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
KernelSolution.cc
Go to the documentation of this file.
1// -*- lsst-c++ -*-
11#include <iterator>
12#include <cmath>
13#include <algorithm>
14#include <limits>
15
16#include <memory>
17#include "boost/timer.hpp"
18
19#include "Eigen/Core"
20#include "Eigen/Cholesky"
21#include "Eigen/QR"
22#include "Eigen/LU"
23#include "Eigen/Eigenvalues"
24#include "Eigen/SVD"
25
26#include "lsst/afw/math.h"
27#include "lsst/afw/geom.h"
28#include "lsst/afw/image.h"
29#include "lsst/afw/detection.h"
30#include "lsst/geom.h"
31#include "lsst/log/Log.h"
33
36
37#include "ndarray.h"
38#include "ndarray/eigen.h"
39
40#define DEBUG_MATRIX 0
41#define DEBUG_MATRIX2 0
42
43namespace geom = lsst::geom;
45namespace afwMath = lsst::afw::math;
46namespace afwGeom = lsst::afw::geom;
47namespace afwImage = lsst::afw::image;
49
50namespace lsst {
51namespace ip {
52namespace diffim {
53
54 /* Unique identifier for solution */
56
58 Eigen::MatrixXd mMat,
59 Eigen::VectorXd bVec,
60 bool fitForBackground
61 ) :
62 _id(++_SolutionId),
63 _mMat(mMat),
64 _bVec(bVec),
65 _aVec(),
66 _solvedBy(NONE),
67 _fitForBackground(fitForBackground)
68 {};
69
71 bool fitForBackground
72 ) :
73 _id(++_SolutionId),
74 _mMat(),
75 _bVec(),
76 _aVec(),
77 _solvedBy(NONE),
78 _fitForBackground(fitForBackground)
79 {};
80
82 _id(++_SolutionId),
83 _mMat(),
84 _bVec(),
85 _aVec(),
86 _solvedBy(NONE),
87 _fitForBackground(true)
88 {};
89
92 }
93
95 return getConditionNumber(_mMat, conditionType);
96 }
97
98 double KernelSolution::getConditionNumber(Eigen::MatrixXd const& mMat,
99 ConditionNumberType conditionType) {
100 switch (conditionType) {
103 Eigen::SelfAdjointEigenSolver<Eigen::MatrixXd> eVecValues(mMat);
104 Eigen::VectorXd eValues = eVecValues.eigenvalues();
105 double eMax = eValues.maxCoeff();
106 double eMin = eValues.minCoeff();
107 LOGL_DEBUG("TRACE3.ip.diffim.KernelSolution.getConditionNumber",
108 "EIGENVALUE eMax / eMin = %.3e", eMax / eMin);
109 return (eMax / eMin);
110 break;
111 }
112 case SVD:
114 Eigen::VectorXd sValues = mMat.jacobiSvd().singularValues();
115 double sMax = sValues.maxCoeff();
116 double sMin = sValues.minCoeff();
117 LOGL_DEBUG("TRACE3.ip.diffim.KernelSolution.getConditionNumber",
118 "SVD eMax / eMin = %.3e", sMax / sMin);
119 return (sMax / sMin);
120 break;
121 }
122 default:
123 {
125 "Undefined ConditionNumberType : only EIGENVALUE, SVD allowed.");
126 break;
127 }
128 }
129 }
130
131 void KernelSolution::solve(Eigen::MatrixXd const& mMat,
132 Eigen::VectorXd const& bVec) {
133
134 if (DEBUG_MATRIX) {
135 std::cout << "M " << std::endl;
136 std::cout << mMat << std::endl;
137 std::cout << "B " << std::endl;
138 std::cout << bVec << std::endl;
139 }
140
141 Eigen::VectorXd aVec = Eigen::VectorXd::Zero(bVec.size());
142
143 boost::timer t;
144 t.restart();
145
146 LOGL_DEBUG("TRACE2.ip.diffim.KernelSolution.solve",
147 "Solving for kernel");
148 _solvedBy = LU;
149 Eigen::FullPivLU<Eigen::MatrixXd> lu(mMat);
150 if (lu.isInvertible()) {
151 aVec = lu.solve(bVec);
152 } else {
153 LOGL_DEBUG("TRACE3.ip.diffim.KernelSolution.solve",
154 "Unable to determine kernel via LU");
155 /* LAST RESORT */
156 try {
157
159 Eigen::SelfAdjointEigenSolver<Eigen::MatrixXd> eVecValues(mMat);
160 Eigen::MatrixXd const& rMat = eVecValues.eigenvectors();
161 Eigen::VectorXd eValues = eVecValues.eigenvalues();
162
163 for (int i = 0; i != eValues.rows(); ++i) {
164 if (eValues(i) != 0.0) {
165 eValues(i) = 1.0/eValues(i);
166 }
167 }
168
169 aVec = rMat * eValues.asDiagonal() * rMat.transpose() * bVec;
170 } catch (pexExcept::Exception& e) {
171
172 _solvedBy = NONE;
173 LOGL_DEBUG("TRACE3.ip.diffim.KernelSolution.solve",
174 "Unable to determine kernel via eigen-values");
175
176 throw LSST_EXCEPT(pexExcept::Exception, "Unable to determine kernel solution");
177 }
178 }
179
180 double time = t.elapsed();
181 LOGL_DEBUG("TRACE3.ip.diffim.KernelSolution.solve",
182 "Compute time for matrix math : %.2f s", time);
183
184 if (DEBUG_MATRIX) {
185 std::cout << "A " << std::endl;
186 std::cout << aVec << std::endl;
187 }
188
189 _aVec = aVec;
190 }
191
192 /*******************************************************************************************************/
193
194 template <typename InputT>
196 lsst::afw::math::KernelList const& basisList,
197 bool fitForBackground
198 )
199 :
200 KernelSolution(fitForBackground),
201 _cMat(),
202 _iVec(),
203 _ivVec(),
204 _kernel(),
205 _background(0.0),
206 _kSum(0.0)
207 {
208 std::vector<double> kValues(basisList.size());
210 new afwMath::LinearCombinationKernel(basisList, kValues)
211 );
212 };
213
214 template <typename InputT>
216 if (_solvedBy == KernelSolution::NONE) {
217 throw LSST_EXCEPT(pexExcept::Exception, "Kernel not solved; cannot return solution");
218 }
219 return _kernel;
220 }
221
222 template <typename InputT>
224 if (_solvedBy == KernelSolution::NONE) {
225 throw LSST_EXCEPT(pexExcept::Exception, "Kernel not solved; cannot return image");
226 }
228 new afwImage::Image<afwMath::Kernel::Pixel>(_kernel->getDimensions())
229 );
230 (void)_kernel->computeImage(*image, false);
231 return image;
232 }
233
234 template <typename InputT>
236 if (_solvedBy == KernelSolution::NONE) {
237 throw LSST_EXCEPT(pexExcept::Exception, "Kernel not solved; cannot return background");
238 }
239 return _background;
240 }
241
242 template <typename InputT>
244 if (_solvedBy == KernelSolution::NONE) {
245 throw LSST_EXCEPT(pexExcept::Exception, "Kernel not solved; cannot return ksum");
246 }
247 return _kSum;
248 }
249
250 template <typename InputT>
253 if (_solvedBy == KernelSolution::NONE) {
254 throw LSST_EXCEPT(pexExcept::Exception, "Kernel not solved; cannot return solution");
255 }
256
257 return std::make_pair(_kernel, _background);
258 }
259
260 template <typename InputT>
262 lsst::afw::image::Image<InputT> const &templateImage,
263 lsst::afw::image::Image<InputT> const &scienceImage,
265 ) {
266
267 afwMath::Statistics varStats = afwMath::makeStatistics(varianceEstimate, afwMath::MIN);
268 if (varStats.getValue(afwMath::MIN) < 0.0) {
270 "Error: variance less than 0.0");
271 }
272 if (varStats.getValue(afwMath::MIN) == 0.0) {
274 "Error: variance equals 0.0, cannot inverse variance weight");
275 }
276
278 std::dynamic_pointer_cast<afwMath::LinearCombinationKernel>(_kernel)->getKernelList();
279
280 unsigned int const nKernelParameters = basisList.size();
281 unsigned int const nBackgroundParameters = _fitForBackground ? 1 : 0;
282 unsigned int const nParameters = nKernelParameters + nBackgroundParameters;
283
284 std::vector<std::shared_ptr<afwMath::Kernel> >::const_iterator kiter = basisList.begin();
285
286 /* Ignore buffers around edge of convolved images :
287 *
288 * If the kernel has width 5, it has center pixel 2. The first good pixel
289 * is the (5-2)=3rd pixel, which is array index 2, and ends up being the
290 * index of the central pixel.
291 *
292 * You also have a buffer of unusable pixels on the other side, numbered
293 * width-center-1. The last good usable pixel is N-width+center+1.
