LSST Applications g070148d5b3+33e5256705,g0d53e28543+25c8b88941,g0da5cf3356+2dd1178308,g1081da9e2a+62d12e78cb,g17e5ecfddb+7e422d6136,g1c76d35bf8+ede3a706f7,g295839609d+225697d880,g2e2c1a68ba+cc1f6f037e,g2ffcdf413f+853cd4dcde,g38293774b4+62d12e78cb,g3b44f30a73+d953f1ac34,g48ccf36440+885b902d19,g4b2f1765b6+7dedbde6d2,g5320a0a9f6+0c5d6105b6,g56b687f8c9+ede3a706f7,g5c4744a4d9+ef6ac23297,g5ffd174ac0+0c5d6105b6,g6075d09f38+66af417445,g667d525e37+2ced63db88,g670421136f+2ced63db88,g71f27ac40c+2ced63db88,g774830318a+463cbe8d1f,g7876bc68e5+1d137996f1,g7985c39107+62d12e78cb,g7fdac2220c+0fd8241c05,g96f01af41f+368e6903a7,g9ca82378b8+2ced63db88,g9d27549199+ef6ac23297,gabe93b2c52+e3573e3735,gb065e2a02a+3dfbe639da,gbc3249ced9+0c5d6105b6,gbec6a3398f+0c5d6105b6,gc9534b9d65+35b9f25267,gd01420fc67+0c5d6105b6,geee7ff78d7+a14128c129,gf63283c776+ede3a706f7,gfed783d017+0c5d6105b6,w.2022.47
LSST Data Management Base Package
Loading...
Searching...
No Matches
GaussianProcess.cc
Go to the documentation of this file.
1// - * - LSST - C++ - * -
2
3/*
4 * LSST Data Management System
5 * Copyright 2008, 2009, 2010 LSST Corporation.
6 *
7 * This product includes software developed by the
8 * LSST Project (http://www.lsst.org/).
9 *
10 * This program is free software: you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation, either version 3 of the License, or
13 * (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
19 *
20 * You should have received a copy of the LSST License Statement and
21 * the GNU General Public License along with this program. If not,
22 * see < http://www.lsstcorp.org/LegalNotices/ > .
23 */
24
25#include <iostream>
26#include <cmath>
27
29
30using namespace std;
31
32namespace lsst {
33namespace afw {
34namespace math {
35
37 _interpolationCount = 0;
38 _iterationTime = 0.0;
39 _eigenTime = 0.0;
40 _searchTime = 0.0;
41 _varianceTime = 0.0;
42 _totalTime = 0.0;
43}
44
46 _interpolationCount = 0;
47 _iterationTime = 0.0;
48 _eigenTime = 0.0;
49 _searchTime = 0.0;
50 _varianceTime = 0.0;
51 _totalTime = 0.0;
52}
53
55 _before = double(lsst::daf::base::DateTime::now().get() * 24 * 60 * 60);
56 _beginning = _before;
57}
58
60 double after;
61 after = double(lsst::daf::base::DateTime::now().get() * 24 * 60 * 60);
62 _eigenTime += after - _before;
63 _before = after;
64}
65
67 double after;
68 after = double(lsst::daf::base::DateTime::now().get() * 24 * 60 * 60);
69 _varianceTime += after - _before;
70 _before = after;
71}
72
74 double after;
75 after = double(lsst::daf::base::DateTime::now().get() * 24 * 60 * 60);
76 _searchTime += after - _before;
77 _before = after;
78}
79
81 double after;
82 after = double(lsst::daf::base::DateTime::now().get() * 24 * 60 * 60);
83 _iterationTime += after - _before;
84 _before = after;
85}
86
88 double after;
89 after = double(lsst::daf::base::DateTime::now().get() * 24 * 60 * 60);
90 _totalTime += after - _beginning;
91 _interpolationCount += i;
92}
93
95 std::cout << "\nSearch time " << _searchTime << "\n";
96 std::cout << "Eigen time " << _eigenTime << "\n";
97 std::cout << "Variance time " << _varianceTime << "\n";
98 std::cout << "Iteration time " << _iterationTime << "\n";
99 std::cout << "Total time " << _totalTime << "\n";
100 std::cout << "Number of interpolations " << _interpolationCount << "\n";
101}
102
103template <typename T>
104void KdTree<T>::Initialize(ndarray::Array<T, 2, 2> const &dt) {
105 int i;
106
107 _npts = dt.template getSize<0>();
108 _dimensions = dt.template getSize<1>();
109
110 // a buffer to use when first building the tree
111 _inn = allocate(ndarray::makeVector(_npts));
112
113 _roomStep = 5000;
114 _room = _npts;
115
116 _data = allocate(ndarray::makeVector(_room, _dimensions));
117
118 _data.deep() = dt;
119
120 _tree = allocate(ndarray::makeVector(_room, 4));
121
122 for (i = 0; i < _npts; i++) {
123 _inn[i] = i;
124 }
125
126 _organize(_inn, _npts, -1, -1);
127
128 i = _testTree();
129 if (i == 0) {
130 throw LSST_EXCEPT(lsst::pex::exceptions::RuntimeError, "Failed to properly initialize KdTree\n");
131 }
132}
133
134template <typename T>
135void KdTree<T>::findNeighbors(ndarray::Array<int, 1, 1> neighdex, ndarray::Array<double, 1, 1> dd,
136 ndarray::Array<const T, 1, 1> const &v, int n_nn) const {
137 if (n_nn > _npts) {
139 "Asked for more neighbors than kd tree contains\n");
140 }
141
142 if (n_nn <= 0) {
144 "Asked for zero or a negative number of neighbors\n");
145 }
146
147 if (neighdex.getNumElements() != static_cast<ndarray::Size>(n_nn)) {
149 "Size of neighdex does not equal n_nn in KdTree.findNeighbors\n");
150 }
151
152 if (dd.getNumElements() != static_cast<ndarray::Size>(n_nn)) {
154 "Size of dd does not equal n_nn in KdTree.findNeighbors\n");
155 }
156
157 int i, start;
158
159 _neighborCandidates = allocate(ndarray::makeVector(n_nn));
160 _neighborDistances = allocate(ndarray::makeVector(n_nn));
161 _neighborsFound = 0;
162 _neighborsWanted = n_nn;
163
164 for (i = 0; i < n_nn; i++) _neighborDistances[i] = -1.0;
165
166 start = _findNode(v);
167
168 _neighborDistances[0] = _distance(v, _data[start]);
169 _neighborCandidates[0] = start;
170 _neighborsFound = 1;
171
172 for (i = 1; i < 4; i++) {
173 if (_tree[start][i] >= 0) {
174 _lookForNeighbors(v, _tree[start][i], start);
175 }
176 }
177
178 for (i = 0; i < n_nn; i++) {
179 neighdex[i] = _neighborCandidates[i];
180 dd[i] = _neighborDistances[i];
181 }
182}
183
184template <typename T>
185T KdTree<T>::getData(int ipt, int idim) const {
186 if (ipt >= _npts || ipt < 0) {
188 "Asked for point that does not exist in KdTree\n");
189 }
190
191 if (idim >= _dimensions || idim < 0) {
193 "KdTree does not have points of that many dimensions\n");
194 }
195
196 return _data[ipt][idim];
197}
198
199template <typename T>
200ndarray::Array<T, 1, 1> KdTree<T>::getData(int ipt) const {
201 if (ipt >= _npts || ipt < 0) {
203 "Asked for point that does not exist in KdTree\n");
204 }
205
206 return _data[ipt];
207}
208
209template <typename T>
210void KdTree<T>::addPoint(ndarray::Array<const T, 1, 1> const &v) {
211 if (v.getNumElements() != static_cast<ndarray::Size>(_dimensions)) {
213 "You are trying to add a point of the incorrect dimensionality to KdTree\n");
214 }
215
216 int i, j, node, dim, dir;
217
218 node = _findNode(v);
219 dim = _tree[node][DIMENSION] + 1;
220 if (dim == _dimensions) dim = 0;
221
222 if (_npts == _room) {
223 ndarray::Array<T, 2, 2> dbuff = allocate(ndarray::makeVector(_npts, _dimensions));
224 ndarray::Array<int, 2, 2> tbuff = allocate(ndarray::makeVector(_npts, 4));
225
226 dbuff.deep() = _data;
227 tbuff.deep() = _tree;
228
229 _room += _roomStep;
230
231 _tree = allocate(ndarray::makeVector(_room, 4));
232 _data = allocate(ndarray::makeVector(_room, _dimensions));
233
234 for (i = 0; i < _npts; i++) {
235 for (j = 0; j < _dimensions; j++) _data[i][j] = dbuff[i][j];
236 for (j = 0; j < 4; j++) _tree[i][j] = tbuff[i][j];
237 }
238 }
239
240 _tree[_npts][DIMENSION] = dim;
241 _tree[_npts][PARENT] = node;
242 i = _tree[node][DIMENSION];
243
244 if (_data[node][i] > v[i]) {
245 if (_tree[node][LT] >= 0) {
247 "Trying to add to KdTree in a node that is already occupied\n");
248 }
249 _tree[node][LT] = _npts;
250 dir = LT;
251 } else {
252 if (_tree[node][GEQ] >= 0) {
254 "Trying to add to KdTree in a node that is already occupied\n");
255 }
256 _tree[node][GEQ] = _npts;
257 dir = GEQ;
258 }
259 _tree[_npts][LT] = -1;
260 _tree[_npts][GEQ] = -1;
261 for (i = 0; i < _dimensions; i++) {
262 _data[_npts][i] = v[i];
263 }
264
265 _npts++;
266
267 i = _walkUpTree(_tree[_npts - 1][PARENT], dir, _npts - 1);
268 if (i != _masterParent) {
269 throw LSST_EXCEPT(lsst::pex::exceptions::RuntimeError, "Adding to KdTree failed\n");
270 }
271}
272
273template <typename T>
275 return _npts;
276}
277
278template <typename T>
279void KdTree<T>::getTreeNode(ndarray::Array<int, 1, 1> const &v, int dex) const {
280 if (dex < 0 || dex >= _npts) {
282 "Asked for tree information on point that does not exist\n");
283 }
284
285 if (v.getNumElements() != 4) {
287 "Need to pass a 4-element ndarray into KdTree.getTreeNode()\n");
288 }
289
290 v[0] = _tree[dex][DIMENSION];
291 v[1] = _tree[dex][LT];
292 v[2] = _tree[dex][GEQ];
293 v[3] = _tree[dex][PARENT];
294}
295
296template <typename T>
297int KdTree<T>::_testTree() const {
298 if (_npts == 1 && _tree[0][PARENT] < 0 && _masterParent == 0) {
299 // there is only one point in the tree
300 return 1;
301 }
302
303 int i, j, output;
304 std::vector<int> isparent;
305
306 output = 1;
307
308 for (i = 0; i < _npts; i++) isparent.push_back(0);
309
310 j = 0;
311 for (i = 0; i < _npts; i++) {
312 if (_tree[i][PARENT] < 0) j++;
313 }
314 if (j != 1) {
