Automatic Differentiation
 
Loading...
Searching...
No Matches
skew_double_exponential_lcdf.hpp
Go to the documentation of this file.
1#ifndef STAN_MATH_OPENCL_PRIM_DOUBLE_SKEW_DOUBLE_EXPONENTIAL_LCDF_HPP
2#define STAN_MATH_OPENCL_PRIM_DOUBLE_SKEW_DOUBLE_EXPONENTIAL_LCDF_HPP
3#ifdef STAN_OPENCL
4
12
13namespace stan {
14namespace math {
15
30template <typename T_y_cl, typename T_loc_cl, typename T_scale_cl,
31 typename T_skewness_cl,
33 T_y_cl, T_loc_cl, T_scale_cl, T_skewness_cl>* = nullptr,
34 require_any_not_stan_scalar_t<T_y_cl, T_loc_cl, T_scale_cl,
35 T_skewness_cl>* = nullptr>
36return_type_t<T_y_cl, T_loc_cl, T_scale_cl, T_skewness_cl>
37skew_double_exponential_lcdf(const T_y_cl& y, const T_loc_cl& mu,
38 const T_scale_cl& sigma,
39 const T_skewness_cl& tau) {
40 static constexpr const char* function
41 = "skew_double_exponential_lcdf(OpenCL)";
42 using T_partials_return
44 using std::isfinite;
45 using std::isnan;
46
47 check_consistent_sizes(function, "Random variable", y, "Location parameter",
48 mu, "Shape parameter", sigma, "Skewness parameter",
49 tau);
50 const size_t N = max_size(y, mu, sigma, tau);
51 if (N == 0) {
52 return 1.0;
53 }
54
55 const auto& y_col = as_column_vector_or_scalar(y);
56 const auto& mu_col = as_column_vector_or_scalar(mu);
57 const auto& sigma_col = as_column_vector_or_scalar(sigma);
58 const auto& tau_col = as_column_vector_or_scalar(tau);
59
60 const auto& y_val = value_of(y_col);
61 const auto& mu_val = value_of(mu_col);
62 const auto& sigma_val = value_of(sigma_col);
63 const auto& tau_val = value_of(tau_col);
64
65 auto check_y_not_nan
66 = check_cl(function, "Random variable", y_val, "not NaN");
67 auto y_not_nan_expr = !isnan(y_val);
68 auto check_mu_finite
69 = check_cl(function, "Location parameter", mu_val, "finite");
70 auto mu_finite_expr = isfinite(mu_val);
71 auto check_sigma_positive_finite
72 = check_cl(function, "Scale parameter", sigma_val, "positive finite");
73 auto sigma_positive_finite_expr = 0.0 < sigma_val && isfinite(sigma_val);
74 auto check_tau_bounded = check_cl(function, "Skewness parameter", tau_val,
75 "in the interval [0, 1]");
76 auto tau_bounded_expr = 0.0 < tau_val && tau_val <= 1.0;
77
78 auto inv_sigma = elt_divide(1.0, sigma_val);
79 auto y_m_mu = y_val - mu_val;
80 auto diff_sign = sign(y_m_mu);
81 auto diff_sign_smaller_0 = diff_sign < 0;
82 auto abs_diff_y_mu = fabs(y_m_mu);
83 auto abs_diff_y_mu_over_sigma = elt_multiply(abs_diff_y_mu, inv_sigma);
84 auto expo
85 = elt_multiply(diff_sign_smaller_0 + elt_multiply(diff_sign, tau_val),
86 abs_diff_y_mu_over_sigma);
87 auto tau_minus_1 = tau_val - 1.0;
88 auto inv_exp_2_expo_tau = elt_divide(1.0, exp(2.0 * expo) + tau_minus_1);
89
90 auto lcdf_expr
91 = colwise_sum(select(y_val <= mu_val, log(tau_val) - 2.0 * expo,
92 log1m_exp(log1m(tau_val) - 2.0 * expo)));
93
94 auto cond = y_val < mu_val;
95 auto y_deriv
96 = select(cond, -2.0 * elt_multiply(inv_sigma, tau_minus_1),
97 -2.0
98 * elt_multiply(elt_multiply(tau_minus_1, tau_val),
99 elt_multiply(inv_sigma, inv_exp_2_expo_tau)));
100 auto mu_deriv = -y_deriv;
101 auto sigma_deriv = select(cond, 2.0 * elt_multiply(inv_sigma, expo),
102 -elt_divide(elt_multiply(y_deriv, expo), tau_val));
103 auto tau_deriv = select(
104 cond,
105 elt_divide(1.0, tau_val)
106 + 2.0 * elt_multiply(elt_multiply(inv_sigma, y_m_mu), diff_sign),
107 elt_multiply(sigma_val - 2.0 * elt_multiply(tau_minus_1, y_m_mu),
108 elt_multiply(inv_sigma, inv_exp_2_expo_tau)));
109
110 matrix_cl<double> lcdf_cl;
111 matrix_cl<double> y_deriv_cl;
112 matrix_cl<double> mu_deriv_cl;
113 matrix_cl<double> sigma_deriv_cl;
114 matrix_cl<double> tau_deriv_cl;
115
116 results(check_y_not_nan, check_mu_finite, check_sigma_positive_finite,
117 check_tau_bounded, lcdf_cl, y_deriv_cl, mu_deriv_cl, sigma_deriv_cl,
118 tau_deriv_cl)
119 = expressions(y_not_nan_expr, mu_finite_expr, sigma_positive_finite_expr,
120 tau_bounded_expr, lcdf_expr,
125
126 T_partials_return lcdf = from_matrix_cl(lcdf_cl).sum();
127
128 auto ops_partials
129 = make_partials_propagator(y_col, mu_col, sigma_col, tau_col);
130
132 partials<0>(ops_partials) = std::move(y_deriv_cl);
133 }
135 partials<1>(ops_partials) = std::move(mu_deriv_cl);
136 }
138 partials<2>(ops_partials) = std::move(sigma_deriv_cl);
139 }
141 partials<3>(ops_partials) = std::move(tau_deriv_cl);
142 }
143
144 return ops_partials.build(lcdf);
145}
146
147} // namespace math
148} // namespace stan
149#endif
150#endif
Represents an arithmetic matrix on the OpenCL device.
