File: sprandsym.m

package info (click to toggle)
octave 9.4.0-1
  • links: PTS, VCS
  • area: main
  • in suites: forky, sid, trixie
  • size: 144,300 kB
  • sloc: cpp: 332,784; ansic: 77,239; fortran: 20,963; objc: 9,396; sh: 8,213; yacc: 4,925; lex: 4,389; perl: 1,544; java: 1,366; awk: 1,259; makefile: 648; xml: 189
file content (190 lines) | stat: -rw-r--r-- 5,819 bytes parent folder | download
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
########################################################################
##
## Copyright (C) 2004-2024 The Octave Project Developers
##
## See the file COPYRIGHT.md in the top-level directory of this
## distribution or <https://octave.org/copyright/>.
##
## This file is part of Octave.
##
## Octave is free software: you can redistribute it and/or modify it
## under the terms of the GNU General Public License as published by
## the Free Software Foundation, either version 3 of the License, or
## (at your option) any later version.
##
## Octave is distributed in the hope that it will be useful, but
## WITHOUT ANY WARRANTY; without even the implied warranty of
## MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
## GNU General Public License for more details.
##
## You should have received a copy of the GNU General Public License
## along with Octave; see the file COPYING.  If not, see
## <https://www.gnu.org/licenses/>.
##
########################################################################

## -*- texinfo -*-
## @deftypefn  {} {@var{S} =} sprandsym (@var{n}, @var{d})
## @deftypefnx {} {@var{S} =} sprandsym (@var{s})
## Generate a symmetric random sparse matrix.
##
## The size of the matrix will be @var{n}x@var{n}, with a density of values
## given by @var{d}.  @var{d} must be between 0 and 1 inclusive.  Values will
## be normally distributed with a mean of zero and a variance of 1.
##
## If called with a single matrix argument, a random sparse matrix is generated
## wherever the matrix @var{s} is nonzero in its lower triangular part.
## @seealso{sprand, sprandn, spones, sparse}
## @end deftypefn

function S = sprandsym (n, d)

  if (nargin < 1)
    print_usage ();
  endif

  if (nargin == 1)
    [i, j] = find (tril (n));
    [nr, nc] = size (n);
    S = sparse (i, j, randn (size (i)), nr, nc);
    S += tril (S, -1)';
    return;
  endif

  if (!(isscalar (n) && n == fix (n) && n >= 0))
    error ("sprandsym: N must be a non-negative integer 0");
  endif

  if (n == 0)
    S = sparse (n, n);
    return;
  endif

  if (d < 0 || d > 1)
    error ("sprandsym: density D must be between 0 and 1");
  endif

  ## Actual number of nonzero entries
  k = round (n^2*d);

  ## Diagonal nonzero entries, same parity as k
  r = pick_rand_diag (n, k);

  ## Off diagonal nonzero entries
  m = (k - r)/2;

  ondiag = randperm (n, r);
  offdiag = randperm (n*(n - 1)/2, m);

  ## Row index
  i = lookup (cumsum (0:n), offdiag - 1) + 1;

  ## Column index
  j = offdiag - (i - 1).*(i - 2)/2;

  diagvals = randn (1, r);
  offdiagvals = randn (1, m);

  S = sparse ([ondiag, i, j], [ondiag, j, i],
              [diagvals, offdiagvals, offdiagvals], n, n);

endfunction

function r = pick_rand_diag (n, k)

  ## Pick a random number R of entries for the diagonal of a sparse NxN
  ## symmetric square matrix with exactly K nonzero entries, ensuring
  ## that this R is chosen uniformly over all such matrices.
  ##
  ## Let D be the number of diagonal entries and M the number of
  ## off-diagonal entries.  Then K = D + 2*M.  Let A = N*(N-1)/2 be the
  ## number of available entries in the upper triangle of the matrix.
  ## Then, by a simple counting argument, there is a total of
  ##
  ##     T = nchoosek (N, D) * nchoosek (A, M)
  ##
  ## symmetric NxN matrices with a total of K nonzero entries and D on
  ## the diagonal.  Letting D range from mod (K,2) through min (N,K), and
  ## dividing by this sum, we obtain the probability P for D to be each
  ## of those values.
  ##
  ## However, we cannot use this form for computation, as the binomial
  ## coefficients become unmanageably large.  Instead, we use the
  ## successive quotients Q(i) = T(i+1)/T(i), which we easily compute to
  ## be
  ##
  ##               (N - D)*(N - D - 1)*M
  ##     Q =  -------------------------------
  ##            (D + 2)*(D + 1)*(A - M + 1)
  ##
  ## Then, after prepending 1, the cumprod of these quotients is
  ##
  ##      C = [ T(1)/T(1), T(2)/T(1), T(3)/T(1), ..., T(N)/T(1) ]
  ##
  ## Their sum is thus S = sum (T)/T(1), and then C(i)/S is the desired
  ## probability P(i) for i=1:N.  The cumsum will finally give the
  ## distribution function for computing the random number of entries on
  ## the diagonal R.
  ##
  ## Thanks to Zsbán Ambrus <ambrus@math.bme.hu> for most of the ideas
  ## of the implementation here, especially how to do the computation
  ## numerically to avoid overflow.

  ## Degenerate case
  if (k == 1)
    r = 1;
    return;
  endif

  ## Compute the stuff described above
  a = n*(n - 1)/2;
  d = [mod(k,2):2:min(n,k)-2];
  m = (k - d)/2;
  q = (n - d).*(n - d - 1).*m ./ (d + 2)./(d + 1)./(a - m + 1);

  ## Slight modification from discussion above: pivot around the max in
  ## order to avoid overflow (underflow is fine, just means effectively
  ## zero probabilities).
  [~, midx] = max (cumsum (log (q)));
  midx += 1;
  lc = fliplr (cumprod (1./q(midx-1:-1:1)));
  rc = cumprod (q(midx:end));

  ## Now c = t(i)/t(midx), so c > 1 == [].
  c = [lc, 1, rc];
  s = sum (c);
  p = c/s;

  ## Add final d
  d(end+1) = d(end) + 2;

  ## Pick a random r using this distribution
  r = d(sum (cumsum (p) < rand) + 1);

endfunction


%!test
%! s = sprandsym (10, 0.1);
%! assert (issparse (s));
%! assert (issymmetric (s));
%! assert (size (s), [10, 10]);
%! assert (nnz (s) / numel (s), 0.1, .01);

## Test 1-input calling form
%!test
%! s = sprandsym (sparse ([1 2 3], [3 2 3], [2 2 2]));
%! [i, j] = find (s);
%! assert (sort (i), [2 3]');
%! assert (sort (j), [2 3]');

## Test empty array creation
%!assert (size (sprandsym (0, 0.5)), [0, 0])

## Test input validation
%!error <Invalid call> sprandsym ()
%!error sprandsym (ones (3), 0.5)
%!error sprandsym (3.5, 0.5)
%!error sprandsym (-1, 0.5)
%!error sprandsym (3, -1)
%!error sprandsym (3, 2)