294 *
295 * Example : the kernel is width = 5, center = 2
296 *
297 * ---|---|-c-|---|---|
298 *
299 * the image is width = N
300 * convolve this with the kernel, and you get
301 *
302 * |-x-|-x-|-g-|---|---| ... |---|---|-g-|-x-|-x-|
303 *
304 * g = first/last good pixel
305 * x = bad
306 *
307 * the first good pixel is the array index that has the value "center", 2
308 * the last good pixel has array index N-(5-2)+1
309 * eg. if N = 100, you want to use up to index 97
310 * 100-3+1 = 98, and the loops use i < 98, meaning the last
311 * index you address is 97.
312 */
313
314 /* NOTE - we are accessing particular elements of Eigen arrays using
315 these coordinates, therefore they need to be in LOCAL coordinates.
316 This was written before ndarray unification.
317 */
318 geom::Box2I goodBBox = (*kiter)->shrinkBBox(templateImage.getBBox(afwImage::LOCAL));
319 unsigned int const startCol = goodBBox.getMinX();
320 unsigned int const startRow = goodBBox.getMinY();
321 // endCol/Row is one past the index of the last good col/row
322 unsigned int endCol = goodBBox.getMaxX() + 1;
323 unsigned int endRow = goodBBox.getMaxY() + 1;
324
325 boost::timer t;
326 t.restart();
327
328 /* Eigen representation of input images; only the pixels that are unconvolved in cimage below */
329 Eigen::MatrixXd eigenTemplate = imageToEigenMatrix(templateImage).block(startRow,
330 startCol,
331 endRow-startRow,
332 endCol-startCol);
333 Eigen::MatrixXd eigenScience = imageToEigenMatrix(scienceImage).block(startRow,
334 startCol,
335 endRow-startRow,
336 endCol-startCol);
337 Eigen::MatrixXd eigeniVariance = imageToEigenMatrix(varianceEstimate).block(
338 startRow, startCol, endRow-startRow, endCol-startCol
339 ).array().inverse().matrix();
340
341 /* Resize into 1-D for later usage */
342 eigenTemplate.resize(eigenTemplate.rows()*eigenTemplate.cols(), 1);
343 eigenScience.resize(eigenScience.rows()*eigenScience.cols(), 1);
344 eigeniVariance.resize(eigeniVariance.rows()*eigeniVariance.cols(), 1);
345
346 /* Holds image convolved with basis function */
347 afwImage::Image<PixelT> cimage(templateImage.getDimensions());
348
349 /* Holds eigen representation of image convolved with all basis functions */
350 std::vector<Eigen::MatrixXd> convolvedEigenList(nKernelParameters);
351
352 /* Iterators over convolved image list and basis list */
353 typename std::vector<Eigen::MatrixXd>::iterator eiter = convolvedEigenList.begin();
354 /* Create C_i in the formalism of Alard & Lupton */
355 afwMath::ConvolutionControl convolutionControl;
356 convolutionControl.setDoNormalize(false);
357 for (kiter = basisList.begin(); kiter != basisList.end(); ++kiter, ++eiter) {
358 afwMath::convolve(cimage, templateImage, **kiter, convolutionControl); /* cimage stores convolved image */
359
360 Eigen::MatrixXd cMat = imageToEigenMatrix(cimage).block(startRow,
361 startCol,
362 endRow-startRow,
363 endCol-startCol);
364 cMat.resize(cMat.size(), 1);
365 *eiter = cMat;
366
367 }
368
369 double time = t.elapsed();
370 LOGL_DEBUG("TRACE3.ip.diffim.StaticKernelSolution.build",
371 "Total compute time to do basis convolutions : %.2f s", time);
372 t.restart();
373
374 /*
375 Load matrix with all values from convolvedEigenList : all images
376 (eigeniVariance, convolvedEigenList) must be the same size
377 */
378 Eigen::MatrixXd cMat(eigenTemplate.col(0).size(), nParameters);
379 typename std::vector<Eigen::MatrixXd>::iterator eiterj = convolvedEigenList.begin();
380 typename std::vector<Eigen::MatrixXd>::iterator eiterE = convolvedEigenList.end();
381 for (unsigned int kidxj = 0; eiterj != eiterE; eiterj++, kidxj++) {
382 cMat.col(kidxj) = eiterj->col(0);
383 }
384 /* Treat the last "image" as all 1's to do the background calculation. */
385 if (_fitForBackground)
386 cMat.col(nParameters-1).fill(1.);
387
388 _cMat = cMat;
389 _ivVec = eigeniVariance.col(0);
390 _iVec = eigenScience.col(0);
391
392 /* Make these outside of solve() so I can check condition number */
393 _mMat = _cMat.transpose() * (_ivVec.asDiagonal() * _cMat);
394 _bVec = _cMat.transpose() * (_ivVec.asDiagonal() * _iVec);
395 }
396
397 template <typename InputT>
399 LOGL_DEBUG("TRACE3.ip.diffim.StaticKernelSolution.solve",
400 "mMat is %d x %d; bVec is %d; cMat is %d x %d; vVec is %d; iVec is %d",
401 _mMat.rows(), _mMat.cols(), _bVec.size(),
402 _cMat.rows(), _cMat.cols(), _ivVec.size(), _iVec.size());
403
404 if (DEBUG_MATRIX) {
405 std::cout << "C" << std::endl;
406 std::cout << _cMat << std::endl;
407 std::cout << "iV" << std::endl;
408 std::cout << _ivVec << std::endl;
409 std::cout << "I" << std::endl;
410 std::cout << _iVec << std::endl;
411 }
412
413 try {
415 } catch (pexExcept::Exception &e) {
416 LSST_EXCEPT_ADD(e, "Unable to solve static kernel matrix");
417 throw e;
418 }
419 /* Turn matrices into _kernel and _background */
420 _setKernel();
421 }
422
423 template <typename InputT>
425 if (_solvedBy == KernelSolution::NONE) {
426 throw LSST_EXCEPT(pexExcept::Exception, "Kernel not solved; cannot make solution");
427 }
428
429 unsigned int const nParameters = _aVec.size();
430 unsigned int const nBackgroundParameters = _fitForBackground ? 1 : 0;
431 unsigned int const nKernelParameters =
432 std::dynamic_pointer_cast<afwMath::LinearCombinationKernel>(_kernel)->getKernelList().size();
433 if (nParameters != (nKernelParameters + nBackgroundParameters))
434 throw LSST_EXCEPT(pexExcept::Exception, "Mismatched sizes in kernel solution");
435
436 /* Fill in the kernel results */
437 std::vector<double> kValues(nKernelParameters);
438 for (unsigned int idx = 0; idx < nKernelParameters; idx++) {
439 if (std::isnan(_aVec(idx))) {
441 str(boost::format("Unable to determine kernel solution %d (nan)") % idx));
442 }
443 kValues[idx] = _aVec(idx);
444 }
445 _kernel->setKernelParameters(kValues);
446
448 new ImageT(_kernel->getDimensions())
449 );
450 _kSum = _kernel->computeImage(*image, false);
451
452 if (_fitForBackground) {
453 if (std::isnan(_aVec(nParameters-1))) {
455 str(boost::format("Unable to determine background solution %d (nan)") %
456 (nParameters-1)));
457 }
458 _background = _aVec(nParameters-1);
459 }
460 }
461
462
463 template <typename InputT>
465 throw LSST_EXCEPT(pexExcept::Exception, "Uncertainty calculation not supported");
466
467 /* Estimate of parameter uncertainties comes from the inverse of the
468 * covariance matrix (noise spectrum).
469 * N.R. 15.4.8 to 15.4.15
470 *
471 * Since this is a linear problem no need to use Fisher matrix
472 * N.R. 15.5.8
473 *
474 * Although I might be able to take advantage of the solution above.
475 * Since this now works and is not the rate limiting step, keep as-is for DC3a.
476 *
477 * Use Cholesky decomposition again.