315 std::cout << "_tree FAILURE " << j << " _masterParents\n";
316 return 0;
317 }
318
319 for (i = 0; i < _npts; i++) {
320 if (_tree[i][PARENT] >= 0) {
321 isparent[_tree[i][PARENT]]++;
322 }
323 }
324
325 for (i = 0; i < _npts; i++) {
326 if (isparent[i] > 2) {
327 std::cout << "_tree FAILURE " << i << " is parent to " << isparent[i] << "\n";
328 return 0;
329 }
330 }
331
332 for (i = 0; i < _npts; i++) {
333 if (_tree[i][PARENT] >= 0) {
334 if (_tree[_tree[i][PARENT]][LT] == i)
335 j = LT;
336 else
337 j = GEQ;
338 output = _walkUpTree(_tree[i][PARENT], j, i);
339
340 if (output != _masterParent) {
341 return 0;
342 }
343 }
344 }
345
346 if (output != _masterParent)
347 return 0;
348 else
349 return 1;
350}
351
352template <typename T>
353void KdTree<T>::_organize(ndarray::Array<int, 1, 1> const &use, int ct, int parent, int dir) {
354 int i, j, k, l, idim, daughter;
355 T mean, var, varbest;
356
358 std::vector<T> toSortElement;
359
360 if (ct > 1) {
361 // below is code to choose the dimension on which the available points
362 // have the greates variance. This will be the dimension on which
363 // the daughter node splits the data
364 idim = 0;
365 varbest = -1.0;
366 for (i = 0; i < _dimensions; i++) {
367 mean = 0.0;
368 var = 0.0;
369 for (j = 0; j < ct; j++) {
370 mean += _data[use[j]][i];
371 var += _data[use[j]][i] * _data[use[j]][i];
372 }
373 mean = mean / double(ct);
374 var = var / double(ct) - mean * mean;
375 if (i == 0 || var > varbest || (var == varbest && parent >= 0 && i > _tree[parent][DIMENSION])) {
376 idim = i;
377 varbest = var;
378 }
379 } // for(i = 0;i < _dimensions;i++ )
380
381 // we need to sort the available data points according to their idim - th element
382 // but we need to keep track of their original indices, so we can refer back to
383 // their positions in _data. Therefore, the code constructs a 2 - dimensional
384 // std::vector. The first dimension contains the element to be sorted.
385 // The second dimension contains the original index information.
386 // After sorting, the (rearranged) index information will be stored back in the
387 // ndarray use[]
388
389 toSortElement.push_back(_data[use[0]][idim]);
390 toSortElement.push_back(T(use[0]));
391 toSort.push_back(toSortElement);
392 for (i = 1; i < ct; i++) {
393 toSortElement[0] = _data[use[i]][idim];
394 toSortElement[1] = T(use[i]);
395 toSort.push_back(toSortElement);
396 }
397
398 std::stable_sort(toSort.begin(), toSort.end());
399
400 k = ct / 2;
401 l = ct / 2;
402 while (k > 0 && toSort[k][0] == toSort[k - 1][0]) k--;
403
404 while (l < ct - 1 && toSort[l][0] == toSort[ct / 2][0]) l++;
405
406 if ((ct / 2 - k) < (l - ct / 2) || l == ct - 1)
407 j = k;
408 else
409 j = l;
410
411 for (i = 0; i < ct; i++) {
412 use[i] = int(toSort[i][1]);
413 }
414 daughter = use[j];
415
416 if (parent >= 0) _tree[parent][dir] = daughter;
417 _tree[daughter][DIMENSION] = idim;
418 _tree[daughter][PARENT] = parent;
419
420 if (j < ct - 1) {
421 _organize(use[ndarray::view(j + 1, use.getSize<0>())], ct - j - 1, daughter, GEQ);
422 } else
423 _tree[daughter][GEQ] = -1;
424
425 if (j > 0) {
426 _organize(use, j, daughter, LT);
427 } else
428 _tree[daughter][LT] = -1;
429
430 } // if(ct > 1)
431 else {
432 daughter = use[0];
433
434 if (parent >= 0) _tree[parent][dir] = daughter;
435
436 idim = _tree[parent][DIMENSION] + 1;
437
438 if (idim >= _dimensions) idim = 0;
439
440 _tree[daughter][DIMENSION] = idim;
441 _tree[daughter][LT] = -1;
442 _tree[daughter][GEQ] = -1;
443 _tree[daughter][PARENT] = parent;
444 }
445
446 if (parent == -1) {
447 _masterParent = daughter;
448 }
449}
450
451template <typename T>
452int KdTree<T>::_findNode(ndarray::Array<const T, 1, 1> const &v) const {
453 int consider, next, dim;
454
455 dim = _tree[_masterParent][DIMENSION];
456
457 if (v[dim] < _data[_masterParent][dim])
458 consider = _tree[_masterParent][LT];
459 else
460 consider = _tree[_masterParent][GEQ];
461
462 next = consider;
463
464 while (next >= 0) {
465 consider = next;
466
467 dim = _tree[consider][DIMENSION];
468 if (v[dim] < _data[consider][dim])
469 next = _tree[consider][LT];
470 else
471 next = _tree[consider][GEQ];
472 }
473
474 return consider;
475}
476
477template <typename T>
478void KdTree<T>::_lookForNeighbors(ndarray::Array<const T, 1, 1> const &v, int consider, int from) const {
479 int i, j, going;
480 double dd;
481
482 dd = _distance(v, _data[consider]);
483
484 if (_neighborsFound < _neighborsWanted || dd < _neighborDistances[_neighborsWanted - 1]) {
485 for (j = 0; j < _neighborsFound && _neighborDistances[j] < dd; j++)
486 ;
487
488 for (i = _neighborsWanted - 1; i > j; i--) {
489 _neighborDistances[i] = _neighborDistances[i - 1];
490 _neighborCandidates[i] = _neighborCandidates[i - 1];
491 }
492
493 _neighborDistances[j] = dd;
494 _neighborCandidates[j] = consider;
495
496 if (_neighborsFound < _neighborsWanted) _neighborsFound++;
497 }
498
499 if (_tree[consider][PARENT] == from) {
500 // you came here from the parent
501
502 i = _tree[consider][DIMENSION];
503 dd = v[i] - _data[consider][i];
504 if ((dd <= _neighborDistances[_neighborsFound - 1] || _neighborsFound < _neighborsWanted) &&
505 _tree[consider][LT] >= 0) {
506 _lookForNeighbors(v, _tree[consider][LT], consider);
507 }
508
509 dd = _data[consider][i] - v[i];
510 if ((dd <= _neighborDistances[_neighborsFound - 1] || _neighborsFound < _neighborsWanted) &&
511 _tree[consider][GEQ] >= 0) {
512 _lookForNeighbors(v, _tree[consider][GEQ], consider);
513 }
514 } else {
515 // you came here from one of the branches
516
517 // descend the other branch
518 if (_tree[consider][LT] == from) {
519 going = GEQ;
520 } else {
521 going = LT;
522 }
523
524 j = _tree[consider][going];
525
526 if (j >= 0) {
527 i = _tree[consider][DIMENSION];
528
529 if (going == 1)
530 dd = v[i] - _data[consider][i];
531 else
532 dd = _data[consider][i] - v[i];
533
534 if (dd <= _neighborDistances[_neighborsFound - 1] || _neighborsFound < _neighborsWanted) {
535 _lookForNeighbors(v, j, consider);
536 }
537 }
538
539 // ascend to the parent
540 if (_tree[consider][PARENT] >= 0) {
541 _lookForNeighbors(v, _tree[consider][PARENT], consider);
542 }
543 }
544}
545
546template <typename T>
547int KdTree<T>::_walkUpTree(int target, int dir, int root) const {
548 // target is the node that you are examining now
549 // dir is where you came from
550 // root is the ultimate point from which you started
551
552 int i, output;
553
554 output = 1;
555
556 if (dir == LT) {
557 if (_data[root][_tree[target][DIMENSION]] >= _data[target][_tree[target][DIMENSION]]) {
558 return 0;
559 }
560 } else {
561 if (_data[root][_tree[target][DIMENSION]] < _data[target][_tree[target][DIMENSION]]) {
562 return 0;
563 }
564 }
565
566 if (_tree[target][PARENT] >= 0) {
567 if (_tree[_tree[target][PARENT]][LT] == target)
568 i = LT;
569 else
570 i = GEQ;
571 output = output * _walkUpTree(_tree[target][PARENT], i, root);
572
573 } else {
574 output = output * target;
575 // so that it will return _masterParent
576 // make sure everything is connected to _masterParent
577 }
578 return output;
579}
580
581template <typename T>
583 if (target < 0 || target >= _npts) {
585 "You are trying to remove a point that doesn't exist from KdTree\n");
586 }
587
588 if (_npts == 1) {
590 "There is only one point left in this KdTree. You cannot call removePoint\n");
591 }
592
593 int nl, nr, i, j, k, side;
594 int root;
595
596 nl = 0;
597 nr = 0;
598 if (_tree[target][LT] >= 0) {
599 nl++;
600 _count(_tree[target][LT], &nl);
601 }
602
603 if (_tree[target][GEQ] >= 0) {
604 nr++;
605 _count(_tree[target][GEQ], &nr);
606 }
607
608 if (nl == 0 && nr == 0) {
609 k = _tree[target][PARENT];
610
611 if (_tree[k][LT] == target)
612 _tree[k][LT] = -1;
613 else if (_tree[k][GEQ] == target)
614 _tree[k][GEQ] = -1;
615
616 } // if target is terminal
617 else if ((nl == 0 && nr > 0) || (nr == 0 && nl > 0)) {
618 if (nl == 0)
619 side = GEQ;
620 else
621 side = LT;
622
623 k = _tree[target][PARENT];
624 if (k >= 0) {
625 if (_tree[k][LT] == target) {
626 _tree[k][LT] = _tree[target][side];
627 _tree[_tree[k][LT]][PARENT] = k;
628
629 } else {
630 _tree[k][GEQ] = _tree[target][side];
631 _tree[_tree[k][GEQ]][PARENT] = k;
632 }
633 } else {
634 _masterParent = _tree[target][side];
635 _tree[_tree[target][side]][PARENT] = -1;
636 }
637 } // if only one side is populated
638 else {
639 if (nl > nr)
640 side = LT;
641 else
642 side = GEQ;
643
644 k = _tree[target][PARENT];
645 if (k < 0) {
646 _masterParent = _tree[target][side];
647 _tree[_masterParent][PARENT] = -1;
648 } else {
649 if (_tree[k][LT] == target) {
650 _tree[k][LT] = _tree[target][side];
