Automatic Differentiation
 
Loading...
Searching...
No Matches
gp_matern32_cov.hpp
Go to the documentation of this file.
1#ifndef STAN_MATH_PRIM_FUN_GP_MATERN32_COV_HPP
2#define STAN_MATH_PRIM_FUN_GP_MATERN32_COV_HPP
3
12#include <cmath>
13#include <vector>
14
15namespace stan {
16namespace math {
17
36template <typename T_x, typename T_s, typename T_l>
37inline typename Eigen::Matrix<return_type_t<T_x, T_s, T_l>, Eigen::Dynamic,
38 Eigen::Dynamic>
39gp_matern32_cov(const std::vector<T_x> &x, const T_s &sigma,
40 const T_l &length_scale) {
41 using std::exp;
42 using std::pow;
43
44 size_t x_size = stan::math::size(x);
45 Eigen::Matrix<return_type_t<T_x, T_s, T_l>, Eigen::Dynamic, Eigen::Dynamic>
46 cov(x_size, x_size);
47
48 if (x_size == 0) {
49 return cov;
50 }
51
52 const char *function = "gp_matern32_cov";
53 size_t x_obs_size = stan::math::size(x[0]);
54 for (size_t i = 0; i < x_size; ++i) {
55 check_size_match(function, "x row", x_obs_size, "x's other row",
56 stan::math::size(x[i]));
57 }
58
59 for (size_t n = 0; n < x_size; ++n) {
60 check_not_nan(function, "x", x[n]);
61 }
62
63 check_positive_finite(function, "magnitude", sigma);
64 check_positive_finite(function, "length scale", length_scale);
65
66 T_s sigma_sq = square(sigma);
67 T_l root_3_inv_l = std::sqrt(3.0) / length_scale;
68
69 size_t block_size = 10;
70 for (size_t jb = 0; jb < x_size; jb += block_size) {
71 for (size_t ib = jb; ib < x_size; ib += block_size) {
72 size_t j_end = std::min(x_size, jb + block_size);
73 for (size_t j = jb; j < j_end; ++j) {
74 cov.coeffRef(j, j) = sigma_sq;
75 size_t i_end = std::min(x_size, ib + block_size);
76 for (size_t i = std::max(ib, j + 1); i < i_end; ++i) {
77 return_type_t<T_x> dist = distance(x[i], x[j]);
78 cov.coeffRef(j, i) = cov.coeffRef(i, j)
79 = sigma_sq * (1.0 + root_3_inv_l * dist)
80 * exp(-root_3_inv_l * dist);
81 }
82 }
83 }
84 }
85 return cov;
86}
87
106template <typename T_x, typename T_s, typename T_l>
107inline typename Eigen::Matrix<return_type_t<T_x, T_s, T_l>, Eigen::Dynamic,
108 Eigen::Dynamic>
109gp_matern32_cov(const std::vector<Eigen::Matrix<T_x, -1, 1>> &x,
110 const T_s &sigma, const std::vector<T_l> &length_scale) {
111 using std::exp;
112
113 size_t x_size = stan::math::size(x);
114 Eigen::Matrix<return_type_t<T_x, T_s, T_l>, Eigen::Dynamic, Eigen::Dynamic>
115 cov(x_size, x_size);
116
117 if (x_size == 0) {
118 return cov;
119 }
120 const char *function = "gp_matern32_cov";
121 for (size_t n = 0; n < x_size; ++n) {
122 check_not_nan(function, "x", x[n]);
123 }
124
125 check_positive_finite(function, "magnitude", sigma);
126 check_positive_finite(function, "length scale", length_scale);
127
128 size_t l_size = length_scale.size();
129 for (size_t n = 0; n < l_size; ++n) {
130 check_not_nan(function, "length scale", length_scale[n]);
131 }
132
133 check_size_match(function, "x dimension", stan::math::size(x[0]),
134 "number of length scales", l_size);
135
136 T_s sigma_sq = square(sigma);
137 double root_3 = std::sqrt(3.0);
138
139 std::vector<Eigen::Matrix<return_type_t<T_x, T_l>, -1, 1>> x_new