478 * Cholkesy:
479 * Cov = L L^t
480 * Cov^(-1) = (L L^t)^(-1)
481 * = (L^T)^-1 L^(-1)
482 *
483 * Code would be:
484 *
485 * Eigen::MatrixXd cov = _mMat.transpose() * _mMat;
486 * Eigen::LLT<Eigen::MatrixXd> llt = cov.llt();
487 * Eigen::MatrixXd error2 = llt.matrixL().transpose().inverse()*llt.matrixL().inverse();
488 */
489 }
490
491 /*******************************************************************************************************/
492
493 template <typename InputT>
495 lsst::afw::math::KernelList const& basisList,
496 bool fitForBackground
497 ) :
498 StaticKernelSolution<InputT>(basisList, fitForBackground)
499 {};
500
501 template <typename InputT>
503 lsst::afw::image::Image<InputT> const &templateImage,
504 lsst::afw::image::Image<InputT> const &scienceImage,
507 ) {
508
509 afwMath::Statistics varStats = afwMath::makeStatistics(varianceEstimate, afwMath::MIN);
510 if (varStats.getValue(afwMath::MIN) < 0.0) {
512 "Error: variance less than 0.0");
513 }
514 if (varStats.getValue(afwMath::MIN) == 0.0) {
516 "Error: variance equals 0.0, cannot inverse variance weight");
517 }
518
519 /* Full footprint of all input images */
521 new afwDet::Footprint(std::make_shared<afwGeom::SpanSet>(templateImage.getBBox())));
522
523 afwMath::KernelList basisList =
524 std::dynamic_pointer_cast<afwMath::LinearCombinationKernel>(this->_kernel)->getKernelList();
525 std::vector<std::shared_ptr<afwMath::Kernel> >::const_iterator kiter = basisList.begin();
526
527 /* Only BAD pixels marked in this mask */
528 afwImage::MaskPixel bitMask =
533
534 /* Create a Footprint that contains all the masked pixels set above */
535 afwDet::Threshold threshold = afwDet::Threshold(bitMask, afwDet::Threshold::BITMASK, true);
536 afwDet::FootprintSet maskFpSet(pixelMask, threshold, true);
537
538 /* And spread it by the kernel half width */
539 int growPix = (*kiter)->getCtr().getX();
540 afwDet::FootprintSet maskedFpSetGrown(maskFpSet, growPix, true);
541
542#if 0
543 for (typename afwDet::FootprintSet::FootprintList::iterator
544 ptr = maskedFpSetGrown.getFootprints()->begin(),
545 end = maskedFpSetGrown.getFootprints()->end();
546 ptr != end;
547 ++ptr) {
548
549 afwDet::setMaskFromFootprint(finalMask,
550 (**ptr),
552 }
553#endif
554
556 for (auto const & foot : *(maskedFpSetGrown.getFootprints())) {
557 foot->getSpans()->setMask(finalMask, afwImage::Mask<afwImage::MaskPixel>::getPlaneBitMask("BAD"));
558 }
559 pixelMask.writeFits("pixelmask.fits");
560 finalMask.writeFits("finalmask.fits");
561
562
563 ndarray::Array<int, 1, 1> maskArray =
564 ndarray::allocate(ndarray::makeVector(fullFp->getArea()));
565 fullFp->getSpans()->flatten(maskArray, finalMask.getArray(), templateImage.getXY0());
566 auto maskEigen = ndarray::asEigenMatrix(maskArray);
567
568 ndarray::Array<InputT, 1, 1> arrayTemplate =
569 ndarray::allocate(ndarray::makeVector(fullFp->getArea()));
570 fullFp->getSpans()->flatten(arrayTemplate, templateImage.getArray(), templateImage.getXY0());
571 auto eigenTemplate0 = ndarray::asEigenMatrix(arrayTemplate);
572
573 ndarray::Array<InputT, 1, 1> arrayScience =
574 ndarray::allocate(ndarray::makeVector(fullFp->getArea()));
575 fullFp->getSpans()->flatten(arrayScience, scienceImage.getArray(), scienceImage.getXY0());
576 auto eigenScience0 = ndarray::asEigenMatrix(arrayScience);
577
578 ndarray::Array<afwImage::VariancePixel, 1, 1> arrayVariance =
579 ndarray::allocate(ndarray::makeVector(fullFp->getArea()));
580 fullFp->getSpans()->flatten(arrayVariance, varianceEstimate.getArray(), varianceEstimate.getXY0());
581 auto eigenVariance0 = ndarray::asEigenMatrix(arrayVariance);
582
583 int nGood = 0;
584 for (int i = 0; i < maskEigen.size(); i++) {
585 if (maskEigen(i) == 0.0)
586 nGood += 1;
587 }
588
589 Eigen::VectorXd eigenTemplate(nGood);
590 Eigen::VectorXd eigenScience(nGood);
591 Eigen::VectorXd eigenVariance(nGood);
592 int nUsed = 0;
593 for (int i = 0; i < maskEigen.size(); i++) {
594 if (maskEigen(i) == 0.0) {
595 eigenTemplate(nUsed) = eigenTemplate0(i);
596 eigenScience(nUsed) = eigenScience0(i);
597 eigenVariance(nUsed) = eigenVariance0(i);
598 nUsed += 1;
599 }
600 }
601
602
603 boost::timer t;
604 t.restart();
605
606 unsigned int const nKernelParameters = basisList.size();
607 unsigned int const nBackgroundParameters = this->_fitForBackground ? 1 : 0;
608 unsigned int const nParameters = nKernelParameters + nBackgroundParameters;
609
610 /* Holds image convolved with basis function */
611 afwImage::Image<InputT> cimage(templateImage.getDimensions());
612
613 /* Holds eigen representation of image convolved with all basis functions */
614 std::vector<Eigen::VectorXd> convolvedEigenList(nKernelParameters);
615
616 /* Iterators over convolved image list and basis list */
617 typename std::vector<Eigen::VectorXd>::iterator eiter = convolvedEigenList.begin();
618
619 /* Create C_i in the formalism of Alard & Lupton */
620 afwMath::ConvolutionControl convolutionControl;
621 convolutionControl.setDoNormalize(false);
622 for (kiter = basisList.begin(); kiter != basisList.end(); ++kiter, ++eiter) {
623 afwMath::convolve(cimage, templateImage, **kiter, convolutionControl); /* cimage stores convolved image */
624
625 ndarray::Array<InputT, 1, 1> arrayC =
626 ndarray::allocate(ndarray::makeVector(fullFp->getArea()));
627 fullFp->getSpans()->flatten(arrayC, cimage.getArray(), cimage.getXY0());
628 auto eigenC0 = ndarray::asEigenMatrix(arrayC);
629
630 Eigen::VectorXd eigenC(nGood);
631 int nUsed = 0;
632 for (int i = 0; i < maskEigen.size(); i++) {
633 if (maskEigen(i) == 0.0) {
634 eigenC(nUsed) = eigenC0(i);
635 nUsed += 1;
636 }
637 }
638
639 *eiter = eigenC;
640 }
641 double time = t.elapsed();
642 LOGL_DEBUG("TRACE3.ip.diffim.StaticKernelSolution.buildWithMask",
643 "Total compute time to do basis convolutions : %.2f s", time);
644 t.restart();
645
646 /* Load matrix with all convolved images */
647 Eigen::MatrixXd cMat(eigenTemplate.size(), nParameters);
648 typename std::vector<Eigen::VectorXd>::iterator eiterj = convolvedEigenList.begin();
649 typename std::vector<Eigen::VectorXd>::iterator eiterE = convolvedEigenList.end();
650 for (unsigned int kidxj = 0; eiterj != eiterE; eiterj++, kidxj++) {
651 cMat.block(0, kidxj, eigenTemplate.size(), 1) =
652 Eigen::MatrixXd(*eiterj).block(0, 0, eigenTemplate.size(), 1);
653 }
654 /* Treat the last "image" as all 1's to do the background calculation. */
655 if (this->_fitForBackground)
656 cMat.col(nParameters-1).fill(1.);
657
658 this->_cMat = cMat;
659 this->_ivVec = eigenVariance.array().inverse().matrix();
660 this->_iVec = eigenScience;
661
662 /* Make these outside of solve() so I can check condition number */
663 this->_mMat = this->_cMat.transpose() * this->_ivVec.asDiagonal() * (this->_cMat);
664 this->_bVec = this->_cMat.transpose() * this->_ivVec.asDiagonal() * (this->_iVec);
665 }
666
667
668 template <typename InputT>
670 lsst::afw::image::Image<InputT> const &templateImage,
671 lsst::afw::image::Image<InputT> const &scienceImage,
674 ) {
675
676 afwMath::Statistics varStats = afwMath::makeStatistics(varianceEstimate, afwMath::MIN);
677 if (varStats.getValue(afwMath::MIN) < 0.0) {
679 "Error: variance less than 0.0");
680 }
681 if (varStats.getValue(afwMath::MIN) == 0.0) {
683 "Error: variance equals 0.0, cannot inverse variance weight");
684 }
685
687 std::dynamic_pointer_cast<afwMath::LinearCombinationKernel>(this->_kernel)->getKernelList();
688 std::vector<std::shared_ptr<afwMath::Kernel> >::const_iterator kiter = basisList.begin();
689
690 /* Only BAD pixels marked in this mask */
692 afwImage::MaskPixel bitMask =
696 sMask &= bitMask;
697 /* TBD: Need to figure out a way to spread this; currently done in Python */
698
699 unsigned int const nKernelParameters = basisList.size();
700 unsigned int const nBackgroundParameters = this->_fitForBackground ? 1 : 0;
701 unsigned int const nParameters = nKernelParameters + nBackgroundParameters;
702
703 /* NOTE - we are accessing particular elements of Eigen arrays using
704 these coordinates, therefore they need to be in LOCAL coordinates.