651 _tree[_tree[k][LT]][PARENT] = k;
652 } else {
653 _tree[k][GEQ] = _tree[target][side];
654 _tree[_tree[k][GEQ]][PARENT] = k;
655 }
656 }
657
658 root = _tree[target][3 - side];
659
660 _descend(root);
661 } // if both sides are populated
662
663 if (target < _npts - 1) {
664 for (i = target + 1; i < _npts; i++) {
665 for (j = 0; j < 4; j++) _tree[i - 1][j] = _tree[i][j];
666
667 for (j = 0; j < _dimensions; j++) _data[i - 1][j] = _data[i][j];
668 }
669
670 for (i = 0; i < _npts; i++) {
671 for (j = 1; j < 4; j++) {
672 if (_tree[i][j] > target) _tree[i][j]--;
673 }
674 }
675
676 if (_masterParent > target) _masterParent--;
677 }
678 _npts--;
679
680 i = _testTree();
681 if (i == 0) {
682 throw LSST_EXCEPT(lsst::pex::exceptions::RuntimeError, "Subtracting from KdTree failed\n");
683 }
684}
685
686template <typename T>
687void KdTree<T>::_count(int where, int *ct) const {
688 // a way to count the number of vital elements on a given branch
689
690 if (_tree[where][LT] >= 0) {
691 ct[0]++;
692 _count(_tree[where][LT], ct);
693 }
694 if (_tree[where][GEQ] >= 0) {
695 ct[0]++;
696 _count(_tree[where][GEQ], ct);
697 }
698}
699
700template <typename T>
701void KdTree<T>::_reassign(int target) {
702 int where, dir, k;
703
704 where = _masterParent;
705 if (_data[target][_tree[where][DIMENSION]] < _data[where][_tree[where][DIMENSION]])
706 dir = LT;
707 else
708 dir = GEQ;
709
710 k = _tree[where][dir];
711 while (k >= 0) {
712 where = k;
713 if (_data[target][_tree[where][DIMENSION]] < _data[where][_tree[where][DIMENSION]])
714 dir = LT;
715 else
716 dir = GEQ;
717 k = _tree[where][dir];
718 }
719
720 _tree[where][dir] = target;
721 _tree[target][PARENT] = where;
722 _tree[target][LT] = -1;
723 _tree[target][GEQ] = -1;
724 _tree[target][DIMENSION] = _tree[where][DIMENSION] + 1;
725
726 if (_tree[target][DIMENSION] == _dimensions) _tree[target][DIMENSION] = 0;
727}
728
729template <typename T>
730void KdTree<T>::_descend(int root) {
731 if (_tree[root][LT] >= 0) _descend(_tree[root][LT]);
732 if (_tree[root][GEQ] >= 0) _descend(_tree[root][GEQ]);
733
734 _reassign(root);
735}
736
737template <typename T>
738double KdTree<T>::_distance(ndarray::Array<const T, 1, 1> const &p1,
739 ndarray::Array<const T, 1, 1> const &p2) const {
740 int i, dd;
741 double ans;
742 ans = 0.0;
743 dd = p1.template getSize<0>();
744
745 for (i = 0; i < dd; i++) ans += (p1[i] - p2[i]) * (p1[i] - p2[i]);
746
747 return ::sqrt(ans);
748}
749
750template <typename T>
751GaussianProcess<T>::GaussianProcess(ndarray::Array<T, 2, 2> const &dataIn, ndarray::Array<T, 1, 1> const &ff,
752 std::shared_ptr<Covariogram<T> > const &covarIn)
753
754{
755 int i;
756
757 _covariogram = covarIn;
758
759 _npts = dataIn.template getSize<0>();
760 _dimensions = dataIn.template getSize<1>();
761
762 _room = _npts;
763 _roomStep = 5000;
764
765 _nFunctions = 1;
766 _function = allocate(ndarray::makeVector(_npts, 1));
767
768 if (ff.getNumElements() != static_cast<ndarray::Size>(_npts)) {
770 "You did not pass in the same number of data points as function values\n");
771 }
772
773 for (i = 0; i < _npts; i++) _function[i][0] = ff[i];
774
775 _krigingParameter = T(1.0);
776 _lambda = T(1.0e-5);
777
778 _useMaxMin = 0;
779
780 _kdTree.Initialize(dataIn);
781
782 _npts = _kdTree.getNPoints();
783}
784
785template <typename T>
786GaussianProcess<T>::GaussianProcess(ndarray::Array<T, 2, 2> const &dataIn, ndarray::Array<T, 1, 1> const &mn,
787 ndarray::Array<T, 1, 1> const &mx, ndarray::Array<T, 1, 1> const &ff,
788 std::shared_ptr<Covariogram<T> > const &covarIn) {
789 int i, j;
790 ndarray::Array<T, 2, 2> normalizedData;
791
792 _covariogram = covarIn;
793
794 _npts = dataIn.template getSize<0>();
795 _dimensions = dataIn.template getSize<1>();
796 _room = _npts;
797 _roomStep = 5000;
798
799 if (ff.getNumElements() != static_cast<ndarray::Size>(_npts)) {
801 "You did not pass in the same number of data points as function values\n");
802 }
803
804 if (mn.getNumElements() != static_cast<ndarray::Size>(_dimensions) ||
805 mx.getNumElements() != static_cast<ndarray::Size>(_dimensions)) {
807 "Your min/max values have different dimensionality than your data points\n");
808 }
809
810 _krigingParameter = T(1.0);
811
812 _lambda = T(1.0e-5);
813 _krigingParameter = T(1.0);
814
815 _max = allocate(ndarray::makeVector(_dimensions));
816 _min = allocate(ndarray::makeVector(_dimensions));
817 _max.deep() = mx;
818 _min.deep() = mn;
819 _useMaxMin = 1;
820 normalizedData = allocate(ndarray::makeVector(_npts, _dimensions));
821 for (i = 0; i < _npts; i++) {
822 for (j = 0; j < _dimensions; j++) {
823 normalizedData[i][j] = (dataIn[i][j] - _min[j]) / (_max[j] - _min[j]);
824 // note the normalization by _max - _min in each dimension
825 }
826 }
827
828 _kdTree.Initialize(normalizedData);
829
830 _npts = _kdTree.getNPoints();
831 _nFunctions = 1;
832 _function = allocate(ndarray::makeVector(_npts, 1));
833 for (i = 0; i < _npts; i++) _function[i][0] = ff[i];
834}
835
836template <typename T>
837GaussianProcess<T>::GaussianProcess(ndarray::Array<T, 2, 2> const &dataIn, ndarray::Array<T, 2, 2> const &ff,
838 std::shared_ptr<Covariogram<T> > const &covarIn) {
839 _covariogram = covarIn;
840
841 _npts = dataIn.template getSize<0>();
842 _dimensions = dataIn.template getSize<1>();
843
844 _room = _npts;
845 _roomStep = 5000;
846
847 if (ff.template getSize<0>() != static_cast<ndarray::Size>(_npts)) {
849 "You did not pass in the same number of data points as function values\n");
850 }
851
852 _nFunctions = ff.template getSize<1>();
853 _function = allocate(ndarray::makeVector(_npts, _nFunctions));
854 _function.deep() = ff;
855
856 _krigingParameter = T(1.0);
857
858 _lambda = T(1.0e-5);
859
860 _useMaxMin = 0;
861
862 _kdTree.Initialize(dataIn);
863
864 _npts = _kdTree.getNPoints();
865}
866
867template <typename T>
868GaussianProcess<T>::GaussianProcess(ndarray::Array<T, 2, 2> const &dataIn, ndarray::Array<T, 1, 1> const &mn,
869 ndarray::Array<T, 1, 1> const &mx, ndarray::Array<T, 2, 2> const &ff,
870 std::shared_ptr<Covariogram<T> > const &covarIn) {
871 int i, j;
872 ndarray::Array<T, 2, 2> normalizedData;
873
874 _covariogram = covarIn;
875
876 _npts = dataIn.template getSize<0>();
877 _dimensions = dataIn.template getSize<1>();
878
879 _room = _npts;
880 _roomStep = 5000;
881
882 if (ff.template getSize<0>() != static_cast<ndarray::Size>(_npts)) {
884 "You did not pass in the same number of data points as function values\n");
885 }
886
887 if (mn.getNumElements() != static_cast<ndarray::Size>(_dimensions) ||
888 mx.getNumElements() != static_cast<ndarray::Size>(_dimensions)) {
890 "Your min/max values have different dimensionality than your data points\n");
891 }
892
893 _krigingParameter = T(1.0);
894
895 _lambda = T(1.0e-5);
896 _krigingParameter = T(1.0);
897
898 _max = allocate(ndarray::makeVector(_dimensions));
899 _min = allocate(ndarray::makeVector(_dimensions));
900 _max.deep() = mx;
901 _min.deep() = mn;
902 _useMaxMin = 1;
903 normalizedData = allocate(ndarray::makeVector(_npts, _dimensions));
904 for (i = 0; i < _npts; i++) {
905 for (j = 0; j < _dimensions; j++) {
906 normalizedData[i][j] = (dataIn[i][j] - _min[j]) / (_max[j] - _min[j]);
907 // note the normalization by _max - _min in each dimension
908 }
909 }
910
911 _kdTree.Initialize(normalizedData);
912 _npts = _kdTree.getNPoints();
913 _nFunctions = ff.template getSize<1>();
914 _function = allocate(ndarray::makeVector(_npts, _nFunctions));
915 _function.deep() = ff;
916}
917
918template <typename T>
920 return _npts;
921}
922
923template <typename T>
925 return _dimensions;
926}
927
928template <typename T>
929void GaussianProcess<T>::getData(ndarray::Array<T, 2, 2> pts, ndarray::Array<T, 1, 1> fn,
930 ndarray::Array<int, 1, 1> indices) const {
931 if (_nFunctions != 1) {
933 "Your function value array does not have enough room for all of the functions "
934 "in your GaussianProcess.\n");
935 }
936
937 if (pts.template getSize<1>() != static_cast<ndarray::Size>(_dimensions)) {
939 "Your pts array is constructed for points of the wrong dimensionality.\n");
940 }
941
942 if (pts.template getSize<0>() != indices.getNumElements()) {
944 "You did not put enough room in your pts array to fit all the points "
945 "you asked for in your indices array.\n");
946 }
947
948 if (fn.template getSize<0>() != indices.getNumElements()) {
950 "You did not provide enough room in your function value array "
951 "for all of the points you requested in your indices array.\n");
952 }
953
954 for (ndarray::Size i = 0; i < indices.template getSize<0>(); i++) {
955 pts[i] = _kdTree.getData(indices[i]); // do this first in case one of the indices is invalid.
