/src/quantlib/ql/experimental/math/laplaceinterpolation.cpp
Line | Count | Source (jump to first uncovered line) |
1 | | /* -*- mode: c++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */ |
2 | | |
3 | | /* |
4 | | Copyright (C) 2015, 2024 Peter Caspers |
5 | | |
6 | | This file is part of QuantLib, a free-software/open-source library |
7 | | for financial quantitative analysts and developers - http://quantlib.org/ |
8 | | |
9 | | QuantLib is free software: you can redistribute it and/or modify it |
10 | | under the terms of the QuantLib license. You should have received a |
11 | | copy of the license along with this program; if not, please email |
12 | | <quantlib-dev@lists.sf.net>. The license is also available online at |
13 | | <http://quantlib.org/license.shtml>. |
14 | | |
15 | | This program is distributed in the hope that it will be useful, but WITHOUT |
16 | | ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS |
17 | | FOR A PARTICULAR PURPOSE. See the license for more details. |
18 | | */ |
19 | | |
20 | | /*! \file laplaceinterpolation.hpp |
21 | | \brief Laplace interpolation of missing values |
22 | | */ |
23 | | |
24 | | #include <ql/experimental/math/laplaceinterpolation.hpp> |
25 | | #include <ql/math/matrix.hpp> |
26 | | #include <ql/math/matrixutilities/bicgstab.hpp> |
27 | | #include <ql/math/matrixutilities/sparsematrix.hpp> |
28 | | #include <ql/methods/finitedifferences/meshers/fdm1dmesher.hpp> |
29 | | #include <ql/methods/finitedifferences/meshers/fdmmeshercomposite.hpp> |
30 | | #include <ql/methods/finitedifferences/meshers/predefined1dmesher.hpp> |
31 | | #include <ql/methods/finitedifferences/operators/fdmlinearopcomposite.hpp> |
32 | | #include <ql/methods/finitedifferences/operators/fdmlinearoplayout.hpp> |
33 | | #include <ql/methods/finitedifferences/operators/secondderivativeop.hpp> |
34 | | #include <ql/methods/finitedifferences/operators/triplebandlinearop.hpp> |
35 | | |
36 | | namespace QuantLib { |
37 | | |
38 | | LaplaceInterpolation::LaplaceInterpolation(std::function<Real(const std::vector<Size>&)> y, |
39 | | std::vector<std::vector<Real>> x, |
40 | | Real relTol, |
41 | | Size maxIterMultiplier) |
42 | 0 | : y_(std::move(y)), x_(std::move(x)), relTol_(relTol), maxIterMultiplier_(maxIterMultiplier) { |
43 | | |
44 | | // set up the mesher |
45 | |
|
46 | 0 | std::vector<Size> dim; |
47 | 0 | coordinateIncluded_.resize(x_.size()); |
48 | 0 | for (Size i = 0; i < x_.size(); ++i) { |
49 | 0 | coordinateIncluded_[i] = x_[i].size() > 1; |
50 | 0 | if (coordinateIncluded_[i]) |
51 | 0 | dim.push_back(x_[i].size()); |
52 | 0 | } |
53 | |
|
54 | 0 | numberOfCoordinatesIncluded_ = dim.size(); |
55 | |
|
56 | 0 | if (numberOfCoordinatesIncluded_ == 0) { |
57 | 0 | return; |
58 | 0 | } |
59 | | |
60 | 0 | QL_REQUIRE(!dim.empty(), "LaplaceInterpolation: singular point or no points given"); |
61 | | |
62 | 0 | layout_ = ext::make_shared<FdmLinearOpLayout>(dim); |
63 | |
|
64 | 0 | std::vector<ext::shared_ptr<Fdm1dMesher>> meshers; |
65 | 0 | for (auto & i : x_) { |
66 | 0 | if (i.size() > 1) |
67 | 0 | meshers.push_back(ext::make_shared<Predefined1dMesher>(i)); |
68 | 0 | } |
69 | |
|
70 | 0 | auto mesher = ext::make_shared<FdmMesherComposite>(layout_, meshers); |
71 | | |
72 | | // set up the Laplace operator and convert it to matrix |
73 | |
|
74 | 0 | struct LaplaceOp : public FdmLinearOpComposite { |
75 | 0 | explicit LaplaceOp(const ext::shared_ptr<FdmMesher>& mesher) { |