705 This was written before ndarray unification.
706 */
707 /* Ignore known EDGE pixels for speed */
708 geom::Box2I shrunkLocalBBox = (*kiter)->shrinkBBox(templateImage.getBBox(afwImage::LOCAL));
709 LOGL_DEBUG("TRACE3.ip.diffim.MaskedKernelSolution.build",
710 "Limits of good pixels after convolution: %d,%d -> %d,%d (local)",
711 shrunkLocalBBox.getMinX(), shrunkLocalBBox.getMinY(),
712 shrunkLocalBBox.getMaxX(), shrunkLocalBBox.getMaxY());
713
714 /* Subimages are addressed in raw pixel coordinates */
715 unsigned int startCol = shrunkLocalBBox.getMinX();
716 unsigned int startRow = shrunkLocalBBox.getMinY();
717 unsigned int endCol = shrunkLocalBBox.getMaxX();
718 unsigned int endRow = shrunkLocalBBox.getMaxY();
719 /* Needed for Eigen block slicing operation */
720 endCol += 1;
721 endRow += 1;
722
723 boost::timer t;
724 t.restart();
725
726 /* Eigen representation of input images; only the pixels that are unconvolved in cimage below */
727 Eigen::MatrixXi eMask = maskToEigenMatrix(sMask).block(startRow,
728 startCol,
729 endRow-startRow,
730 endCol-startCol);
731
732 Eigen::MatrixXd eigenTemplate = imageToEigenMatrix(templateImage).block(startRow,
733 startCol,
734 endRow-startRow,
735 endCol-startCol);
736 Eigen::MatrixXd eigenScience = imageToEigenMatrix(scienceImage).block(startRow,
737 startCol,
738 endRow-startRow,
739 endCol-startCol);
740 Eigen::MatrixXd eigeniVariance = imageToEigenMatrix(varianceEstimate).block(
741 startRow, startCol, endRow-startRow, endCol-startCol
742 ).array().inverse().matrix();
743
744 /* Resize into 1-D for later usage */
745 eMask.resize(eMask.rows()*eMask.cols(), 1);
746 eigenTemplate.resize(eigenTemplate.rows()*eigenTemplate.cols(), 1);
747 eigenScience.resize(eigenScience.rows()*eigenScience.cols(), 1);
748 eigeniVariance.resize(eigeniVariance.rows()*eigeniVariance.cols(), 1);
749
750 /* Do the masking, slowly for now... */
751 Eigen::MatrixXd maskedEigenTemplate(eigenTemplate.rows(), 1);
752 Eigen::MatrixXd maskedEigenScience(eigenScience.rows(), 1);
753 Eigen::MatrixXd maskedEigeniVariance(eigeniVariance.rows(), 1);
754 int nGood = 0;
755 for (int i = 0; i < eMask.rows(); i++) {
756 if (eMask(i, 0) == 0) {
757 maskedEigenTemplate(nGood, 0) = eigenTemplate(i, 0);
758 maskedEigenScience(nGood, 0) = eigenScience(i, 0);
759 maskedEigeniVariance(nGood, 0) = eigeniVariance(i, 0);
760 nGood += 1;
761 }
762 }
763 /* Can't resize() since all values are lost; use blocks */
764 eigenTemplate = maskedEigenTemplate.block(0, 0, nGood, 1);
765 eigenScience = maskedEigenScience.block(0, 0, nGood, 1);
766 eigeniVariance = maskedEigeniVariance.block(0, 0, nGood, 1);
767
768
769 /* Holds image convolved with basis function */
770 afwImage::Image<InputT> cimage(templateImage.getDimensions());
771
772 /* Holds eigen representation of image convolved with all basis functions */
773 std::vector<Eigen::MatrixXd> convolvedEigenList(nKernelParameters);
774
775 /* Iterators over convolved image list and basis list */
776 typename std::vector<Eigen::MatrixXd>::iterator eiter = convolvedEigenList.begin();
777 /* Create C_i in the formalism of Alard & Lupton */
778 afwMath::ConvolutionControl convolutionControl;
779 convolutionControl.setDoNormalize(false);
780 for (kiter = basisList.begin(); kiter != basisList.end(); ++kiter, ++eiter) {
781 afwMath::convolve(cimage, templateImage, **kiter, convolutionControl); /* cimage stores convolved image */
782
783 Eigen::MatrixXd cMat = imageToEigenMatrix(cimage).block(startRow,
784 startCol,
785 endRow-startRow,
786 endCol-startCol);
787 cMat.resize(cMat.size(), 1);
788
789 /* Do masking */
790 Eigen::MatrixXd maskedcMat(cMat.rows(), 1);
791 int nGood = 0;
792 for (int i = 0; i < eMask.rows(); i++) {
793 if (eMask(i, 0) == 0) {
794 maskedcMat(nGood, 0) = cMat(i, 0);
795 nGood += 1;
796 }
797 }
798 cMat = maskedcMat.block(0, 0, nGood, 1);
799 *eiter = cMat;
800 }
801
802 double time = t.elapsed();
803 LOGL_DEBUG("TRACE3.ip.diffim.StaticKernelSolution.build",
804 "Total compute time to do basis convolutions : %.2f s", time);
805 t.restart();
806
807 /*
808 Load matrix with all values from convolvedEigenList : all images
809 (eigeniVariance, convolvedEigenList) must be the same size
810 */
811 Eigen::MatrixXd cMat(eigenTemplate.col(0).size(), nParameters);
812 typename std::vector<Eigen::MatrixXd>::iterator eiterj = convolvedEigenList.begin();
813 typename std::vector<Eigen::MatrixXd>::iterator eiterE = convolvedEigenList.end();
814 for (unsigned int kidxj = 0; eiterj != eiterE; eiterj++, kidxj++) {
815 cMat.col(kidxj) = eiterj->col(0);
816 }
817 /* Treat the last "image" as all 1's to do the background calculation. */
818 if (this->_fitForBackground)
819 cMat.col(nParameters-1).fill(1.);
820
821 this->_cMat = cMat;
822 this->_ivVec = eigeniVariance.col(0);
823 this->_iVec = eigenScience.col(0);
824
825 /* Make these outside of solve() so I can check condition number */
826 this->_mMat = this->_cMat.transpose() * this->_ivVec.asDiagonal() * this->_cMat;
827 this->_bVec = this->_cMat.transpose() * this->_ivVec.asDiagonal() * this->_iVec;
828
829 }
830
831 /* NOTE - this was written before the ndarray unification. I am rewriting
832 buildSingleMask to use the ndarray stuff */
833 template <typename InputT>
835 lsst::afw::image::Image<InputT> const &templateImage,
836 lsst::afw::image::Image<InputT> const &scienceImage,
838 lsst::geom::Box2I maskBox
839 ) {
840
841 afwMath::Statistics varStats = afwMath::makeStatistics(varianceEstimate, afwMath::MIN);
842 if (varStats.getValue(afwMath::MIN) < 0.0) {
844 "Error: variance less than 0.0");
845 }
846 if (varStats.getValue(afwMath::MIN) == 0.0) {
848 "Error: variance equals 0.0, cannot inverse variance weight");
849 }
850
852 std::dynamic_pointer_cast<afwMath::LinearCombinationKernel>(this->_kernel)->getKernelList();
853
854 unsigned int const nKernelParameters = basisList.size();
855 unsigned int const nBackgroundParameters = this->_fitForBackground ? 1 : 0;
856 unsigned int const nParameters = nKernelParameters + nBackgroundParameters;
857
858 std::vector<std::shared_ptr<afwMath::Kernel> >::const_iterator kiter = basisList.begin();
859
860 /*
861 NOTE : If we are using these views in Afw's Image space, we need to
862 make sure and compensate for the XY0 of the image:
863
864 geom::Box2I fullBBox = templateImage.getBBox();
865 int maskStartCol = maskBox.getMinX();
866 int maskStartRow = maskBox.getMinY();
867 int maskEndCol = maskBox.getMaxX();
868 int maskEndRow = maskBox.getMaxY();
869
870
871 If we are going to be doing the slicing in Eigen matrices derived from
872 the images, ignore the XY0.
873
874 geom::Box2I fullBBox = templateImage.getBBox(afwImage::LOCAL);
875
876 int maskStartCol = maskBox.getMinX() - templateImage.getX0();
877 int maskStartRow = maskBox.getMinY() - templateImage.getY0();
878 int maskEndCol = maskBox.getMaxX() - templateImage.getX0();
879 int maskEndRow = maskBox.getMaxY() - templateImage.getY0();
880
881 */
882
883
884 geom::Box2I shrunkBBox = (*kiter)->shrinkBBox(templateImage.getBBox());
885
886 LOGL_DEBUG("TRACE3.ip.diffim.MaskedKernelSolution.build",
887 "Limits of good pixels after convolution: %d,%d -> %d,%d",
888 shrunkBBox.getMinX(), shrunkBBox.getMinY(),
889 shrunkBBox.getMaxX(), shrunkBBox.getMaxY());
890
891 unsigned int const startCol = shrunkBBox.getMinX();
892 unsigned int const startRow = shrunkBBox.getMinY();
893 unsigned int const endCol = shrunkBBox.getMaxX();
894 unsigned int const endRow = shrunkBBox.getMaxY();
895
896 /* NOTE: no endCol/endRow += 1 for block slicing, since we are doing the
897 slicing using Afw, not Eigen
898
899 Eigen arrays have index 0,0 in the upper right, while LSST images
900 have 0,0 in the lower left. The y-axis is flipped. When translating
901 Images to Eigen matrices in ipDiffim::imageToEigenMatrix this is
902 accounted for. However, we still need to be aware of this fact if
903 addressing subregions of an Eigen matrix. This is why the slicing is
904 done in Afw, its cleaner.
905
906 Please see examples/maskedKernel.cc for elaboration on some of the
907 tests done to make sure this indexing gets done right.