140 = divide_columns(x, length_scale);
141
142 size_t block_size = 10;
143 for (size_t jb = 0; jb < x_size; jb += block_size) {
144 for (size_t ib = jb; ib < x_size; ib += block_size) {
145 size_t j_end = std::min(x_size, jb + block_size);
146 for (size_t j = jb; j < j_end; ++j) {
147 size_t i_end = std::min(x_size, ib + block_size);
148 for (size_t i = std::max(ib, j); i < i_end; ++i) {
149 return_type_t<T_x, T_l> dist = distance(x_new[i], x_new[j]);
150 cov.coeffRef(j, i) = cov.coeffRef(i, j)
151 = sigma_sq * (1.0 + root_3 * dist) * exp(-root_3 * dist);
152 }
153 }
154 }
155 }
156 return cov;
157}
158
183template <typename T_x1, typename T_x2, typename T_s, typename T_l>
184inline typename Eigen::Matrix<return_type_t<T_x1, T_x2, T_s, T_l>,
185 Eigen::Dynamic, Eigen::Dynamic>
186gp_matern32_cov(const std::vector<T_x1> &x1, const std::vector<T_x2> &x2,
187 const T_s &sigma, const T_l &length_scale) {
188 using std::exp;
189
190 size_t x1_size = stan::math::size(x1);
191 size_t x2_size = stan::math::size(x2);
192 Eigen::Matrix<return_type_t<T_x1, T_x2, T_s, T_l>, Eigen::Dynamic,
193 Eigen::Dynamic>
194 cov(x1_size, x2_size);
195
196 if (x1_size == 0 || x2_size == 0) {
197 return cov;
198 }
199
200 const char *function = "gp_matern32_cov";
201 size_t x1_obs_size = stan::math::size(x1[0]);
202 for (size_t i = 0; i < x1_size; ++i) {
203 check_size_match(function, "x1's row", x1_obs_size, "x1's other row",
204 stan::math::size(x1[i]));
205 }
206 for (size_t i = 0; i < x2_size; ++i) {
207 check_size_match(function, "x1's row", x1_obs_size, "x2's other row",
208 stan::math::size(x2[i]));
209 }
210
211 for (size_t n = 0; n < x1_size; ++n) {
212 check_not_nan(function, "x1", x1[n]);
213 }
214 for (size_t n = 0; n < x2_size; ++n) {
215 check_not_nan(function, "x2", x2[n]);
216 }
217
218 check_positive_finite(function, "magnitude", sigma);
219 check_positive_finite(function, "length scale", length_scale);
220
221 T_s sigma_sq = square(sigma);
222 T_l root_3_inv_l_sq = std::sqrt(3.0) / length_scale;
223
224 size_t block_size = 10;
225 for (size_t ib = 0; ib < x1.size(); ib += block_size) {
226 for (size_t jb = 0; jb < x2.size(); jb += block_size) {
227 size_t j_end = std::min(x2_size, jb + block_size);
228 for (size_t j = jb; j < j_end; ++j) {
229 size_t i_end = std::min(x1_size, ib + block_size);
230 for (size_t i = ib; i < i_end; ++i) {
231 return_type_t<T_x1, T_x2> dist = distance(x1[i], x2[j]);
232 cov(i, j) = sigma_sq * (1.0 + root_3_inv_l_sq * dist)
233 * exp(-root_3_inv_l_sq * dist);
234 }
235 }
236 }
237 }
238 return cov;
239}
240
265template <typename T_x1, typename T_x2, typename T_s, typename T_l>
266inline typename Eigen::Matrix<return_type_t<T_x1, T_x2, T_s, T_l>,
267 Eigen::Dynamic, Eigen::Dynamic>
268gp_matern32_cov(const std::vector<Eigen::Matrix<T_x1, -1, 1>> &x1,
269 const std::vector<Eigen::Matrix<T_x2, -1, 1>> &x2,
270 const T_s &sigma, const std::vector<T_l> &length_scale) {
271 using std::exp;
272
273 size_t x1_size = stan::math::size(x1);
274 size_t x2_size = stan::math::size(x2);