76 | 0 | for (Size direction = 0; direction < mesher->layout()->dim().size(); ++direction) { |
77 | 0 | if (mesher->layout()->dim()[direction] > 1) |
78 | 0 | map_.push_back(SecondDerivativeOp(direction, mesher)); |
79 | 0 | } |
80 | 0 | } |
81 | 0 | std::vector<TripleBandLinearOp> map_; |
82 | |
|
83 | 0 | Size size() const override { QL_FAIL("no impl"); } |
84 | 0 | void setTime(Time t1, Time t2) override { QL_FAIL("no impl"); } |
85 | 0 | Array apply(const array_type& r) const override { QL_FAIL("no impl"); } |
86 | 0 | Array apply_mixed(const Array& r) const override { QL_FAIL("no impl"); } |
87 | 0 | Array apply_direction(Size direction, const Array& r) const override { |
88 | 0 | QL_FAIL("no impl"); |
89 | 0 | } |
90 | 0 | Array solve_splitting(Size direction, const Array& r, Real s) const override { |
91 | 0 | QL_FAIL("no impl"); |
92 | 0 | } |
93 | 0 | Array preconditioner(const Array& r, Real s) const override { QL_FAIL("no impl"); } |
94 | 0 | std::vector<SparseMatrix> toMatrixDecomp() const override { |
95 | 0 | std::vector<SparseMatrix> decomp; |
96 | 0 | decomp.reserve(map_.size()); |
97 | 0 | for (auto const& m : map_) |
98 | 0 | decomp.push_back(m.toMatrix()); |
99 | 0 | return decomp; |
100 | 0 | } |
101 | 0 | }; |
102 | |
|
103 | 0 | SparseMatrix op = LaplaceOp(mesher).toMatrix(); |
104 | | |
105 | | // set up the linear system to solve |
106 | |
|
107 | 0 | Size N = layout_->size(); |
108 | |
|
109 | 0 | SparseMatrix g(N, N, 5 * N); |
110 | 0 | Array rhs(N, 0.0), guess(N, 0.0); |
111 | 0 | Real guessTmp = 0.0; |
112 | |
|
113 | 0 | struct f_A { |
114 | 0 | const SparseMatrix& g; |
115 | 0 | explicit f_A(const SparseMatrix& g) : g(g) {} |
116 | 0 | Array operator()(const Array& x) const { return prod(g, x); } |
117 | 0 | }; |
118 | |
|
119 | 0 | auto rowit = op.begin1(); |
120 | 0 | Size count = 0; |
121 | 0 | std::vector<Real> corner_h(dim.size()); |
122 | 0 | std::vector<Size> corner_neighbour_index(dim.size()); |
123 | 0 | for (auto const& pos : *layout_) { |
124 | 0 | const auto& coord = pos.coordinates(); |
125 | 0 | Real val = |
126 | 0 | y_(numberOfCoordinatesIncluded_ == x_.size() ? coord : fullCoordinates(coord)); |
127 | 0 | QL_REQUIRE(rowit != op.end1() && rowit.index1() == count, |
128 | 0 | "LaplaceInterpolation: op matrix row iterator (" |
129 | 0 | << (rowit != op.end1() ? std::to_string(rowit.index1()) : "na") |
130 | 0 | << ") does not match expected row count (" << count << ")"); |
131 | 0 | if (val == Null<Real>()) { |
132 | 0 | bool isCorner = true; |
133 | 0 | for (Size d = 0; d < dim.size() && isCorner; ++d) { |
134 | 0 | if (coord[d] == 0) { |
135 | 0 | corner_h[d] = meshers[d]->dplus(0); |
136 | 0 | corner_neighbour_index[d] = 1; |
137 | 0 | } else if (coord[d] == layout_->dim()[d] - 1) { |
138 | 0 | corner_h[d] = meshers[d]->dminus(dim[d] - 1); |
139 | 0 | corner_neighbour_index[d] = dim[d] - 2; |
140 | 0 | } else { |
141 | 0 | isCorner = false; |
142 | 0 | } |
143 | 0 | } |
144 | 0 | if (isCorner) { |
145 | | // handling of the "corners", all second derivs are zero in the op |
146 | | // this directly generalizes Numerical Recipes, 3rd ed, eq 3.8.6 |
147 | 0 | Real sum_corner_h = |
148 | 0 | std::accumulate(corner_h.begin(), corner_h.end(), Real(0.0), std::plus<>()); |
149 | 0 | for (Size j = 0; j < dim.size(); ++j) { |
150 | 0 | std::vector<Size> coord_j(coord); |
151 | 0 | coord_j[j] = corner_neighbour_index[j]; |
152 | 0 | Real weight = 0.0; |
153 | 0 | for (Size i = 0; i < dim.size(); ++i) { |