956 // _kdTree.getData() will raise an exception in that case
957 fn[i] = _function[indices[i]][0];
958 if (_useMaxMin == 1) {
959 for (int j = 0; j < _dimensions; j++) {
960 pts[i][j] *= (_max[j] - _min[j]);
961 pts[i][j] += _min[j];
962 }
963 }
964 }
965}
966
967template <typename T>
968void GaussianProcess<T>::getData(ndarray::Array<T, 2, 2> pts, ndarray::Array<T, 2, 2> fn,
969 ndarray::Array<int, 1, 1> indices) const {
970 if (fn.template getSize<1>() != static_cast<ndarray::Size>(_nFunctions)) {
972 "Your function value array does not have enough room for all of the functions "
973 "in your GaussianProcess.\n");
974 }
975
976 if (pts.template getSize<1>() != static_cast<ndarray::Size>(_dimensions)) {
978 "Your pts array is constructed for points of the wrong dimensionality.\n");
979 }
980
981 if (pts.template getSize<0>() != indices.getNumElements()) {
983 "You did not put enough room in your pts array to fit all the points "
984 "you asked for in your indices array.\n");
985 }
986
987 if (fn.template getSize<0>() != indices.getNumElements()) {
989 "You did not provide enough room in your function value array "
990 "for all of the points you requested in your indices array.\n");
991 }
992
993 for (ndarray::Size i = 0; i < indices.template getSize<0>(); i++) {
994 pts[i] = _kdTree.getData(indices[i]); // do this first in case one of the indices is invalid.
995 // _kdTree.getData() will raise an exception in that case
996 for (int j = 0; j < _nFunctions; j++) {
997 fn[i][j] = _function[indices[i]][j];
998 }
999 if (_useMaxMin == 1) {
1000 for (int j = 0; j < _dimensions; j++) {
1001 pts[i][j] *= (_max[j] - _min[j]);
1002 pts[i][j] += _min[j];
1003 }
1004 }
1005 }
1006}
1007
1008template <typename T>
1009T GaussianProcess<T>::interpolate(ndarray::Array<T, 1, 1> variance, ndarray::Array<T, 1, 1> const &vin,
1010 int numberOfNeighbors) const {
1011 if (_nFunctions > 1) {
1013 "You need to call the version of GaussianProcess.interpolate() "
1014 "that accepts mu and variance arrays (which it populates with results). "
1015 "You are interpolating more than one function.");
1016 }
1017
1018 if (numberOfNeighbors <= 0) {
1020 "Asked for zero or negative number of neighbors\n");
1021 }
1022
1023 if (numberOfNeighbors > _kdTree.getNPoints()) {
1025 "Asked for more neighbors than you have data points\n");
1026 }
1027
1028 if (variance.getNumElements() != static_cast<ndarray::Size>(_nFunctions)) {
1030 "Your variance array is the incorrect size for the number "
1031 "of functions you are trying to interpolate\n");
1032 }
1033
1034 if (vin.getNumElements() != static_cast<ndarray::Size>(_dimensions)) {
1036 "You are interpolating at a point with different dimensionality than you data\n");
1037 }
1038
1039 int i, j;
1040 T fbar, mu;
1041
1042 ndarray::Array<T, 1, 1> covarianceTestPoint;
1043 ndarray::Array<int, 1, 1> neighbors;
1044 ndarray::Array<double, 1, 1> neighborDistances, vv;
1045
1046 Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> covariance, bb, xx;
1047 Eigen::LDLT<Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> > ldlt;
1048
1049 _timer.start();
1050
1051 bb.resize(numberOfNeighbors, 1);
1052 xx.resize(numberOfNeighbors, 1);
1053
1054 covariance.resize(numberOfNeighbors, numberOfNeighbors);
1055
1056 covarianceTestPoint = allocate(ndarray::makeVector(numberOfNeighbors));
1057
1058 neighbors = allocate(ndarray::makeVector(numberOfNeighbors));
1059
1060 neighborDistances = allocate(ndarray::makeVector(numberOfNeighbors));
1061
1062 vv = allocate(ndarray::makeVector(_dimensions));
1063 if (_useMaxMin == 1) {
1064 // if you constructed this Gaussian process with minimum and maximum
1065 // values for the dimensions of your parameter space,
1066 // the point you are interpolating must be scaled to match the data so
1067 // that the selected nearest neighbors are appropriate
1068
1069 for (i = 0; i < _dimensions; i++) vv[i] = (vin[i] - _min[i]) / (_max[i] - _min[i]);
1070 } else {
1071 vv = vin;
1072 }
1073
1074 _kdTree.findNeighbors(neighbors, neighborDistances, vv, numberOfNeighbors);
1075
1076 _timer.addToSearch();
1077
1078 fbar = 0.0;
1079 for (i = 0; i < numberOfNeighbors; i++) fbar += _function[neighbors[i]][0];
1080 fbar = fbar / double(numberOfNeighbors);
1081
1082 for (i = 0; i < numberOfNeighbors; i++) {
1083 covarianceTestPoint[i] = (*_covariogram)(vv, _kdTree.getData(neighbors[i]));
1084
1085 covariance(i, i) =
1086 (*_covariogram)(_kdTree.getData(neighbors[i]), _kdTree.getData(neighbors[i])) + _lambda;
1087
1088 for (j = i + 1; j < numberOfNeighbors; j++) {
1089 covariance(i, j) = (*_covariogram)(_kdTree.getData(neighbors[i]), _kdTree.getData(neighbors[j]));
1090 covariance(j, i) = covariance(i, j);
1091 }
1092 }
1093
1094 _timer.addToIteration();
1095
1096 // use Eigen's ldlt solver in place of matrix inversion (for speed purposes)
1097 ldlt.compute(covariance);
1098
1099 for (i = 0; i < numberOfNeighbors; i++) bb(i, 0) = _function[neighbors[i]][0] - fbar;
1100
1101 xx = ldlt.solve(bb);
1102 _timer.addToEigen();
1103
1104 mu = fbar;
1105
1106 for (i = 0; i < numberOfNeighbors; i++) {
1107 mu += covarianceTestPoint[i] * xx(i, 0);
1108 }
1109
1110 _timer.addToIteration();
1111
1112 variance(0) = (*_covariogram)(vv, vv) + _lambda;
1113
1114 for (i = 0; i < numberOfNeighbors; i++) bb(i) = covarianceTestPoint[i];
1115
1116 xx = ldlt.solve(bb);
1117
1118 for (i = 0; i < numberOfNeighbors; i++) {
1119 variance(0) -= covarianceTestPoint[i] * xx(i, 0);
1120 }
1121
1122 variance(0) = variance(0) * _krigingParameter;
1123
1124 _timer.addToVariance();
1125 _timer.addToTotal(1);
1126
1127 return mu;
1128}
1129
1130template <typename T>
1131void GaussianProcess<T>::interpolate(ndarray::Array<T, 1, 1> mu, ndarray::Array<T, 1, 1> variance,
1132 ndarray::Array<T, 1, 1> const &vin, int numberOfNeighbors) const {
1133 if (numberOfNeighbors <= 0) {
1135 "Asked for zero or negative number of neighbors\n");
1136 }
1137
1138 if (numberOfNeighbors > _kdTree.getNPoints()) {
1140 "Asked for more neighbors than you have data points\n");
1141 }
1142
1143 if (vin.getNumElements() != static_cast<ndarray::Size>(_dimensions)) {
1145 "You are interpolating at a point with different dimensionality than you data\n");
1146 }
1147
1148 if (mu.getNumElements() != static_cast<ndarray::Size>(_nFunctions) ||
1149 variance.getNumElements() != static_cast<ndarray::Size>(_nFunctions)) {
1151 "Your mu and/or var arrays are improperly sized for the number of functions "
1152 "you are interpolating\n");
1153 }
1154
1155 int i, j, ii;
1156 T fbar;
1157
1158 ndarray::Array<T, 1, 1> covarianceTestPoint;
1159 ndarray::Array<int, 1, 1> neighbors;
1160 ndarray::Array<double, 1, 1> neighborDistances, vv;
1161
1162 Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> covariance, bb, xx;
1163 Eigen::LDLT<Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> > ldlt;
1164
1165 _timer.start();
1166
1167 bb.resize(numberOfNeighbors, 1);
1168 xx.resize(numberOfNeighbors, 1);
1169 covariance.resize(numberOfNeighbors, numberOfNeighbors);
1170 covarianceTestPoint = allocate(ndarray::makeVector(numberOfNeighbors));
1171 neighbors = allocate(ndarray::makeVector(numberOfNeighbors));
1172 neighborDistances = allocate(ndarray::makeVector(numberOfNeighbors));
1173
1174 vv = allocate(ndarray::makeVector(_dimensions));
1175
1176 if (_useMaxMin == 1) {
1177 // if you constructed this Gaussian process with minimum and maximum
1178 // values for the dimensions of your parameter space,
1179 // the point you are interpolating must be scaled to match the data so
1180 // that the selected nearest neighbors are appropriate
1181
1182 for (i = 0; i < _dimensions; i++) vv[i] = (vin[i] - _min[i]) / (_max[i] - _min[i]);
1183 } else {
1184 vv = vin;
1185 }
1186
1187 _kdTree.findNeighbors(neighbors, neighborDistances, vv, numberOfNeighbors);
1188
1189 _timer.addToSearch();
1190
1191 for (i = 0; i < numberOfNeighbors; i++) {