275 Eigen::Matrix<return_type_t<T_x1, T_x2, T_s, T_l>, Eigen::Dynamic,
276 Eigen::Dynamic>
277 cov(x1_size, x2_size);
278
279 if (x1_size == 0 || x2_size == 0) {
280 return cov;
281 }
282
283 const char *function = "gp_matern_32_cov";
284 for (size_t n = 0; n < x1_size; ++n) {
285 check_not_nan(function, "x1", x1[n]);
286 }
287 for (size_t n = 0; n < x2_size; ++n) {
288 check_not_nan(function, "x2", x2[n]);
289 }
290
291 check_positive_finite(function, "magnitude", sigma);
292 check_positive_finite(function, "length scale", length_scale);
293
294 size_t l_size = length_scale.size();
295 for (size_t n = 0; n < l_size; ++n) {
296 check_not_nan(function, "length scale", length_scale[n]);
297 }
298
299 for (size_t i = 0; i < x1_size; ++i) {
300 check_size_match(function, "x1's row", stan::math::size(x1[i]),
301 "number of length scales", l_size);
302 }
303 for (size_t i = 0; i < x2_size; ++i) {
304 check_size_match(function, "x2's row", stan::math::size(x2[i]),
305 "number of length scales", l_size);
306 }
307
308 T_s sigma_sq = square(sigma);
309 double root_3 = std::sqrt(3.0);
310
311 std::vector<Eigen::Matrix<return_type_t<T_x1, T_l>, -1, 1>> x1_new
312 = divide_columns(x1, length_scale);
313 std::vector<Eigen::Matrix<return_type_t<T_x2, T_l>, -1, 1>> x2_new
314 = divide_columns(x2, length_scale);
315
316 size_t block_size = 10;
317
318 for (size_t ib = 0; ib < x1.size(); ib += block_size) {
319 for (size_t jb = 0; jb < x2.size(); jb += block_size) {
320 size_t j_end = std::min(x2_size, jb + block_size);
321 for (size_t j = jb; j < j_end; ++j) {
322 size_t i_end = std::min(x1_size, ib + block_size);
323 for (size_t i = ib; i < i_end; ++i) {
324 return_type_t<T_x1, T_x2, T_l> dist = distance(x1_new[i], x2_new[j]);
325 cov(i, j) = sigma_sq * (1.0 + root_3 * dist) * exp(-root_3 * dist);
326 }
327 }
328 }
329 }
330 return cov;
331}
332} // namespace math
333} // namespace stan
334#endif
void divide_columns(matrix_cl< T1 > &A, const matrix_cl< T2 > &B)
Divides each column of a matrix by a vector.
matrix_cl< return_type_t< T1, T2, T3 > > gp_matern32_cov(const T1 &x, const T2 sigma, const T3 length_scale)
Matern 3/2 kernel on the GPU.
typename return_type< Ts... >::type return_type_t
Convenience type for the return type of the specified template parameters.
int64_t size(const T &m)
Returns the size (number of the elements) of a matrix_cl or var_value<matrix_cl<T>>.
Definition size.hpp:19
auto distance(const T_a &a, const T_b &b)
Returns the distance between the specified vectors.
Definition distance.hpp:33
void check_not_nan(const char *function, const char *name, const T_y &y)
Check if y is not NaN.
void check_size_match(const char *function, const char *name_i, T_size1 i, const char *name_j, T_size2 j)
Check if the provided sizes match.
void check_positive_finite(const char *function, const char *name, const T_y &y)
Check if y is positive and finite.
fvar< T > square(const fvar< T > &x)
Definition square.hpp:12
fvar< T > exp(const fvar< T > &x)
Definition exp.hpp:13
The lgamma implementation in stan-math is based on either the reentrant safe lgamma_r implementation ...