908
909 */
910
911
912 /* Inner limits; of mask box */
913 int maskStartCol = maskBox.getMinX();
914 int maskStartRow = maskBox.getMinY();
915 int maskEndCol = maskBox.getMaxX();
916 int maskEndRow = maskBox.getMaxY();
917
918 /*
919
920 |---------------------------|
921 | Kernel Boundary |
922 | |---------------------| |
923 | | Top | |
924 | |......_________......| |
925 | | | | | |
926 | | L | Box | R | |
927 | | | | | |
928 | |......---------......| |
929 | | Bottom | |
930 | |---------------------| |
931 | |
932 |---------------------------|
933
934 4 regions we want to extract from the pixels: top bottom left right
935
936 */
937 geom::Box2I tBox = geom::Box2I(geom::Point2I(startCol, maskEndRow + 1),
938 geom::Point2I(endCol, endRow));
939
940 geom::Box2I bBox = geom::Box2I(geom::Point2I(startCol, startRow),
941 geom::Point2I(endCol, maskStartRow - 1));
942
943 geom::Box2I lBox = geom::Box2I(geom::Point2I(startCol, maskStartRow),
944 geom::Point2I(maskStartCol - 1, maskEndRow));
945
946 geom::Box2I rBox = geom::Box2I(geom::Point2I(maskEndCol + 1, maskStartRow),
947 geom::Point2I(endCol, maskEndRow));
948
949 LOGL_DEBUG("TRACE3.ip.diffim.MaskedKernelSolution.build",
950 "Upper good pixel region: %d,%d -> %d,%d",
951 tBox.getMinX(), tBox.getMinY(), tBox.getMaxX(), tBox.getMaxY());
952 LOGL_DEBUG("TRACE3.ip.diffim.MaskedKernelSolution.build",
953 "Bottom good pixel region: %d,%d -> %d,%d",
954 bBox.getMinX(), bBox.getMinY(), bBox.getMaxX(), bBox.getMaxY());
955 LOGL_DEBUG("TRACE3.ip.diffim.MaskedKernelSolution.build",
956 "Left good pixel region: %d,%d -> %d,%d",
957 lBox.getMinX(), lBox.getMinY(), lBox.getMaxX(), lBox.getMaxY());
958 LOGL_DEBUG("TRACE3.ip.diffim.MaskedKernelSolution.build",
959 "Right good pixel region: %d,%d -> %d,%d",
960 rBox.getMinX(), rBox.getMinY(), rBox.getMaxX(), rBox.getMaxY());
961
963 boxArray.push_back(tBox);
964 boxArray.push_back(bBox);
965 boxArray.push_back(lBox);
966 boxArray.push_back(rBox);
967
968 int totalSize = tBox.getWidth() * tBox.getHeight();
969 totalSize += bBox.getWidth() * bBox.getHeight();
970 totalSize += lBox.getWidth() * lBox.getHeight();
971 totalSize += rBox.getWidth() * rBox.getHeight();
972
973 Eigen::MatrixXd eigenTemplate(totalSize, 1);
974 Eigen::MatrixXd eigenScience(totalSize, 1);
975 Eigen::MatrixXd eigeniVariance(totalSize, 1);
976 eigenTemplate.setZero();
977 eigenScience.setZero();
978 eigeniVariance.setZero();
979
980 boost::timer t;
981 t.restart();
982
983 int nTerms = 0;
984 typename std::vector<geom::Box2I>::iterator biter = boxArray.begin();
985 for (; biter != boxArray.end(); ++biter) {
986 int area = (*biter).getWidth() * (*biter).getHeight();
987
988 afwImage::Image<InputT> siTemplate(templateImage, *biter);
989 afwImage::Image<InputT> siScience(scienceImage, *biter);
990 afwImage::Image<InputT> sVarEstimate(varianceEstimate, *biter);
991
992 Eigen::MatrixXd eTemplate = imageToEigenMatrix(siTemplate);
993 Eigen::MatrixXd eScience = imageToEigenMatrix(siScience);
994 Eigen::MatrixXd eiVarEstimate = imageToEigenMatrix(sVarEstimate).array().inverse().matrix();
995
996 eTemplate.resize(area, 1);
997 eScience.resize(area, 1);
998 eiVarEstimate.resize(area, 1);
999
1000 eigenTemplate.block(nTerms, 0, area, 1) = eTemplate.block(0, 0, area, 1);
1001 eigenScience.block(nTerms, 0, area, 1) = eScience.block(0, 0, area, 1);
1002 eigeniVariance.block(nTerms, 0, area, 1) = eiVarEstimate.block(0, 0, area, 1);
1003
1004 nTerms += area;
1005 }
1006
1007 afwImage::Image<InputT> cimage(templateImage.getDimensions());
1008
1009 std::vector<Eigen::MatrixXd> convolvedEigenList(nKernelParameters);
1010 typename std::vector<Eigen::MatrixXd>::iterator eiter = convolvedEigenList.begin();
1011 /* Create C_i in the formalism of Alard & Lupton */
1012 afwMath::ConvolutionControl convolutionControl;
1013 convolutionControl.setDoNormalize(false);
1014 for (kiter = basisList.begin(); kiter != basisList.end(); ++kiter, ++eiter) {
1015 afwMath::convolve(cimage, templateImage, **kiter, convolutionControl); /* cimage stores convolved image */
1016 Eigen::MatrixXd cMat(totalSize, 1);
1017 cMat.setZero();
1018
1019 int nTerms = 0;
1020 typename std::vector<geom::Box2I>::iterator biter = boxArray.begin();
1021 for (; biter != boxArray.end(); ++biter) {
1022 int area = (*biter).getWidth() * (*biter).getHeight();
1023
1024 afwImage::Image<InputT> csubimage(cimage, *biter);
1025 Eigen::MatrixXd esubimage = imageToEigenMatrix(csubimage);
1026 esubimage.resize(area, 1);
1027 cMat.block(nTerms, 0, area, 1) = esubimage.block(0, 0, area, 1);
1028
1029 nTerms += area;
1030 }
1031
1032 *eiter = cMat;
1033
1034 }
1035
1036 double time = t.elapsed();
1037 LOGL_DEBUG("TRACE3.ip.diffim.MaskedKernelSolution.build",
1038 "Total compute time to do basis convolutions : %.2f s", time);
1039 t.restart();
1040
1041 /*
1042 Load matrix with all values from convolvedEigenList : all images
1043 (eigeniVariance, convolvedEigenList) must be the same size
1044 */
1045 Eigen::MatrixXd cMat(eigenTemplate.col(0).size(), nParameters);
1046 typename std::vector<Eigen::MatrixXd>::iterator eiterj = convolvedEigenList.begin();
1047 typename std::vector<Eigen::MatrixXd>::iterator eiterE = convolvedEigenList.end();
1048 for (unsigned int kidxj = 0; eiterj != eiterE; eiterj++, kidxj++) {
1049 cMat.col(kidxj) = eiterj->col(0);
1050 }
1051 /* Treat the last "image" as all 1's to do the background calculation. */
1052 if (this->_fitForBackground)
1053 cMat.col(nParameters-1).fill(1.);
1054
1055 this->_cMat = cMat;
1056 this->_ivVec = eigeniVariance.col(0);
1057 this->_iVec = eigenScience.col(0);
1058
1059 /* Make these outside of solve() so I can check condition number */
1060 this->_mMat = this->_cMat.transpose() * this->_ivVec.asDiagonal() * this->_cMat;
1061 this->_bVec = this->_cMat.transpose() * this->_ivVec.asDiagonal() * this->_iVec;
1062 }
1063 /*******************************************************************************************************/
1064
1065
1066 template <typename InputT>
1068 lsst::afw::math::KernelList const& basisList,
1069 bool fitForBackground,
1070 Eigen::MatrixXd const& hMat,
1072 )
1073 :
1074 StaticKernelSolution<InputT>(basisList, fitForBackground),
1075 _hMat(hMat),
1076 _ps(ps.deepCopy())
1077 {};
1078
1079 template <typename InputT>
1081 Eigen::MatrixXd vMat = this->_cMat.jacobiSvd().matrixV();
1082 Eigen::MatrixXd vMatvMatT = vMat * vMat.transpose();
1083
1084 /* Find pseudo inverse of mMat, which may be ill conditioned */
1085 Eigen::SelfAdjointEigenSolver<Eigen::MatrixXd> eVecValues(this->_mMat);
1086 Eigen::MatrixXd const& rMat = eVecValues.eigenvectors();
1087 Eigen::VectorXd eValues = eVecValues.eigenvalues();
1088 double eMax = eValues.maxCoeff();
1089 for (int i = 0; i != eValues.rows(); ++i) {
1090 if (eValues(i) == 0.0) {
1091 eValues(i) = 0.0;
1092 }
1093 else if ((eMax / eValues(i)) > maxCond) {
1094 LOGL_DEBUG("TRACE3.ip.diffim.RegularizedKernelSolution.estimateRisk",
1095 "Truncating eValue %d; %.5e / %.5e = %.5e vs. %.5e",
1096 i, eMax, eValues(i), eMax / eValues(i), maxCond);
1097 eValues(i) = 0.0;
1098 }
1099 else {
1100 eValues(i) = 1.0 / eValues(i);
1101 }
1102 }
1103 Eigen::MatrixXd mInv = rMat * eValues.asDiagonal() * rMat.transpose();
1104
1105 std::vector<double> lambdas = _createLambdaSteps();
1106 std::vector<double> risks;
1107 for (unsigned int i = 0; i < lambdas.size(); i++) {
1108 double l = lambdas[i];
1109 Eigen::MatrixXd mLambda = this->_mMat + l * _hMat;