1192 covarianceTestPoint[i] = (*_covariogram)(vv, _kdTree.getData(neighbors[i]));
1193
1194 covariance(i, i) =
1195 (*_covariogram)(_kdTree.getData(neighbors[i]), _kdTree.getData(neighbors[i])) + _lambda;
1196
1197 for (j = i + 1; j < numberOfNeighbors; j++) {
1198 covariance(i, j) = (*_covariogram)(_kdTree.getData(neighbors[i]), _kdTree.getData(neighbors[j]));
1199 covariance(j, i) = covariance(i, j);
1200 }
1201 }
1202
1203 _timer.addToIteration();
1204
1205 // use Eigen's ldlt solver in place of matrix inversion (for speed purposes)
1206 ldlt.compute(covariance);
1207
1208 for (ii = 0; ii < _nFunctions; ii++) {
1209 fbar = 0.0;
1210 for (i = 0; i < numberOfNeighbors; i++) fbar += _function[neighbors[i]][ii];
1211 fbar = fbar / double(numberOfNeighbors);
1212
1213 for (i = 0; i < numberOfNeighbors; i++) bb(i, 0) = _function[neighbors[i]][ii] - fbar;
1214 xx = ldlt.solve(bb);
1215
1216 mu[ii] = fbar;
1217
1218 for (i = 0; i < numberOfNeighbors; i++) {
1219 mu[ii] += covarianceTestPoint[i] * xx(i, 0);
1220 }
1221
1222 } // ii = 0 through _nFunctions
1223
1224 _timer.addToEigen();
1225
1226 variance[0] = (*_covariogram)(vv, vv) + _lambda;
1227
1228 for (i = 0; i < numberOfNeighbors; i++) bb(i) = covarianceTestPoint[i];
1229
1230 xx = ldlt.solve(bb);
1231
1232 for (i = 0; i < numberOfNeighbors; i++) {
1233 variance[0] -= covarianceTestPoint[i] * xx(i, 0);
1234 }
1235 variance[0] = variance[0] * _krigingParameter;
1236
1237 for (i = 1; i < _nFunctions; i++) variance[i] = variance[0];
1238
1239 _timer.addToVariance();
1240 _timer.addToTotal(1);
1241}
1242
1243template <typename T>
1244T GaussianProcess<T>::selfInterpolate(ndarray::Array<T, 1, 1> variance, int dex,
1245 int numberOfNeighbors) const {
1246 if (_nFunctions > 1) {
1248 "You need to call the version of GaussianProcess.selfInterpolate() "
1249 "that accepts mu and variance arrays (which it populates with results). "
1250 "You are interpolating more than one function.");
1251 }
1252
1253 if (variance.getNumElements() != static_cast<ndarray::Size>(_nFunctions)) {
1255 "Your variance array is the incorrect size for the number "
1256 "of functions you are trying to interpolate\n");
1257 }
1258
1259 if (numberOfNeighbors <= 0) {
1261 "Asked for zero or negative number of neighbors\n");
1262 }
1263
1264 if (numberOfNeighbors + 1 > _kdTree.getNPoints()) {
1266 "Asked for more neighbors than you have data points\n");
1267 }
1268
1269 if (dex < 0 || dex >= _kdTree.getNPoints()) {
1271 "Asked to self interpolate on a point that does not exist\n");
1272 }
1273
1274 int i, j;
1275 T fbar, mu;
1276
1277 ndarray::Array<T, 1, 1> covarianceTestPoint;
1278 ndarray::Array<int, 1, 1> selfNeighbors;
1279 ndarray::Array<double, 1, 1> selfDistances;
1280 ndarray::Array<int, 1, 1> neighbors;
1281 ndarray::Array<double, 1, 1> neighborDistances;
1282
1283 Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> covariance, bb, xx;
1284 Eigen::LDLT<Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> > ldlt;
1285
1286 _timer.start();
1287
1288 bb.resize(numberOfNeighbors, 1);
1289 xx.resize(numberOfNeighbors, 1);
1290 covariance.resize(numberOfNeighbors, numberOfNeighbors);
1291 covarianceTestPoint = allocate(ndarray::makeVector(numberOfNeighbors));
1292 neighbors = allocate(ndarray::makeVector(numberOfNeighbors));
1293 neighborDistances = allocate(ndarray::makeVector(numberOfNeighbors));
1294
1295 selfNeighbors = allocate(ndarray::makeVector(numberOfNeighbors + 1));
1296 selfDistances = allocate(ndarray::makeVector(numberOfNeighbors + 1));
1297
1298 // we don't use _useMaxMin because the data has already been normalized
1299
1300 _kdTree.findNeighbors(selfNeighbors, selfDistances, _kdTree.getData(dex), numberOfNeighbors + 1);
1301
1302 _timer.addToSearch();
1303
1304 if (selfNeighbors[0] != dex) {
1306 "Nearest neighbor search in selfInterpolate did not find self\n");
1307 }
1308
1309 // SelfNeighbors[0] will be the point itself (it is its own nearest neighbor)
1310 // We discard that for the interpolation calculation
1311 //
1312 // If you do not wish to do this, simply call the usual ::interpolate() method instead of
1313 //::selfInterpolate()
1314 for (i = 0; i < numberOfNeighbors; i++) {
1315 neighbors[i] = selfNeighbors[i + 1];
1316 neighborDistances[i] = selfDistances[i + 1];
1317 }
1318
1319 fbar = 0.0;
1320 for (i = 0; i < numberOfNeighbors; i++) fbar += _function[neighbors[i]][0];
1321 fbar = fbar / double(numberOfNeighbors);
1322
1323 for (i = 0; i < numberOfNeighbors; i++) {
1324 covarianceTestPoint[i] = (*_covariogram)(_kdTree.getData(dex), _kdTree.getData(neighbors[i]));
1325
1326 covariance(i, i) =
1327 (*_covariogram)(_kdTree.getData(neighbors[i]), _kdTree.getData(neighbors[i])) + _lambda;
1328
1329 for (j = i + 1; j < numberOfNeighbors; j++) {
1330 covariance(i, j) = (*_covariogram)(_kdTree.getData(neighbors[i]), _kdTree.getData(neighbors[j]));
1331 covariance(j, i) = covariance(i, j);
1332 }
1333 }
1334 _timer.addToIteration();
1335
1336 // use Eigen's ldlt solver in place of matrix inversion (for speed purposes)
1337 ldlt.compute(covariance);
1338
1339 for (i = 0; i < numberOfNeighbors; i++) bb(i, 0) = _function[neighbors[i]][0] - fbar;
1340 xx = ldlt.solve(bb);
1341 _timer.addToEigen();
1342
1343 mu = fbar;
1344
1345 for (i = 0; i < numberOfNeighbors; i++) {
1346 mu += covarianceTestPoint[i] * xx(i, 0);
1347 }
1348
1349 variance(0) = (*_covariogram)(_kdTree.getData(dex), _kdTree.getData(dex)) + _lambda;
1350
1351 for (i = 0; i < numberOfNeighbors; i++) bb(i) = covarianceTestPoint[i];
1352
1353 xx = ldlt.solve(bb);
1354
1355 for (i = 0; i < numberOfNeighbors; i++) {
1356 variance(0) -= covarianceTestPoint[i] * xx(i, 0);
1357 }
1358
1359 variance(0) = variance(0) * _krigingParameter;
1360 _timer.addToVariance();
1361 _timer.addToTotal(1);
1362
1363 return mu;
1364}
1365
1366template <typename T>
1367void GaussianProcess<T>::selfInterpolate(ndarray::Array<T, 1, 1> mu, ndarray::Array<T, 1, 1> variance,
1368 int dex, int numberOfNeighbors) const {
1369 if (mu.getNumElements() != static_cast<ndarray::Size>(_nFunctions) ||
1370 variance.getNumElements() != static_cast<ndarray::Size>(_nFunctions)) {
1372 "Your mu and/or var arrays are improperly sized for the number of functions "
1373 "you are interpolating\n");
1374 }
1375
1376 if (numberOfNeighbors <= 0) {
1378 "Asked for zero or negative number of neighbors\n");
1379 }
1380
1381 if (numberOfNeighbors + 1 > _kdTree.getNPoints()) {
1383 "Asked for more neighbors than you have data points\n");
1384 }
1385
1386 if (dex < 0 || dex >= _kdTree.getNPoints()) {
1388 "Asked to self interpolate on a point that does not exist\n");
1389 }
1390
1391 int i, j, ii;
1392 T fbar;
1393
1394 ndarray::Array<T, 1, 1> covarianceTestPoint;
1395 ndarray::Array<int, 1, 1> selfNeighbors;
1396 ndarray::Array<double, 1, 1> selfDistances;
1397 ndarray::Array<int, 1, 1> neighbors;
1398 ndarray::Array<double, 1, 1> neighborDistances;
1399
1400 Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> covariance, bb, xx;
1401 Eigen::LDLT<Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> > ldlt;
1402
1403 _timer.start();
1404
1405 bb.resize(numberOfNeighbors, 1);
1406 xx.resize(numberOfNeighbors, 1);
1407 covariance.resize(numberOfNeighbors, numberOfNeighbors);
1408 covarianceTestPoint = allocate(ndarray::makeVector(numberOfNeighbors));
1409 neighbors = allocate(ndarray::makeVector(numberOfNeighbors));
1410 neighborDistances = allocate(ndarray::makeVector(numberOfNeighbors));
1411
1412 selfNeighbors = allocate(ndarray::makeVector(numberOfNeighbors + 1));
1413 selfDistances = allocate(ndarray::makeVector(numberOfNeighbors + 1));
1414
1415 // we don't use _useMaxMin because the data has already been normalized
1416
1417 _kdTree.findNeighbors(selfNeighbors, selfDistances, _kdTree.getData(dex), numberOfNeighbors + 1);
1418
1419 _timer.addToSearch();
1420
1421 if (selfNeighbors[0] != dex) {
1423 "Nearest neighbor search in selfInterpolate did not find self\n");
1424 }
1425
