 = -y_div_sigma_pow_alpha;
+  auto logp2 = static_select<include_summand<propto, T_shape_cl>::value>(
+      logp1 + log(alpha_val), logp1);
+  auto logp3
+      = static_select<include_summand<propto, T_y_cl, T_shape_cl>::value>(
+          logp2 + elt_multiply(alpha_val - 1.0, log_y), logp2);
+  auto logp_expr = colwise_sum(
+      static_select<include_summand<propto, T_shape_cl, T_scale_cl>::value>(
+          logp3 - elt_multiply(alpha_val, log_sigma), logp3));
+
+  auto one_m_y_div_sigma_pow_alpha = 1.0 - y_div_sigma_pow_alpha;
+  auto y_deriv = elt_divide(
+      elt_multiply(alpha_val, one_m_y_div_sigma_pow_alpha) - 1.0, y_val);
+  auto alpha_deriv
+      = elt_divide(1.0, alpha_val)
+        + elt_multiply(one_m_y_div_sigma_pow_alpha, log_y - log_sigma);
+  auto sigma_deriv = elt_multiply(elt_multiply(alpha_val, inv_sigma),
+                                  -one_m_y_div_sigma_pow_alpha);
+
+  matrix_cl<int> any_y_negative_cl;
+  matrix_cl<double> logp_cl;
+  matrix_cl<double> alpha_deriv_cl;
+  matrix_cl<double> y_deriv_cl;
+  matrix_cl<double> sigma_deriv_cl;
+
+  results(check_y_finite, check_alpha_positive_finite,
+          check_sigma_positive_finite, any_y_negative_cl, logp_cl, y_deriv_cl,
+          alpha_deriv_cl, sigma_deriv_cl)
+      = expressions(y_finite, alpha_positive_finite, sigma_positive_finite,
+                    any_y_negative, logp_expr,
+                    calc_if<!is_constant<T_y_cl>::value>(y_deriv),
+                    calc_if<!is_constant<T_shape_cl>::value>(alpha_deriv),
+                    calc_if<!is_constant<T_scale_cl>::value>(sigma_deriv));
+
+  if (from_matrix_cl(any_y_negative_cl).any()) {
+    return LOG_ZERO;
+  }
+
+  T_partials_return logp = sum(from_matrix_cl(logp_cl));
+
+  operands_and_partials<T_y_cl, T_shape_cl, T_scale_cl> ops_partials(y, alpha,
+                                                                     sigma);
+
+  if (!is_constant<T_y_cl>::value) {
+    ops_partials.edge1_.partials_ = std::move(y_deriv_cl);
+  }
+  if (!is_constant<T_shape_cl>::value) {
+    ops_partials.edge2_.partials_ = std::move(alpha_deriv_cl);
+  }
+  if (!is_constant<T_scale_cl>::value) {
+    ops_partials.edge3_.partials_ = std::move(sigma_deriv_cl);
+  }
+  return ops_partials.build(logp);
+}
+
+}  // namespace math
+}  // namespace stan
+#endif
+#endif