view scripts/general/sortrows.m @ 18754:0ede4dbb37f1

Overhaul interp1, interp2, interp3 functions. * NEWS: Announce change in 'cubic' interpolation method for interp2 to match Matlab. * bicubic.m: Use interp2 (..., "spline") in %!tests. * interp1.m: Improve docstring. Use switch statement instead of if/elseif tree for simpler code. Use more informative error message than 'table too short'. Add titles to demo plots. Add new demo block showing difference between 'pchip' and 'spline' methods. * interp2.m: Rewrite docstring. Use variable 'extrap' instead of 'extrapval' to match documentation. Use clearer messages in error() calls. Make 'cubic' use the same algorithm as 'pchip' for Matlab compatibility. Use Octave coding conventions regarding spaces between variable and parenthesis. Added input validation tests. * interp3.m: Rewrite docstring. Use clearer messages in error() calls. Make 'cubic' use the same algorithm as 'pchip' for Matlab compatibility. Simplify input processing. Rewrite some %!tests for clarity. Added input validation tests.
author Rik <rik@octave.org>
date Sun, 30 Mar 2014 14:18:43 -0700
parents d63878346099
children 446c46af4b42
line wrap: on
line source

## Copyright (C) 2000-2013 Daniel Calvelo
## Copyright (C) 2009 Jaroslav Hajek
##
## 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
## <http://www.gnu.org/licenses/>.

## -*- texinfo -*-
## @deftypefn  {Function File} {[@var{s}, @var{i}] =} sortrows (@var{A})
## @deftypefnx {Function File} {[@var{s}, @var{i}] =} sortrows (@var{A}, @var{c})
## Sort the rows of the matrix @var{A} according to the order of the
## columns specified in @var{c}.  If @var{c} is omitted, a
## lexicographical sort is used.  By default ascending order is used
## however if elements of @var{c} are negative then the corresponding
## column is sorted in descending order.
## @seealso{sort}
## @end deftypefn

## Author: Daniel Calvelo, Paul Kienzle
## Adapted-by: jwe

function [s, i] = sortrows (A, c)

  if (nargin < 1 || nargin > 2)
    print_usage ();
  endif

  if (nargin == 2)
    if (! (isnumeric (c) && isvector (c))) 
      error ("sortrows: C must be a numeric vector");
    elseif (any (c == 0) || any (abs (c) > columns (A)))
      error ("sortrows: all elements of C must be in the range [1, columns (A)]");
    endif
  endif

  default_mode = "ascend";
  reverse_mode = "descend";

  if (issparse (A))
    ## FIXME: Eliminate this case once __sort_rows_idx__ is fixed to
    ##        handle sparse matrices.
    if (nargin == 1)
      i = sort_rows_idx_generic (default_mode, reverse_mode, A);
    else
      i = sort_rows_idx_generic (default_mode, reverse_mode, A, c);
    endif
  elseif (nargin == 1)
    i = __sort_rows_idx__ (A, default_mode);
  elseif (all (c > 0))
    i = __sort_rows_idx__ (A(:,c), default_mode);
  elseif (all (c < 0))
    i = __sort_rows_idx__ (A(:,-c), reverse_mode);
  else
    ## Otherwise, fall back to the old algorithm.
    i = sort_rows_idx_generic (default_mode, reverse_mode, A, c);
  endif

  ## Only bother to compute s if needed.
  if (isargout (1))
    s = A(i,:);
  endif

endfunction

function i = sort_rows_idx_generic (default_mode, reverse_mode, m, c)

  if (nargin == 3)
    indices = [1:columns(m)]';
    mode(1:columns(m)) = {default_mode};
  else
    for j = 1:length (c);
      if (c(j) < 0)
        mode{j} = reverse_mode;
      else
        mode{j} = default_mode;
      endif
    endfor
    indices = abs (c(:));
  endif

  ## Since sort is 'stable' the order of identical elements will be
  ## preserved, so by traversing the sort indices in reverse order we
  ## will make sure that identical elements in index i are subsorted by
  ## index j.
  indices = flipud (indices);
  mode = flipud (mode');
  i = [1:rows(m)]';
  for j = 1:length (indices);
    [~, idx] = sort (m(i, indices(j)), mode{j});
    i = i(idx);
  endfor

endfunction


%!test
%! m = [1, 1; 1, 2; 3, 6; 2, 7];
%! c = [1, -2];
%! [x, idx] = sortrows (m, c);
%! [sx, sidx] = sortrows (sparse (m), c);
%! assert (x, [1, 2; 1, 1; 2, 7; 3, 6]);
%! assert (idx, [2; 1; 4; 3]);
%! assert (issparse (sx));
%! assert (x, full (sx));
%! assert (idx, sidx);

%!test
%! m = [1, 0, 0, 4];
%! c = 1;
%! [x, idx] = sortrows (m, c);
%! [sx, sidx] = sortrows (sparse (m), c);
%! assert (x, m);
%! assert (idx, 1);
%! assert (issparse (sx));
%! assert (x, full (sx));
%! assert (idx, sidx);

%% Test input validation
%!error sortrows ()
%!error sortrows (1, 2, 3)
%!error sortrows (1, "ascend")
%!error sortrows (1, ones (2,2))
%!error sortrows (1, 0)
%!error sortrows (1, 2)