1426 // SelfNeighbors[0] will be the point itself (it is its own nearest neighbor)
1427 // We discard that for the interpolation calculation
1428 //
1429 // If you do not wish to do this, simply call the usual ::interpolate() method instead of
1430 //::selfInterpolate()
1431 for (i = 0; i < numberOfNeighbors; i++) {
1432 neighbors[i] = selfNeighbors[i + 1];
1433 neighborDistances[i] = selfDistances[i + 1];
1434 }
1435
1436 for (i = 0; i < numberOfNeighbors; i++) {
1437 covarianceTestPoint[i] = (*_covariogram)(_kdTree.getData(dex), _kdTree.getData(neighbors[i]));
1438
1439 covariance(i, i) =
1440 (*_covariogram)(_kdTree.getData(neighbors[i]), _kdTree.getData(neighbors[i])) + _lambda;
1441
1442 for (j = i + 1; j < numberOfNeighbors; j++) {
1443 covariance(i, j) = (*_covariogram)(_kdTree.getData(neighbors[i]), _kdTree.getData(neighbors[j]));
1444 covariance(j, i) = covariance(i, j);
1445 }
1446 }
1447 _timer.addToIteration();
1448
1449 // use Eigen's ldlt solver in place of matrix inversion (for speed purposes)
1450 ldlt.compute(covariance);
1451
1452 for (ii = 0; ii < _nFunctions; ii++) {
1453 fbar = 0.0;
1454 for (i = 0; i < numberOfNeighbors; i++) fbar += _function[neighbors[i]][ii];
1455 fbar = fbar / double(numberOfNeighbors);
1456
1457 for (i = 0; i < numberOfNeighbors; i++) bb(i, 0) = _function[neighbors[i]][ii] - fbar;
1458 xx = ldlt.solve(bb);
1459
1460 mu[ii] = fbar;
1461
1462 for (i = 0; i < numberOfNeighbors; i++) {
1463 mu[ii] += covarianceTestPoint[i] * xx(i, 0);
1464 }
1465 } // ii = 0 through _nFunctions
1466
1467 _timer.addToEigen();
1468
1469 variance[0] = (*_covariogram)(_kdTree.getData(dex), _kdTree.getData(dex)) + _lambda;
1470
1471 for (i = 0; i < numberOfNeighbors; i++) bb(i) = covarianceTestPoint[i];
1472
1473 xx = ldlt.solve(bb);
1474
1475 for (i = 0; i < numberOfNeighbors; i++) {
1476 variance[0] -= covarianceTestPoint[i] * xx(i, 0);
1477 }
1478
1479 variance[0] = variance[0] * _krigingParameter;
1480
1481 for (i = 1; i < _nFunctions; i++) variance[i] = variance[0];
1482
1483 _timer.addToVariance();
1484 _timer.addToTotal(1);
1485}
1486
1487template <typename T>
1488void GaussianProcess<T>::batchInterpolate(ndarray::Array<T, 1, 1> mu, ndarray::Array<T, 1, 1> variance,
1489 ndarray::Array<T, 2, 2> const &queries) const {
1490 int i, j;
1491
1492 ndarray::Size nQueries = queries.template getSize<0>();
1493
1494 if (_nFunctions != 1) {
1496 "Your mu and variance arrays do not have room for all of the functions "
1497 "as you are trying to interpolate\n");
1498 }
1499
1500 if (mu.getNumElements() != nQueries || variance.getNumElements() != nQueries) {
1502 "Your mu and variance arrays do not have room for all of the points "
1503 "at which you are trying to interpolate your function.\n");
1504 }
1505
1506 if (queries.template getSize<1>() != static_cast<ndarray::Size>(_dimensions)) {
1508 "The points you passed to batchInterpolate are of the wrong "
1509 "dimensionality for your Gaussian Process\n");
1510 }
1511
1512 T fbar;
1513 Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> batchCovariance, batchbb, batchxx;
1514 Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> queryCovariance;
1515 Eigen::LDLT<Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> > ldlt;
1516
1517 ndarray::Array<T, 1, 1> v1;
1518
1519 _timer.start();
1520
1521 v1 = allocate(ndarray::makeVector(_dimensions));
1522 batchbb.resize(_npts, 1);
1523 batchxx.resize(_npts, 1);
1524 batchCovariance.resize(_npts, _npts);
1525 queryCovariance.resize(_npts, 1);
1526
1527 for (i = 0; i < _npts; i++) {
1528 batchCovariance(i, i) = (*_covariogram)(_kdTree.getData(i), _kdTree.getData(i)) + _lambda;
1529 for (j = i + 1; j < _npts; j++) {
1530 batchCovariance(i, j) = (*_covariogram)(_kdTree.getData(i), _kdTree.getData(j));
1531 batchCovariance(j, i) = batchCovariance(i, j);
1532 }
1533 }
1534 _timer.addToIteration();
1535
1536 ldlt.compute(batchCovariance);
1537
1538 fbar = 0.0;
1539 for (i = 0; i < _npts; i++) {
1540 fbar += _function[i][0];
1541 }
1542 fbar = fbar / T(_npts);
1543
1544 for (i = 0; i < _npts; i++) {
1545 batchbb(i, 0) = _function[i][0] - fbar;
1546 }
1547 batchxx = ldlt.solve(batchbb);
1548 _timer.addToEigen();
1549
1550 for (ndarray::Size ii = 0; ii < nQueries; ii++) {
1551 for (i = 0; i < _dimensions; i++) v1[i] = queries[ii][i];
1552 if (_useMaxMin == 1) {
1553 for (i = 0; i < _dimensions; i++) v1[i] = (v1[i] - _min[i]) / (_max[i] - _min[i]);
1554 }
1555 mu(ii) = fbar;
1556 for (i = 0; i < _npts; i++) {
1557 mu(ii) += batchxx(i) * (*_covariogram)(v1, _kdTree.getData(i));
1558 }
1559 }
1560 _timer.addToIteration();
1561
1562 for (ndarray::Size ii = 0; ii < nQueries; ii++) {
1563 for (i = 0; i < _dimensions; i++) v1[i] = queries[ii][i];
1564 if (_useMaxMin == 1) {
1565 for (i = 0; i < _dimensions; i++) v1[i] = (v1[i] - _min[i]) / (_max[i] - _min[i]);
1566 }
1567
1568 for (i = 0; i < _npts; i++) {
1569 batchbb(i, 0) = (*_covariogram)(v1, _kdTree.getData(i));
1570 queryCovariance(i, 0) = batchbb(i, 0);
1571 }
1572 batchxx = ldlt.solve(batchbb);
1573
1574 variance[ii] = (*_covariogram)(v1, v1) + _lambda;
1575
1576 for (i = 0; i < _npts; i++) {
1577 variance[ii] -= queryCovariance(i, 0) * batchxx(i);
1578 }
1579
1580 variance[ii] = variance[ii] * _krigingParameter;
1581 }
1582
1583 _timer.addToVariance();
1584 _timer.addToTotal(nQueries);
1585}
1586
1587template <typename T>
1588void GaussianProcess<T>::batchInterpolate(ndarray::Array<T, 2, 2> mu, ndarray::Array<T, 2, 2> variance,
1589 ndarray::Array<T, 2, 2> const &queries) const {
1590 int i, j, ifn;
1591
1592 ndarray::Size nQueries = queries.template getSize<0>();
1593
1594 if (mu.template getSize<0>() != nQueries || variance.template getSize<0>() != nQueries) {
1596 "Your output arrays do not have room for all of the points at which "
1597 "you are interpolating your functions.\n");
1598 }
1599
1600 if (mu.template getSize<1>() != static_cast<ndarray::Size>(_nFunctions) ||
1601 variance.template getSize<1>() != static_cast<ndarray::Size>(_nFunctions)) {
1603 "Your output arrays do not have room for all of the functions you are "
1604 "interpolating\n");
1605 }
1606
1607 if (queries.template getSize<1>() != static_cast<ndarray::Size>(_dimensions)) {
1609 "The points at which you are interpolating your functions have the "
1610 "wrong dimensionality.\n");
1611 }
1612
1613 T fbar;
1614 Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> batchCovariance, batchbb, batchxx;
1615 Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> queryCovariance;
1616 Eigen::LDLT<Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> > ldlt;
1617
1618 ndarray::Array<T, 1, 1> v1;
1619
1620 _timer.start();
1621
1622 v1 = allocate(ndarray::makeVector(_dimensions));
1623 batchbb.resize(_npts, 1);
1624 batchxx.resize(_npts, 1);
1625 batchCovariance.resize(_npts, _npts);
1626 queryCovariance.resize(_npts, 1);
1627
1628 for (i = 0; i < _npts; i++) {
1629 batchCovariance(i, i) = (*_covariogram)(_kdTree.getData(i), _kdTree.getData(i)) + _lambda;
1630 for (j = i + 1; j < _npts; j++) {
1631 batchCovariance(i, j) = (*_covariogram)(_kdTree.getData(i), _kdTree.getData(j));
1632 batchCovariance(j, i) = batchCovariance(i, j);
1633 }
1634 }
1635
1636 _timer.addToIteration();
1637
1638 ldlt.compute(batchCovariance);
1639
1640 _timer.addToEigen();
1641
1642 for (ifn = 0; ifn < _nFunctions; ifn++) {
1643 fbar = 0.0;
1644 for (i = 0; i < _npts; i++) {
1645 fbar += _function[i][ifn];
1646 }
1647 fbar = fbar / T(_npts);
1648 _timer.addToIteration();
1649
1650 for (i = 0; i < _npts; i++) {
1651 batchbb(i, 0) = _function[i][ifn] - fbar;
1652 }
1653 batchxx = ldlt.solve(batchbb);
1654 _timer.addToEigen();
1655
1656 for (ndarray::Size ii = 0; ii < nQueries; ii++) {
1657 for (i = 0; i < _dimensions; i++) v1[i] = queries[ii][i];
1658 if (_useMaxMin == 1) {
1659 for (i = 0; i < _dimensions; i++) v1[i] = (v1[i] - _min[i]) / (_max[i] - _min[i]);
1660 }
1661 mu[ii][ifn] = fbar;
1662 for (i = 0; i < _npts; i++) {