1110
1111 try {
1112 KernelSolution::solve(mLambda, this->_bVec);
1113 } catch (pexExcept::Exception &e) {
1114 LSST_EXCEPT_ADD(e, "Unable to solve regularized kernel matrix");
1115 throw e;
1116 }
1117 Eigen::VectorXd term1 = (this->_aVec.transpose() * vMatvMatT * this->_aVec);
1118 if (term1.size() != 1)
1119 throw LSST_EXCEPT(pexExcept::Exception, "Matrix size mismatch");
1120
1121 double term2a = (vMatvMatT * mLambda.inverse()).trace();
1122
1123 Eigen::VectorXd term2b = (this->_aVec.transpose() * (mInv * this->_bVec));
1124 if (term2b.size() != 1)
1125 throw LSST_EXCEPT(pexExcept::Exception, "Matrix size mismatch");
1126
1127 double risk = term1(0) + 2 * (term2a - term2b(0));
1128 LOGL_DEBUG("TRACE4.ip.diffim.RegularizedKernelSolution.estimateRisk",
1129 "Lambda = %.3f, Risk = %.5e",
1130 l, risk);
1131 LOGL_DEBUG("TRACE5.ip.diffim.RegularizedKernelSolution.estimateRisk",
1132 "%.5e + 2 * (%.5e - %.5e)",
1133 term1(0), term2a, term2b(0));
1134 risks.push_back(risk);
1135 }
1136 std::vector<double>::iterator it = min_element(risks.begin(), risks.end());
1137 int index = distance(risks.begin(), it);
1138 LOGL_DEBUG("TRACE3.ip.diffim.RegularizedKernelSolution.estimateRisk",
1139 "Minimum Risk = %.3e at lambda = %.3e", risks[index], lambdas[index]);
1140
1141 return lambdas[index];
1142
1143 }
1144
1145 template <typename InputT>
1146 Eigen::MatrixXd RegularizedKernelSolution<InputT>::getM(bool includeHmat) {
1147 if (includeHmat == true) {
1148 return this->_mMat + _lambda * _hMat;
1149 }
1150 else {
1151 return this->_mMat;
1152 }
1153 }
1154
1155 template <typename InputT>
1157
1158 LOGL_DEBUG("TRACE3.ip.diffim.RegularizedKernelSolution.solve",
1159 "cMat is %d x %d; vVec is %d; iVec is %d; hMat is %d x %d",
1160 this->_cMat.rows(), this->_cMat.cols(), this->_ivVec.size(),
1161 this->_iVec.size(), _hMat.rows(), _hMat.cols());
1162
1163 if (DEBUG_MATRIX2) {
1164 std::cout << "ID: " << (this->_id) << std::endl;
1165 std::cout << "C:" << std::endl;
1166 std::cout << this->_cMat << std::endl;
1167 std::cout << "Sigma^{-1}:" << std::endl;
1168 std::cout << Eigen::MatrixXd(this->_ivVec.asDiagonal()) << std::endl;
1169 std::cout << "Y:" << std::endl;
1170 std::cout << this->_iVec << std::endl;
1171 std::cout << "H:" << std::endl;
1172 std::cout << _hMat << std::endl;
1173 }
1174
1175
1176 this->_mMat = this->_cMat.transpose() * this->_ivVec.asDiagonal() * this->_cMat;
1177 this->_bVec = this->_cMat.transpose() * this->_ivVec.asDiagonal() * this->_iVec;
1178
1179
1180 /* See N.R. 18.5
1181
1182 Matrix equation to solve is Y = C a + N
1183 Y = vectorized version of I (I = image to not convolve)
1184 C_i = K_i (x) R (R = image to convolve)
1185 a = kernel coefficients
1186 N = noise
1187
1188 If we reweight everything by the inverse square root of the noise
1189 covariance, we get a linear model with the identity matrix for
1190 the noise. The problem can then be solved using least squares,
1191 with the normal equations
1192
1193 C^T Y = C^T C a
1194
1195 or
1196
1197 b = M a
1198
1199 with
1200
1201 b = C^T Y
1202 M = C^T C
1203 a = (C^T C)^{-1} C^T Y
1204
1205
1206 We can regularize the least square problem
1207
1208 Y = C a + lambda a^T H a (+ N, which can be ignored)
1209
1210 or the normal equations
1211
1212 (C^T C + lambda H) a = C^T Y
1213
1214
1215 The solution to the regularization of the least squares problem is
1216
1217 a = (C^T C + lambda H)^{-1} C^T Y
1218
1219 The approximation to Y is
1220
1221 C a = C (C^T C + lambda H)^{-1} C^T Y
1222
1223 with smoothing matrix
1224
1225 S = C (C^T C + lambda H)^{-1} C^T
1226
1227 */
1228
1229 std::string lambdaType = _ps->getAsString("lambdaType");
1230 if (lambdaType == "absolute") {
1231 _lambda = _ps->getAsDouble("lambdaValue");
1232 }
1233 else if (lambdaType == "relative") {
1234 _lambda = this->_mMat.trace() / this->_hMat.trace();
1235 _lambda *= _ps->getAsDouble("lambdaScaling");
1236 }
1237 else if (lambdaType == "minimizeBiasedRisk") {
1238 double tol = _ps->getAsDouble("maxConditionNumber");
1239 _lambda = estimateRisk(tol);
1240 }
1241 else if (lambdaType == "minimizeUnbiasedRisk") {
1242 _lambda = estimateRisk(std::numeric_limits<double>::max());
1243 }
1244 else {
1245 throw LSST_EXCEPT(pexExcept::Exception, "lambdaType in PropertySet not recognized");
1246 }
1247
1248 LOGL_DEBUG("TRACE3.ip.diffim.RegularizedKernelSolution.solve",
1249 "Applying kernel regularization with lambda = %.2e", _lambda);
1250
1251
1252 try {
1253 KernelSolution::solve(this->_mMat + _lambda * _hMat, this->_bVec);
1254 } catch (pexExcept::Exception &e) {
1255 LSST_EXCEPT_ADD(e, "Unable to solve static kernel matrix");
1256 throw e;
1257 }
1258 /* Turn matrices into _kernel and _background */
1260 }
1261
1262 template <typename InputT>
1264 std::vector<double> lambdas;
1265
1266 std::string lambdaStepType = _ps->getAsString("lambdaStepType");
1267 if (lambdaStepType == "linear") {
1268 double lambdaLinMin = _ps->getAsDouble("lambdaLinMin");
1269 double lambdaLinMax = _ps->getAsDouble("lambdaLinMax");
1270 double lambdaLinStep = _ps->getAsDouble("lambdaLinStep");
1271 for (double l = lambdaLinMin; l <= lambdaLinMax; l += lambdaLinStep) {
1272 lambdas.push_back(l);
1273 }
1274 }
1275 else if (lambdaStepType == "log") {
1276 double lambdaLogMin = _ps->getAsDouble("lambdaLogMin");
1277 double lambdaLogMax = _ps->getAsDouble("lambdaLogMax");
1278 double lambdaLogStep = _ps->getAsDouble("lambdaLogStep");
1279 for (double l = lambdaLogMin; l <= lambdaLogMax; l += lambdaLogStep) {
1280 lambdas.push_back(pow(10, l));
1281 }
1282 }
1283 else {
1284 throw LSST_EXCEPT(pexExcept::Exception, "lambdaStepType in PropertySet not recognized");
1285 }
1286 return lambdas;
1287 }
1288
1289 /*******************************************************************************************************/
1290
1292 lsst::afw::math::KernelList const& basisList,
1293 lsst::afw::math::Kernel::SpatialFunctionPtr spatialKernelFunction,
1296 ) :
1298 _spatialKernelFunction(spatialKernelFunction),
1299 _constantFirstTerm(false),
1300 _kernel(),
1301 _background(background),
1302 _kSum(0.0),
1303 _ps(ps.deepCopy()),
1304 _nbases(0),
1305 _nkt(0),
1306 _nbt(0),
1307 _nt(0) {
1308
1309 bool isAlardLupton = _ps->getAsString("kernelBasisSet") == "alard-lupton";
1310 bool usePca = _ps->getAsBool("usePcaForSpatialKernel");
1311 if (isAlardLupton || usePca) {
1312 _constantFirstTerm = true;
1313 }
1314 this->_fitForBackground = _ps->getAsBool("fitForBackground");
1315
1316 _nbases = basisList.size();
1317 _nkt = _spatialKernelFunction->getParameters().size();
1318 _nbt = _fitForBackground ? _background->getParameters().size() : 0;
1319 _nt = 0;
1320 if (_constantFirstTerm) {
1321 _nt = (_nbases - 1) * _nkt + 1 + _nbt;
1322 } else {
1323 _nt = _nbases * _nkt + _nbt;
1324 }
1325
1326 Eigen::MatrixXd mMat(_nt, _nt);
1327 Eigen::VectorXd bVec(_nt);
1328 mMat.setZero();
1329 bVec.setZero();
1330
1331 this->_mMat = mMat;
1332 this->_bVec = bVec;
1333
1335 new afwMath::LinearCombinationKernel(basisList, *_spatialKernelFunction)
1336 );
1337
1338 LOGL_DEBUG("TRACE3.ip.diffim.SpatialKernelSolution",
1339 "Initializing with size %d %d %d and constant first term = %s",
1340 _nkt, _nbt, _nt,
1341 _constantFirstTerm ? "true" : "false");
1342
1343 }
1344
1345 void SpatialKernelSolution::addConstraint(float xCenter, float yCenter,
1346 Eigen::MatrixXd const& qMat,
1347 Eigen::VectorXd const& wVec) {
1348