154 | 0 | if (i != j) |
155 | 0 | weight += corner_h[i]; |
156 | 0 | } |
157 | 0 | weight = dim.size() == 1 ? Real(1.0) : Real(weight / sum_corner_h); |
158 | 0 | g(count, layout_->index(coord_j)) = -weight; |
159 | 0 | } |
160 | 0 | g(count, count) = 1.0; |
161 | 0 | } else { |
162 | | // point with at least one dimension with non-trivial second derivative |
163 | 0 | for (auto colit = rowit.begin(); colit != rowit.end(); ++colit) |
164 | 0 | g(count, colit.index2()) = *colit; |
165 | 0 | } |
166 | 0 | rhs[count] = 0.0; |
167 | 0 | guess[count] = guessTmp; |
168 | 0 | } else { |
169 | 0 | g(count, count) = 1; |
170 | 0 | rhs[count] = val; |
171 | 0 | guess[count] = guessTmp = val; |
172 | 0 | } |
173 | 0 | ++count; |
174 | 0 | ++rowit; |
175 | 0 | } |
176 | | |
177 | 0 | interpolatedValues_ = BiCGstab(f_A(g), maxIterMultiplier_ * N, relTol_).solve(rhs, guess).x; |
178 | 0 | } |
179 | | |
180 | | std::vector<Size> |
181 | 0 | LaplaceInterpolation::projectedCoordinates(const std::vector<Size>& coordinates) const { |
182 | 0 | std::vector<Size> tmp; |
183 | 0 | for (Size i = 0; i < coordinates.size(); ++i) { |
184 | 0 | if (coordinateIncluded_[i]) |
185 | 0 | tmp.push_back(coordinates[i]); |
186 | 0 | } |
187 | 0 | return tmp; |
188 | 0 | } |
189 | | |
190 | | std::vector<Size> |
191 | 0 | LaplaceInterpolation::fullCoordinates(const std::vector<Size>& projectedCoordinates) const { |
192 | 0 | std::vector<Size> tmp(coordinateIncluded_.size(), 0); |
193 | 0 | for (Size i = 0, count = 0; i < coordinateIncluded_.size(); ++i) { |
194 | 0 | if (coordinateIncluded_[i]) |
195 | 0 | tmp[i] = projectedCoordinates[count++]; |
196 | 0 | } |
197 | 0 | return tmp; |
198 | 0 | } |
199 | | |
200 | 0 | Real LaplaceInterpolation::operator()(const std::vector<Size>& coordinates) const { |
201 | 0 | QL_REQUIRE(coordinates.size() == x_.size(), "LaplaceInterpolation::operator(): expected " |
202 | 0 | << x_.size() << " coordinates, got " |
203 | 0 | << coordinates.size()); |
204 | 0 | if (numberOfCoordinatesIncluded_ == 0) { |
205 | 0 | Real val = y_(coordinates); |
206 | 0 | return val == Null<Real>() ? 0.0 : val; |
207 | 0 | } else { |
208 | 0 | return interpolatedValues_[layout_->index(numberOfCoordinatesIncluded_ == x_.size() ? |
209 | 0 | coordinates : |
210 | 0 | projectedCoordinates(coordinates))]; |
211 | 0 | } |
212 | 0 | } |
213 | | |
214 | | void laplaceInterpolation(Matrix& A, |
215 | | const std::vector<Real>& x, |
216 | | const std::vector<Real>& y, |
217 | | Real relTol, |
218 | 0 | Size maxIterMultiplier) { |
219 | |
|
220 | 0 | std::vector<std::vector<Real>> tmp; |
221 | 0 | tmp.push_back(y); |
222 | 0 | tmp.push_back(x); |
223 | |
|
224 | 0 | if (y.empty()) { |
225 | 0 | tmp[0].resize(A.rows()); |
226 | 0 | std::iota(tmp[0].begin(), tmp[0].end(), 0.0); |
227 | 0 | } |
228 | |
|
229 | 0 | if (x.empty()) { |
230 | 0 | tmp[1].resize(A.columns()); |
231 | 0 | std::iota(tmp[1].begin(), tmp[1].end(), 0.0); |
232 | 0 | } |
233 | |
|
234 | 0 | LaplaceInterpolation interpolation( |
235 | 0 | [&A](const std::vector<Size>& coordinates) { |
236 | 0 | return A(coordinates[0], coordinates[1]); |
237 | 0 | }, |
238 | 0 | tmp, relTol, maxIterMultiplier); |
239 | |
|
240 | 0 | for (Size i = 0; i < A.rows(); ++i) { |
241 | 0 | for (Size j = 0; j < A.columns(); ++j) { |
242 | 0 | if (A(i, j) == Null<Real>()) |
243 | 0 | A(i, j) = interpolation({i, j}); |
244 | 0 | } |
245 | 0 | } |
246 | 0 | } |
247 | | |
248 | | } // namespace QuantLib |