1663 mu[ii][ifn] += batchxx(i) * (*_covariogram)(v1, _kdTree.getData(i));
1664 }
1665 }
1666
1667 } // ifn = 0 to _nFunctions
1668
1669 _timer.addToIteration();
1670 for (ndarray::Size ii = 0; ii < nQueries; ii++) {
1671 for (i = 0; i < _dimensions; i++) v1[i] = queries[ii][i];
1672 if (_useMaxMin == 1) {
1673 for (i = 0; i < _dimensions; i++) v1[i] = (v1[i] - _min[i]) / (_max[i] - _min[i]);
1674 }
1675
1676 for (i = 0; i < _npts; i++) {
1677 batchbb(i, 0) = (*_covariogram)(v1, _kdTree.getData(i));
1678 queryCovariance(i, 0) = batchbb(i, 0);
1679 }
1680 batchxx = ldlt.solve(batchbb);
1681
1682 variance[ii][0] = (*_covariogram)(v1, v1) + _lambda;
1683
1684 for (i = 0; i < _npts; i++) {
1685 variance[ii][0] -= queryCovariance(i, 0) * batchxx(i);
1686 }
1687
1688 variance[ii][0] = variance[ii][0] * _krigingParameter;
1689 for (i = 1; i < _nFunctions; i++) variance[ii][i] = variance[ii][0];
1690 }
1691 _timer.addToVariance();
1692 _timer.addToTotal(nQueries);
1693}
1694
1695template <typename T>
1696void GaussianProcess<T>::batchInterpolate(ndarray::Array<T, 1, 1> mu,
1697 ndarray::Array<T, 2, 2> const &queries) const {
1698 int i, j;
1699
1700 ndarray::Size nQueries = queries.template getSize<0>();
1701
1702 if (_nFunctions != 1) {
1704 "Your output array does not have enough room for all of the functions "
1705 "you are trying to interpolate.\n");
1706 }
1707
1708 if (queries.template getSize<1>() != static_cast<ndarray::Size>(_dimensions)) {
1710 "The points at which you are trying to interpolate your function are "
1711 "of the wrong dimensionality.\n");
1712 }
1713
1714 if (mu.getNumElements() != nQueries) {
1716 "Your output array does not have enough room for all of the points "
1717 "at which you are trying to interpolate your function.\n");
1718 }
1719
1720 T fbar;
1721 Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> batchCovariance, batchbb, batchxx;
1722 Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> queryCovariance;
1723 Eigen::LDLT<Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> > ldlt;
1724
1725 ndarray::Array<T, 1, 1> v1;
1726
1727 _timer.start();
1728
1729 v1 = allocate(ndarray::makeVector(_dimensions));
1730
1731 batchbb.resize(_npts, 1);
1732 batchxx.resize(_npts, 1);
1733 batchCovariance.resize(_npts, _npts);
1734 queryCovariance.resize(_npts, 1);
1735
1736 for (i = 0; i < _npts; i++) {
1737 batchCovariance(i, i) = (*_covariogram)(_kdTree.getData(i), _kdTree.getData(i)) + _lambda;
1738 for (j = i + 1; j < _npts; j++) {
1739 batchCovariance(i, j) = (*_covariogram)(_kdTree.getData(i), _kdTree.getData(j));
1740 batchCovariance(j, i) = batchCovariance(i, j);
1741 }
1742 }
1743 _timer.addToIteration();
1744
1745 ldlt.compute(batchCovariance);
1746
1747 fbar = 0.0;
1748 for (i = 0; i < _npts; i++) {
1749 fbar += _function[i][0];
1750 }
1751 fbar = fbar / T(_npts);
1752
1753 for (i = 0; i < _npts; i++) {
1754 batchbb(i, 0) = _function[i][0] - fbar;
1755 }
1756 batchxx = ldlt.solve(batchbb);
1757 _timer.addToEigen();
1758
1759 for (ndarray::Size ii = 0; ii < nQueries; ii++) {
1760 for (i = 0; i < _dimensions; i++) v1[i] = queries[ii][i];
1761 if (_useMaxMin == 1) {
1762 for (i = 0; i < _dimensions; i++) v1[i] = (v1[i] - _min[i]) / (_max[i] - _min[i]);
1763 }
1764
1765 mu(ii) = fbar;
1766 for (i = 0; i < _npts; i++) {
1767 mu(ii) += batchxx(i) * (*_covariogram)(v1, _kdTree.getData(i));
1768 }
1769 }
1770 _timer.addToIteration();
1771 _timer.addToTotal(nQueries);
1772}
1773
1774template <typename T>
1775void GaussianProcess<T>::batchInterpolate(ndarray::Array<T, 2, 2> mu,
1776 ndarray::Array<T, 2, 2> const &queries) const {
1777 int i, j, ifn;
1778
1779 ndarray::Size nQueries = queries.template getSize<0>();
1780
1781 if (mu.template getSize<0>() != nQueries) {
1783 "Your output array does not have enough room for all of the points "
1784 "at which you want to interpolate your functions.\n");
1785 }
1786
1787 if (mu.template getSize<1>() != static_cast<ndarray::Size>(_nFunctions)) {
1789 "Your output array does not have enough room for all of the functions "
1790 "you are trying to interpolate.\n");
1791 }
1792
1793 if (queries.template getSize<1>() != static_cast<ndarray::Size>(_dimensions)) {
1795 "The points at which you are interpolating your functions do not "
1796 "have the correct dimensionality.\n");
1797 }
1798
1799 T fbar;
1800 Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> batchCovariance, batchbb, batchxx;
1801 Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> queryCovariance;
1802 Eigen::LDLT<Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> > ldlt;
1803
1804 ndarray::Array<T, 1, 1> v1;
1805
1806 _timer.start();
1807
1808 v1 = allocate(ndarray::makeVector(_dimensions));
1809
1810 batchbb.resize(_npts, 1);
1811 batchxx.resize(_npts, 1);
1812 batchCovariance.resize(_npts, _npts);
1813 queryCovariance.resize(_npts, 1);
1814
1815 for (i = 0; i < _npts; i++) {
1816 batchCovariance(i, i) = (*_covariogram)(_kdTree.getData(i), _kdTree.getData(i)) + _lambda;
1817 for (j = i + 1; j < _npts; j++) {
1818 batchCovariance(i, j) = (*_covariogram)(_kdTree.getData(i), _kdTree.getData(j));
1819 batchCovariance(j, i) = batchCovariance(i, j);
1820 }
1821 }
1822
1823 _timer.addToIteration();
1824
1825 ldlt.compute(batchCovariance);
1826
1827 _timer.addToEigen();
1828
1829 for (ifn = 0; ifn < _nFunctions; ifn++) {
1830 fbar = 0.0;
1831 for (i = 0; i < _npts; i++) {
1832 fbar += _function[i][ifn];
1833 }
1834 fbar = fbar / T(_npts);
1835
1836 _timer.addToIteration();
1837
1838 for (i = 0; i < _npts; i++) {
1839 batchbb(i, 0) = _function[i][ifn] - fbar;
1840 }
1841 batchxx = ldlt.solve(batchbb);
1842 _timer.addToEigen();
1843
1844 for (ndarray::Size ii = 0; ii < nQueries; ii++) {
1845 for (i = 0; i < _dimensions; i++) v1[i] = queries[ii][i];
1846 if (_useMaxMin == 1) {
1847 for (i = 0; i < _dimensions; i++) v1[i] = (v1[i] - _min[i]) / (_max[i] - _min[i]);
1848 }
1849
1850 mu[ii][ifn] = fbar;
1851 for (i = 0; i < _npts; i++) {
1852 mu[ii][ifn] += batchxx(i) * (*_covariogram)(v1, _kdTree.getData(i));
1853 }
1854 }
1855
1856 } // ifn = 0 through _nFunctions
1857
1858 _timer.addToTotal(nQueries);
1859}
1860
1861template <typename T>
1862void GaussianProcess<T>::addPoint(ndarray::Array<T, 1, 1> const &vin, T f) {
1863 int i, j;
1864
1865 if (_nFunctions != 1) {
1867 "You are calling the wrong addPoint; you need a vector of functions\n");
1868 }
1869
1870 if (vin.getNumElements() != static_cast<ndarray::Size>(_dimensions)) {
1872 "You are trying to add a point of the wrong dimensionality to "
1873 "your GaussianProcess.\n");
1874 }
1875
1876 ndarray::Array<T, 1, 1> v;
1877 v = allocate(ndarray::makeVector(_dimensions));
1878
1879 for (i = 0; i < _dimensions; i++) {
1880 v[i] = vin[i];
1881 if (_useMaxMin == 1) {
1882 v[i] = (v[i] - _min[i]) / (_max[i] - _min[i]);
1883 }
1884 }
1885
1886 if (_npts == _room) {
1887 ndarray::Array<T, 2, 2> buff;
1888 buff = allocate(ndarray::makeVector(_npts, _nFunctions));
1889 buff.deep() = _function;
1890
1891 _room += _roomStep;
1892 _function = allocate(ndarray::makeVector(_room, _nFunctions));
1893 for (i = 0; i < _npts; i++) {
1894 for (j = 0; j < _nFunctions; j++) {
1895 _function[i][j] = buff[i][j];
1896 }
1897 }
1898 }
1899
1900 _function[_npts][0] = f;
1901
1902 _kdTree.addPoint(v);
1903 _npts = _kdTree.getNPoints();
1904}
1905
1906template <typename T>
1907void GaussianProcess<T>::addPoint(ndarray::Array<T, 1, 1> const &vin, ndarray::Array<T, 1, 1> const &f) {
1908 if (vin.getNumElements() != static_cast<ndarray::Size>(_dimensions)) {
1910 "You are trying to add a point of the wrong dimensionality to "
1911 "your GaussianProcess.\n");
1912 }
1913
1914 if (f.template getSize<0>() != static_cast<ndarray::Size>(_nFunctions)) {
1916 "You are not adding the correct number of function values to "
1917 "your GaussianProcess.\n");
1918 }
1919
1920 int i, j;
1921
1922 ndarray::Array<T, 1, 1> v;
1923 v = allocate(ndarray::makeVector(_dimensions));