1349 LOGL_DEBUG("TRACE5.ip.diffim.SpatialKernelSolution.addConstraint",
1350 "Adding candidate at %f, %f", xCenter, yCenter);
1351
1352 /* Calculate P matrices */
1353 /* Pure kernel terms */
1354 Eigen::VectorXd pK(_nkt);
1355 std::vector<double> paramsK = _spatialKernelFunction->getParameters();
1356 for (int idx = 0; idx < _nkt; idx++) { paramsK[idx] = 0.0; }
1357 for (int idx = 0; idx < _nkt; idx++) {
1358 paramsK[idx] = 1.0;
1359 _spatialKernelFunction->setParameters(paramsK);
1360 pK(idx) = (*_spatialKernelFunction)(xCenter, yCenter); /* Assume things don't vary over stamp */
1361 paramsK[idx] = 0.0;
1362 }
1363 Eigen::MatrixXd pKpKt = (pK * pK.transpose());
1364
1365 Eigen::VectorXd pB;
1366 Eigen::MatrixXd pBpBt;
1367 Eigen::MatrixXd pKpBt;
1368 if (_fitForBackground) {
1369 pB = Eigen::VectorXd(_nbt);
1370
1371 /* Pure background terms */
1372 std::vector<double> paramsB = _background->getParameters();
1373 for (int idx = 0; idx < _nbt; idx++) { paramsB[idx] = 0.0; }
1374 for (int idx = 0; idx < _nbt; idx++) {
1375 paramsB[idx] = 1.0;
1376 _background->setParameters(paramsB);
1377 pB(idx) = (*_background)(xCenter, yCenter); /* Assume things don't vary over stamp */
1378 paramsB[idx] = 0.0;
1379 }
1380 pBpBt = (pB * pB.transpose());
1381
1382 /* Cross terms */
1383 pKpBt = (pK * pB.transpose());
1384 }
1385
1386 if (DEBUG_MATRIX) {
1387 std::cout << "Spatial weights" << std::endl;
1388 std::cout << "pKpKt " << pKpKt << std::endl;
1389 if (_fitForBackground) {
1390 std::cout << "pBpBt " << pBpBt << std::endl;
1391 std::cout << "pKpBt " << pKpBt << std::endl;
1392 }
1393 }
1394
1395 if (DEBUG_MATRIX) {
1396 std::cout << "Spatial matrix inputs" << std::endl;
1397 std::cout << "M " << qMat << std::endl;
1398 std::cout << "B " << wVec << std::endl;
1399 }
1400
1401 /* first index to start the spatial blocks; default=0 for non-constant first term */
1402 int m0 = 0;
1403 /* how many rows/cols to adjust the matrices/vectors; default=0 for non-constant first term */
1404 int dm = 0;
1405 /* where to start the background terms; this is always true */
1406 int mb = _nt - _nbt;
1407
1408 if (_constantFirstTerm) {
1409 m0 = 1; /* we need to manually fill in the first (non-spatial) terms below */
1410 dm = _nkt-1; /* need to shift terms due to lack of spatial variation in first term */
1411
1412 _mMat(0, 0) += qMat(0,0);
1413 for(int m2 = 1; m2 < _nbases; m2++) {
1414 _mMat.block(0, m2*_nkt-dm, 1, _nkt) += qMat(0,m2) * pK.transpose();
1415 }
1416 _bVec(0) += wVec(0);
1417
1418 if (_fitForBackground) {
1419 _mMat.block(0, mb, 1, _nbt) += qMat(0,_nbases) * pB.transpose();
1420 }
1421 }
1422
1423 /* Fill in the spatial blocks */
1424 for(int m1 = m0; m1 < _nbases; m1++) {
1425 /* Diagonal kernel-kernel term; only use upper triangular part of pKpKt */
1426 _mMat.block(m1*_nkt-dm, m1*_nkt-dm, _nkt, _nkt) +=
1427 (pKpKt * qMat(m1,m1)).triangularView<Eigen::Upper>();
1428
1429 /* Kernel-kernel terms */
1430 for(int m2 = m1+1; m2 < _nbases; m2++) {
1431 _mMat.block(m1*_nkt-dm, m2*_nkt-dm, _nkt, _nkt) += qMat(m1,m2) * pKpKt;
1432 }
1433
1434 if (_fitForBackground) {
1435 /* Kernel cross terms with background */
1436 _mMat.block(m1*_nkt-dm, mb, _nkt, _nbt) += qMat(m1,_nbases) * pKpBt;
1437 }
1438
1439 /* B vector */
1440 _bVec.segment(m1*_nkt-dm, _nkt) += wVec(m1) * pK;
1441 }
1442
1443 if (_fitForBackground) {
1444 /* Background-background terms only */
1445 _mMat.block(mb, mb, _nbt, _nbt) +=
1446 (pBpBt * qMat(_nbases,_nbases)).triangularView<Eigen::Upper>();
1447 _bVec.segment(mb, _nbt) += wVec(_nbases) * pB;
1448 }
1449
1450 if (DEBUG_MATRIX) {
1451 std::cout << "Spatial matrix outputs" << std::endl;
1452 std::cout << "mMat " << _mMat << std::endl;
1453 std::cout << "bVec " << _bVec << std::endl;
1454 }
1455
1456 }
1457
1460 throw LSST_EXCEPT(pexExcept::Exception, "Kernel not solved; cannot return image");
1461 }
1464 );
1465 (void)_kernel->computeImage(*image, false, pos[0], pos[1]);
1466 return image;
1467 }
1468
1470 /* Fill in the other half of mMat */
1471 for (int i = 0; i < _nt; i++) {
1472 for (int j = i+1; j < _nt; j++) {
1473 _mMat(j,i) = _mMat(i,j);
1474 }
1475 }
1476
1477 try {
1479 } catch (pexExcept::Exception &e) {
1480 LSST_EXCEPT_ADD(e, "Unable to solve spatial kernel matrix");
1481 throw e;
1482 }
1483 /* Turn matrices into _kernel and _background */
1484 _setKernel();
1485 }
1486
1488 afwMath::Kernel::SpatialFunctionPtr> SpatialKernelSolution::getSolutionPair() {
1490 throw LSST_EXCEPT(pexExcept::Exception, "Kernel not solved; cannot return solution");
1491 }
1492
1493 return std::make_pair(_kernel, _background);
1494 }
1495
1496 void SpatialKernelSolution::_setKernel() {
1497 /* Report on condition number */
1498 double cNumber = this->getConditionNumber(EIGENVALUE);
1499
1500 if (_nkt == 1) {
1501 /* Not spatially varying; this fork is a specialization for convolution speed--up */
1502
1503 /* Set the basis coefficients */
1504 std::vector<double> kCoeffs(_nbases);
1505 for (int i = 0; i < _nbases; i++) {
1506 if (std::isnan(_aVec(i))) {
1507 throw LSST_EXCEPT(
1509 str(boost::format(
1510 "I. Unable to determine spatial kernel solution %d (nan). Condition number = %.3e") % i % cNumber));
1511 }
1512 kCoeffs[i] = _aVec(i);
1513 }
1514 lsst::afw::math::KernelList basisList =
1515 std::dynamic_pointer_cast<afwMath::LinearCombinationKernel>(_kernel)->getKernelList();
1516 _kernel.reset(
1517 new afwMath::LinearCombinationKernel(basisList, kCoeffs)
1518 );
1519 }
1520 else {
1521
1522 /* Set the kernel coefficients */
1524 kCoeffs.reserve(_nbases);
1525 for (int i = 0, idx = 0; i < _nbases; i++) {
1526 kCoeffs.push_back(std::vector<double>(_nkt));
1527
1528 /* Deal with the possibility the first term doesn't vary spatially */
1529 if ((i == 0) && (_constantFirstTerm)) {
1530 if (std::isnan(_aVec(idx))) {
1531 throw LSST_EXCEPT(
1533 str(boost::format(
1534 "II. Unable to determine spatial kernel solution %d (nan). Condition number = %.3e") % idx % cNumber));
1535 }
1536 kCoeffs[i][0] = _aVec(idx++);
1537 }
1538 else {
1539 for (int j = 0; j < _nkt; j++) {
1540 if (std::isnan(_aVec(idx))) {
1541 throw LSST_EXCEPT(
1543 str(boost::format(
1544 "III. Unable to determine spatial kernel solution %d (nan). Condition number = %.3e") % idx % cNumber));
1545 }
1546 kCoeffs[i][j] = _aVec(idx++);
1547 }
1548 }
1549 }
1550 _kernel->setSpatialParameters(kCoeffs);
1551 }
1552
1553 /* Set kernel Sum */
1555 _kSum = _kernel->computeImage(*image, false);
1556
1557 /* Set the background coefficients */
1558 std::vector<double> bgCoeffs(_fitForBackground ? _nbt : 1);
1559 if (_fitForBackground) {
1560 for (int i = 0; i < _nbt; i++) {
1561 int idx = _nt - _nbt + i;
1562 if (std::isnan(_aVec(idx))) {
1564 str(boost::format(
1565 "Unable to determine spatial background solution %d (nan)") % (idx)));
1566 }
1567 bgCoeffs[i] = _aVec(idx);
1568 }
1569 }
1570 else {
1571 bgCoeffs[0] = 0.;
1572 }
1573 _background->setParameters(bgCoeffs);
1574 }
1575
1576/***********************************************************************************************************/
1577//
1578// Explicit instantiations
1579//
1580 typedef float InputT;
1581
1582 template class StaticKernelSolution<InputT>;
1583 template class MaskedKernelSolution<InputT>;
1584 template class RegularizedKernelSolution<InputT>;
1585
1586}}} // end of namespace lsst::ip::diffim
int end
#define LSST_EXCEPT_ADD(e, m)
Add the current location and a message to an existing exception before rethrowing it.
