1 ## Copyright (C) 2008-2012 Bill Denney
2 ## Copyright (C) 2008 Jaroslav Hajek
3 ## Copyright (C) 2009 VZLU Prague
5 ## This file is part of Octave.
7 ## Octave is free software; you can redistribute it and/or modify it
8 ## under the terms of the GNU General Public License as published by
9 ## the Free Software Foundation; either version 3 of the License, or (at
10 ## your option) any later version.
12 ## Octave is distributed in the hope that it will be useful, but
13 ## WITHOUT ANY WARRANTY; without even the implied warranty of
14 ## MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 ## General Public License for more details.
17 ## You should have received a copy of the GNU General Public License
18 ## along with Octave; see the file COPYING. If not, see
19 ## <http://www.gnu.org/licenses/>.
22 ## @deftypefn {Function File} {@var{x} =} lsqnonneg (@var{c}, @var{d})
23 ## @deftypefnx {Function File} {@var{x} =} lsqnonneg (@var{c}, @var{d}, @var{x0})
24 ## @deftypefnx {Function File} {[@var{x}, @var{resnorm}] =} lsqnonneg (@dots{})
25 ## @deftypefnx {Function File} {[@var{x}, @var{resnorm}, @var{residual}] =} lsqnonneg (@dots{})
26 ## @deftypefnx {Function File} {[@var{x}, @var{resnorm}, @var{residual}, @var{exitflag}] =} lsqnonneg (@dots{})
27 ## @deftypefnx {Function File} {[@var{x}, @var{resnorm}, @var{residual}, @var{exitflag}, @var{output}] =} lsqnonneg (@dots{})
28 ## @deftypefnx {Function File} {[@var{x}, @var{resnorm}, @var{residual}, @var{exitflag}, @var{output}, @var{lambda}] =} lsqnonneg (@dots{})
29 ## Minimize @code{norm (@var{c}*@var{x} - d)} subject to
30 ## @code{@var{x} >= 0}. @var{c} and @var{d} must be real. @var{x0} is an
31 ## optional initial guess for @var{x}.
37 ## The squared 2-norm of the residual: norm(@var{c}*@var{x}-@var{d})^2
41 ## The residual: @var{d}-@var{c}*@var{x}
45 ## An indicator of convergence. 0 indicates that the iteration count
46 ## was exceeded, and therefore convergence was not reached; >0 indicates
47 ## that the algorithm converged. (The algorithm is stable and will
48 ## converge given enough iterations.)
52 ## A structure with two fields:
54 ## @item "algorithm": The algorithm used ("nnls")
56 ## @item "iterations": The number of iterations taken.
63 ## @seealso{optimset, pqpnonneg}
66 ## PKG_ADD: ## Discard result to avoid polluting workspace with ans at startup.
67 ## PKG_ADD: [~] = __all_opts__ ("lsqnonneg");
69 ## This is implemented from Lawson and Hanson's 1973 algorithm on page
70 ## 161 of Solving Least Squares Problems.
72 function [x, resnorm, residual, exitflag, output, lambda] = lsqnonneg (c, d, x = [], options = struct ())
74 if (nargin == 1 && ischar (c) && strcmp (c, 'defaults'))
75 x = optimset ("MaxIter", 1e5);
79 if (! (nargin >= 2 && nargin <= 4 && ismatrix (c) && ismatrix (d) && isstruct (options)))
83 ## Lawson-Hanson Step 1 (LH1): initialize the variables.
87 ## Initial guess is 0s.
90 ## ensure nonnegative guess.
95 max_iter = optimget (options, "MaxIter", 1e5);
97 ## Initialize P, according to zero pattern of x.
100 ## Initialize the QR factorization, economized form.
101 [q, r] = qr (c(:,p), 0);
106 ## LH3: test for completion.
107 while (iter < max_iter)
108 while (iter < max_iter)
111 ## LH6: compute the positive matrix and find the min norm solution
112 ## of the positive problem.
118 idx = find (xtmp < 0);
121 ## LH7: tmp solution found, iterate.
126 ## LH8, LH9: find the scaling factor.
128 sf = x(pidx)./(x(pidx) - xtmp(idx));
134 ## LH11: move from P to Z all X == 0.
135 ## This corresponds to those indices where minimum of sf is attained.
136 idx = idx (sf == alpha);
139 ## update the QR factorization.
140 [q, r] = qrdelete (q, r, idx);
145 ## compute the gradient.
150 ## verify the solution achieved using qr updating.
151 ## in the best case, this should only take a single step.
160 ## find the maximum gradient.
161 idx = find (w == max (w));
163 warning ("lsqnonneg:nonunique",
164 "a non-unique solution may be returned due to equal gradients");
167 ## move the index from Z to P. Keep P sorted.
168 z = [1:n]; z(p) = [];
170 jdx = 1 + lookup (p, zidx);
171 p = [p(1:jdx-1), zidx, p(jdx:end)];
173 ## insert the column into the QR factorization.
174 [q, r] = qrinsert (q, r, jdx, c(:,zidx));
180 ## Generate the additional output arguments.
182 resnorm = norm (c*x - d) ^ 2;
188 if (nargout > 3 && iter >= max_iter)
192 output = struct ("algorithm", "nnls", "iterations", iter);
195 lambda = zeros (size (x));
203 %! C = [1 0;0 1;2 1];
205 %! assert (lsqnonneg (C, d), [0;0.5], 100*eps)
208 %! C = [0.0372 0.2869;0.6861 0.7071;0.6233 0.6245;0.6344 0.6170];
209 %! d = [0.8587;0.1781;0.0747;0.8405];
210 %! xnew = [0;0.6929];
211 %! assert (lsqnonneg (C, d), xnew, 0.0001)