1924
1925 for (i = 0; i < _dimensions; i++) {
1926 v[i] = vin[i];
1927 if (_useMaxMin == 1) {
1928 v[i] = (v[i] - _min[i]) / (_max[i] - _min[i]);
1929 }
1930 }
1931
1932 if (_npts == _room) {
1933 ndarray::Array<T, 2, 2> buff;
1934 buff = allocate(ndarray::makeVector(_npts, _nFunctions));
1935 buff.deep() = _function;
1936
1937 _room += _roomStep;
1938 _function = allocate(ndarray::makeVector(_room, _nFunctions));
1939 for (i = 0; i < _npts; i++) {
1940 for (j = 0; j < _nFunctions; j++) {
1941 _function[i][j] = buff[i][j];
1942 }
1943 }
1944 }
1945 for (i = 0; i < _nFunctions; i++) _function[_npts][i] = f[i];
1946
1947 _kdTree.addPoint(v);
1948 _npts = _kdTree.getNPoints();
1949}
1950
1951template <typename T>
1953 int i, j;
1954
1955 _kdTree.removePoint(dex);
1956
1957 for (i = dex; i < _npts; i++) {
1958 for (j = 0; j < _nFunctions; j++) {
1959 _function[i][j] = _function[i + 1][j];
1960 }
1961 }
1962 _npts = _kdTree.getNPoints();
1963}
1964
1965template <typename T>
1967 _krigingParameter = kk;
1968}
1969
1970template <typename T>
1972 _covariogram = covar;
1973}
1974
1975template <typename T>
1977 _lambda = lambda;
1978}
1979
1980template <typename T>
1982 return _timer;
1983}
1984
1985template <typename T>
1987
1988template <typename T>
1989T Covariogram<T>::operator()(ndarray::Array<const T, 1, 1> const &p1,
1990 ndarray::Array<const T, 1, 1> const &p2) const {
1991 std::cout << "by the way, you are calling the wrong operator\n";
1992 exit(1);
1993 return T(1.0);
1994}
1995
1996template <typename T>
1998
1999template <typename T>
2001 _ellSquared = 1.0;
2002}
2003
2004template <typename T>
2006 _ellSquared = ellSquared;
2007}
2008
2009template <typename T>
2010T SquaredExpCovariogram<T>::operator()(ndarray::Array<const T, 1, 1> const &p1,
2011 ndarray::Array<const T, 1, 1> const &p2) const {
2012 T d = 0.0;
2013 for (ndarray::Size i = 0; i < p1.template getSize<0>(); i++) {
2014 d += (p1[i] - p2[i]) * (p1[i] - p2[i]);
2015 }
2016
2017 d = d / _ellSquared;
2018 return T(exp(-0.5 * d));
2019}
2020
2021template <typename T>
2023
2024template <typename T>
2026 _sigma0 = 1.0;
2027 _sigma1 = 1.0;
2028}
2029
2030template <typename T>
2031T NeuralNetCovariogram<T>::operator()(ndarray::Array<const T, 1, 1> const &p1,
2032 ndarray::Array<const T, 1, 1> const &p2) const {
2033 int i, dim;
2034 double num, denom1, denom2, arg;
2035
2036 dim = p1.template getSize<0>();
2037
2038 num = 2.0 * _sigma0;
2039 denom1 = 1.0 + 2.0 * _sigma0;
2040 denom2 = 1.0 + 2.0 * _sigma0;
2041
2042 for (i = 0; i < dim; i++) {
2043 num += 2.0 * p1[i] * p2[i] * _sigma1;
2044 denom1 += 2.0 * p1[i] * p1[i] * _sigma1;
2045 denom2 += 2.0 * p2[i] * p2[i] * _sigma1;
2046 }
2047
2048 arg = num / ::sqrt(denom1 * denom2);
2049 return T(2.0 * (::asin(arg)) / 3.141592654);
2050}
2051
2052template <typename T>
2054 _sigma0 = sigma0;
2055}
2056
2057template <typename T>
2059 _sigma1 = sigma1;
2060}
2061
2062#define INSTANTIATEGP(T) \
2063 template class KdTree<T>; \
2064 template class GaussianProcess<T>; \
2065 template class Covariogram<T>; \
2066 template class SquaredExpCovariogram<T>; \
2067 template class NeuralNetCovariogram<T>;
2068
2069INSTANTIATEGP(double);
2070} // namespace math
2071} // namespace afw
2072} // namespace lsst
Key< Flag > const & target
#define LSST_EXCEPT(type,...)
Create an exception with a given type.
Definition: Exception.h:48
table::Key< double > sigma1
#define INSTANTIATEGP(T)
afw::table::Key< afw::table::Array< VariancePixelT > > variance
T asin(T... args)
T begin(T... args)
The parent class of covariogram functions for use in Gaussian Process interpolation.
virtual T operator()(ndarray::Array< const T, 1, 1 > const &p1, ndarray::Array< const T, 1, 1 > const &p2) const
Actually evaluate the covariogram function relating two points you want to interpolate from.
GaussianProcessTimer & getTimes() const
Give the user acces to _timer, an object keeping track of the time spent on various processes within ...
void setLambda(T lambda)
set the value of the hyperparameter _lambda
int getDim() const
return the dimensionality of data points stored in the GaussianProcess
void setKrigingParameter(T kk)
Assign a value to the Kriging paramter.
void addPoint(ndarray::Array< T, 1, 1 > const &vin, T f)
Add a point to the pool of data used by GaussianProcess for interpolation.
void batchInterpolate(ndarray::Array< T, 1, 1 > mu, ndarray::Array< T, 1, 1 > variance, ndarray::Array< T, 2, 2 > const &queries) const
Interpolate a list of query points using all of the input data (rather than nearest neighbors)
GaussianProcess(const GaussianProcess &)=delete
T interpolate(ndarray::Array< T, 1, 1 > variance, ndarray::Array< T, 1, 1 > const &vin, int numberOfNeighbors) const
Interpolate the function value at one point using a specified number of nearest neighbors.
T selfInterpolate(ndarray::Array< T, 1, 1 > variance, int dex, int numberOfNeighbors) const
This method will interpolate the function on a data point for purposes of optimizing hyper parameters...
void removePoint(int dex)
This will remove a point from the data set.
void setCovariogram(std::shared_ptr< Covariogram< T > > const &covar)
Assign a different covariogram to this GaussianProcess.
void getData(ndarray::Array< T, 2, 2 > pts, ndarray::Array< T, 1, 1 > fn, ndarray::Array< int, 1, 1 > indices) const
Return a sub-sample the data underlying the Gaussian Process.
int getNPoints() const
return the number of data points stored in the GaussianProcess
This is a structure for keeping track of how long the interpolation methods spend on different parts ...
void addToSearch()
Adds time to _searchTime.
void addToVariance()
Adds time to _varianceTime.
void addToIteration()
Adds time to _iterationTime.
void start()
Starts the timer for an individual call to an interpolation routine.
void display()
Displays the current values of all times and _interpolationCount.
void reset()
Resets all of the data members of GaussianProcessTimer to zero.
void addToEigen()
Adds time to _eigenTime.
void addToTotal(int i)
Adds time to _totalTime and adds counts to _interpolationCount.
The data for GaussianProcess is stored in a KD tree to facilitate nearest-neighbor searches.
void removePoint(int dex)
Remove a point from the tree.
void findNeighbors(ndarray::Array< int, 1, 1 > neighdex, ndarray::Array< double, 1, 1 > dd, ndarray::Array< const T, 1, 1 > const &v, int n_nn) const
Find the nearest neighbors of a point.
void addPoint(ndarray::Array< const T, 1, 1 > const &v)
Add a point to the tree.
void Initialize(ndarray::Array< T, 2, 2 > const &dt)
Build a KD Tree to store the data for GaussianProcess.
void getTreeNode(ndarray::Array< int, 1, 1 > const &v, int dex) const
Return the _tree information for a given data point.
int getNPoints() const
return the number of data points stored in the tree
T getData(int ipt, int idim) const
Return one element of one node on the tree.
void setSigma0(double sigma0)
set the _sigma0 hyper parameter
T operator()(ndarray::Array< const T, 1, 1 > const &, ndarray::Array< const T, 1, 1 > const &) const override
Actually evaluate the covariogram function relating two points you want to interpolate from.
void setSigma1(double sigma1)
set the _sigma1 hyper parameter
void setEllSquared(double ellSquared)
set the _ellSquared hyper parameter (the square of the characteristic length scale of the covariogram...
T operator()(ndarray::Array< const T, 1, 1 > const &, ndarray::Array< const T, 1, 1 > const &) const override
Actually evaluate the covariogram function relating two points you want to interpolate from.
static DateTime now(void)
Return current time as a DateTime.
Definition: DateTime.cc:518
Reports errors that are due to events beyond the control of the program.
Definition: Runtime.h:104
T end(T... args)
T exit(T... args)
T exp(T... args)
STL namespace.
T next(T... args)
T push_back(T... args)
MatrixQ covariance
Definition: simpleShape.cc:152
T stable_sort(T... args)