Definition matrix_cl.hpp:47
elt_multiply_< as_operation_cl_t< T_a >, as_operation_cl_t< T_b > > elt_multiply(T_a &&a, T_b &&b)
isfinite_< as_operation_cl_t< T > > isfinite(T &&a)
select_< as_operation_cl_t< T_condition >, as_operation_cl_t< T_then >, as_operation_cl_t< T_else > > select(T_condition &&condition, T_then &&then, T_else &&els)
Selection operation on kernel generator expressions.
Definition select.hpp:148
auto check_cl(const char *function, const char *var_name, T &&y, const char *must_be)
Constructs a check on opencl matrix or expression.
Definition check_cl.hpp:219
results_cl< T_results... > results(T_results &&... results)
Deduces types for constructing results_cl object.
auto as_column_vector_or_scalar(T &&a)
as_column_vector_or_scalar of a kernel generator expression.
elt_divide_< as_operation_cl_t< T_a >, as_operation_cl_t< T_b > > elt_divide(T_a &&a, T_b &&b)
calc_if_< true, as_operation_cl_t< T > > calc_if(T &&a)
Definition calc_if.hpp:121
auto colwise_sum(T &&a)
Column wise sum - reduction of a kernel generator expression.
expressions_cl< T_expressions... > expressions(T_expressions &&... expressions)
Deduces types for constructing expressions_cl object.
return_type_t< T_y_cl, T_loc_cl, T_scale_cl, T_skewness_cl > skew_double_exponential_lcdf(const T_y_cl &y, const T_loc_cl &mu, const T_scale_cl &sigma, const T_skewness_cl &tau)
Returns the skew double exponential cumulative density function.
auto from_matrix_cl(const T &src)
Copies the source matrix that is stored on the OpenCL device to the destination Eigen matrix.
Definition copy.hpp:61
require_all_t< is_prim_or_rev_kernel_expression< std::decay_t< Types > >... > require_all_prim_or_rev_kernel_expression_t
Require type satisfies is_prim_or_rev_kernel_expression.
require_any_not_t< is_stan_scalar< std::decay_t< Types > >... > require_any_not_stan_scalar_t
Require at least one of the types do not satisfy is_stan_scalar.
fvar< T > log1m_exp(const fvar< T > &x)
Return the natural logarithm of one minus the exponentiation of the specified argument.
Definition log1m_exp.hpp:23
auto sign(const T &x)
Returns signs of the arguments.
Definition sign.hpp:18
T value_of(const fvar< T > &v)
Return the value of the specified variable.
Definition value_of.hpp:18
fvar< T > log(const fvar< T > &x)
Definition log.hpp:15
void check_consistent_sizes(const char *)
Trivial no input case, this function is a no-op.
int64_t max_size(const T1 &x1, const Ts &... xs)
Calculate the size of the largest input.
Definition max_size.hpp:20
fvar< T > log1m(const fvar< T > &x)
Definition log1m.hpp:12
auto make_partials_propagator(Ops &&... ops)
Construct an partials_propagator.
fvar< T > fabs(const fvar< T > &x)
Definition fabs.hpp:15
fvar< T > exp(const fvar< T > &x)
Definition exp.hpp:13
typename partials_return_type< Args... >::type partials_return_t
The lgamma implementation in stan-math is based on either the reentrant safe lgamma_r implementation ...
bool isnan(const stan::math::var &a)
Checks if the given number is NaN.
Definition std_isnan.hpp:18
Metaprogramming struct to detect whether a given type is constant in the mathematical sense (not the ...