Definition: Exception.h:54
#define LSST_EXCEPT(type,...)
Create an exception with a given type.
Definition: Exception.h:48
Image Subtraction helper functions.
Declaration of classes to store the solution for convolution kernels.
LSST DM logging module built on log4cxx.
#define LOGL_DEBUG(logger, message...)
Log a debug-level message using a varargs/printf style interface.
Definition: Log.h:515
uint64_t * ptr
Definition: RangeSet.cc:88
T begin(T... args)
Class to describe the properties of a detected object from an image.
Definition: Footprint.h:63
A set of Footprints, associated with a MaskedImage.
Definition: FootprintSet.h:55
std::shared_ptr< FootprintList > getFootprints()
: Return the Footprints of detected objects
Definition: FootprintSet.h:162
A Threshold is used to pass a threshold value to detection algorithms.
Definition: Threshold.h:43
lsst::geom::Box2I getBBox(ImageOrigin origin=PARENT) const
Definition: ImageBase.h:445
lsst::geom::Extent2I getDimensions() const
Return the image's size; useful for passing to constructors.
Definition: ImageBase.h:356
lsst::geom::Point2I getXY0() const
Return the image's origin.
Definition: ImageBase.h:323
A class to represent a 2-dimensional array of pixels.
Definition: Image.h:51
Represent a 2-dimensional array of bitmask pixels.
Definition: Mask.h:77
void writeFits(std::string const &fileName, daf::base::PropertySet const *metadata=nullptr, std::string const &mode="w") const
Write a mask to a regular FITS file.
Parameters to control convolution.
Definition: ConvolveImage.h:50
void setDoNormalize(bool doNormalize)
Definition: ConvolveImage.h:66
void setParameters(std::vector< double > const &params)
Set all function parameters.
Definition: Function.h:156
std::vector< double > const & getParameters() const noexcept
Return all function parameters.
Definition: Function.h:129
lsst::geom::Extent2I const getDimensions() const
Return the Kernel's dimensions (width, height)
Definition: Kernel.h:212
void setSpatialParameters(const std::vector< std::vector< double > > params)
Set the parameters of all spatial functions.
Definition: Kernel.cc:110
double computeImage(lsst::afw::image::Image< Pixel > &image, bool doNormalize, double x=0.0, double y=0.0) const
Compute an image (pixellized representation of the kernel) in place.
Definition: Kernel.cc:76
A kernel that is a linear combination of fixed basis kernels.
Definition: Kernel.h:704
A class to evaluate image statistics.
Definition: Statistics.h:222
double getValue(Property const prop=NOTHING) const
Return the value of the desired property (if specified in the constructor)
Definition: Statistics.cc:1079
Class for storing generic metadata.
Definition: PropertySet.h:66
An integer coordinate rectangle.
Definition: Box.h:55
int getMinY() const noexcept
Definition: Box.h:158
int getHeight() const noexcept
Definition: Box.h:188
int getMinX() const noexcept
Definition: Box.h:157
int getWidth() const noexcept
Definition: Box.h:187
int getMaxX() const noexcept
Definition: Box.h:161
int getMaxY() const noexcept
Definition: Box.h:162
bool _fitForBackground
Background terms included in fit.
virtual double getConditionNumber(ConditionNumberType conditionType)
Eigen::VectorXd _bVec
Derived least squares B vector.
Eigen::MatrixXd const & getM()
Eigen::VectorXd _aVec
Derived least squares solution matrix.
KernelSolvedBy _solvedBy
Type of algorithm used to make solution.
Eigen::MatrixXd _mMat
Derived least squares M matrix.
static int _SolutionId
Unique identifier for solution.
lsst::afw::image::Image< lsst::afw::math::Kernel::Pixel > ImageT
virtual void buildWithMask(lsst::afw::image::Image< InputT > const &templateImage, lsst::afw::image::Image< InputT > const &scienceImage, lsst::afw::image::Image< lsst::afw::image::VariancePixel > const &varianceEstimate, lsst::afw::image::Mask< lsst::afw::image::MaskPixel > const &pixelMask)
virtual void buildSingleMaskOrig(lsst::afw::image::Image< InputT > const &templateImage, lsst::afw::image::Image< InputT > const &scienceImage, lsst::afw::image::Image< lsst::afw::image::VariancePixel > const &varianceEstimate, lsst::geom::Box2I maskBox)
MaskedKernelSolution(lsst::afw::math::KernelList const &basisList, bool fitForBackground)
virtual void buildOrig(lsst::afw::image::Image< InputT > const &templateImage, lsst::afw::image::Image< InputT > const &scienceImage, lsst::afw::image::Image< lsst::afw::image::VariancePixel > const &varianceEstimate, lsst::afw::image::Mask< lsst::afw::image::MaskPixel > pixelMask)
RegularizedKernelSolution(lsst::afw::math::KernelList const &basisList, bool fitForBackground, Eigen::MatrixXd const &hMat, lsst::daf::base::PropertySet const &ps)
std::pair< std::shared_ptr< lsst::afw::math::LinearCombinationKernel >, lsst::afw::math::Kernel::SpatialFunctionPtr > getSolutionPair()
std::shared_ptr< lsst::afw::image::Image< lsst::afw::math::Kernel::Pixel > > makeKernelImage(lsst::geom::Point2D const &pos)
SpatialKernelSolution(lsst::afw::math::KernelList const &basisList, lsst::afw::math::Kernel::SpatialFunctionPtr spatialKernelFunction, lsst::afw::math::Kernel::SpatialFunctionPtr background, lsst::daf::base::PropertySet const &ps)
void addConstraint(float xCenter, float yCenter, Eigen::MatrixXd const &qMat, Eigen::VectorXd const &wVec)
virtual std::pair< std::shared_ptr< lsst::afw::math::Kernel >, double > getSolutionPair()
void _setKernel()
Set kernel after solution.
virtual std::shared_ptr< lsst::afw::math::Kernel > getKernel()
std::shared_ptr< lsst::afw::math::Kernel > _kernel
Derived single-object convolution kernel.
void _setKernelUncertainty()
Not implemented.
virtual std::shared_ptr< lsst::afw::image::Image< lsst::afw::math::Kernel::Pixel > > makeKernelImage()
StaticKernelSolution(lsst::afw::math::KernelList const &basisList, bool fitForBackground)
virtual void build(lsst::afw::image::Image< InputT > const &templateImage, lsst::afw::image::Image< InputT > const &scienceImage, lsst::afw::image::Image< lsst::afw::image::VariancePixel > const &varianceEstimate)
Provides consistent interface for LSST exceptions.
Definition: Exception.h:107
Reports invalid arguments.
Definition: Runtime.h:66
T end(T... args)
T endl(T... args)
T isnan(T... args)
T make_pair(T... args)
Statistics makeStatistics(lsst::afw::image::Image< Pixel > const &img, lsst::afw::image::Mask< image::MaskPixel > const &msk, int const flags, StatisticsControl const &sctrl=StatisticsControl())
Handle a watered-down front-end to the constructor (no variance)
Definition: Statistics.h:361
void convolve(OutImageT &convolvedImage, InImageT const &inImage, KernelT const &kernel, ConvolutionControl const &convolutionControl=ConvolutionControl())
Convolve an Image or MaskedImage with a Kernel, setting pixels of an existing output image.
@ MIN
estimate sample minimum
Definition: Statistics.h:66
Eigen::MatrixXd imageToEigenMatrix(lsst::afw::image::Image< PixelT > const &img)
Turns a 2-d Image into a 2-d Eigen Matrix.
Eigen::MatrixXi maskToEigenMatrix(lsst::afw::image::Mask< lsst::afw::image::MaskPixel > const &mask)
T push_back(T... args)
T reserve(T... args)
T reset(T... args)
T size(T... args)
#define DEBUG_MATRIX
#define DEBUG_MATRIX2